Home

thesisPivot - University of New Mexico

1. Leg Misclassified Error MagXZ Classifier Left 57 41 91 MeanClassifier Right 42 30 88 MeanClassifier Left 45 33 08 MaxClassifier Right 42 30 88 MaxClassifier Left 46 33 82 MinMaxClassifier Right 47 34 55 MinMaxClassifier Left 46 33 82 MeanMaxClassifier Right 43 31 61 MeanMaxClassifier Table 5 12 XZ Magnitude metric classification results 5 7 2 Non parametric classification K Nearest Neighbors Previous section introduced some classification efforts purely based on statistical metrics pulled out of the testing data As simplistic as they were they managed to be close to 70 accurate while classifying all the patients from the trails This section takes another approach towards trying to classify patients based on the data produced by the pivot shift prototype The method implemented is a known algorithm commonly used in classification 30 and regression analysis The method used is the K nearest neighbors algorithm To implement this algo rithm and following the constrain of using the XZ acceleration data we can say that the classification problem involves e 2 dimensions X Z e 2 classes healthy injured therefore the data coming from the patient trails is used in a different way this time instead of looking at a patient file as a acceleration vs time graph now it is seen as a collection of X Z sample points where X takes the horizontal axis and Z the vert
2. 0 20 40 60 80 100 120 Test Duration Figure 5 7 Compensation of internal tibial rotation for X axis readings where the red signal is the original signal of a left leg test As shown before left leg internal rotation will be measured as a negative X axis acceleration while a right one comes as positive Yellow signal corresponds to the result of applying the operation described by equation 5 1 which at the end creates the effect as if the rotation had the same direction for both left and right pivot shift tests TT Chapter 5 Data Acquisition and Analysis Figure 5 8 shows the difference in comparing X axis results using the raw data and after compensating rotation Pivot Shift Test Beta Thu Feb 26 18 39 30 2015 Pivot Shift Test Beta Thu Feb 26 18 39 21 2015 P39Pl csv P39PD csv P39lflip csv P39PD csv 300 300 xLeftleg x Left leg transformed x Right leg x Right leg 200 man 100 100 E o 8 100 E 2 E 100 200 E 200 400 300 o 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140 Test Duration Test Duration Figure 5 8 Side to side X axis comparison left side with original data right side compensating rotation Analyzing the resulting graph right side it show how the magnitude of the acceleration for both maneuvers left and right side reaches similar maximum values Now we can see how it makes more sense to perform this tra
3. For k 3 outcome is 028 class ACL i i n 0 52 0 50 0 48 0 46 0 44 0 42 0 40 x c Zoomed plot example knn with k 3 Figure 5 18 Testing and training set examples training set the final classification results were the one reporting the minimum error rate For example the results just seen from P12 were k 3 with an error rate equal to 10 Optimal result obtained for this patient was using k 155 with an error rate of 1 76 this patient did pretty good Tables 5 13 and 5 14 are the results obtained while classifying all the patients both right and left legs using knn rule From the results we have that for the right leg the average number of patients 99 Chapter 5 missclassified was of 15 07 and 25 55 for left legs The mean average error rate for all patients was around 26 this is considering all outputs both classified and missclassified results The metric Avg Error Class represents the mean error rate present in the patients correctly classified as Avg Error Miss does the same for the missclassified part The last metric from the tables specifies the percentage of patients that were correctly classified with more than 70 confidence or in other Data Acquisition and Analysis e e class ACL e e class Healthy L5L e a 2 e T Ju mt e uu e o p N e e ur 2 e e 0 5 wh pr 0 5 i o 1 0 0 8 0 6 0 4 0 2 0 0 0 2 0 4 0 6 x
4. 1 5665 1 9335 3 5000 0 0105 Table 5 3 Summary for X Y and Z data for patients by groups These previous tables describe the data of left leg and right leg results combined Prefix H specifies healthy patients O patients with observations and ACL patients with the ligament injury From the data can be seen that both X and Y have negative mean values while Z stays positive Out of the three axis Y is the one with the highest standard deviation Graph 5 11 displays the same statistics but divided by healthy and ACL patients from here it can be seen some differences in the mean value between axis These metrics like mean maximum and minimum values taken out of each group of patients healthy and with the injury will be used later on in this chapter with the goal of seeing if a patient can be recognized as a healthy or injured patient based on these reported values 84 Chapter 5 Data Acquisition and Analysis acceleration g s Acceleration g s 0 7 0 6 0 5 0 4 0 3 E mean 0 2 Esd 01 E median 0 01 ALLX ALLY ALLZ 0 2 0 3 Axis Figure 5 10 Mean median and sd for complete data set by axis 0 5 0 4 0 3 E mean 0 2 sd 0 1 E median H HY HZ CL CLY ACLZ o 0 1 0 2 0 3 Axis Figure 5 11 Mean median and sd by healthy and ACL patients 85 Chapter 5 Data Acquisition and Analysis Other metric calculated out of the tests was related to the maximum and mini mum ac
5. Chapter 2 The Pivot Shift Prototype Hardware Design Term Description Transmitter The device which sends data to the bus Receiver The device which receives data from the bus Master The device that initiates a transfer generates clock sig nals and terminates transfers Slave Device addressed by the master Table 2 4 TC terminology supports two main data rates called normal and fast mode running at 100kHz and 400 kHz respectively There are two signals associated with the I C bus the serial clock line SCL and serial data line SDA The SDA line is a bidirectional line that can be used for sending and receiving data both by master and slave devices These lines use pull up resistors to keep a high state when the bus is free The bus drivers are open drain which means that they can pull the signal line to a low state but cannot turn it to high This implementation prevents potential damage to the drivers and excessive power dissipation disables what is known as bus contention where a device tries to drive the line high while other tries to pull it low Bus operation The operation on the bus starts when the master performs a start condition which is defined as a high to low transition on SDA data line while SCL clock line is high at this point the bus is considered busy The next byte of data contains the address of the slave in the first 7 bits the 8th bit determines whether the devic
6. Figure 5 19 Classification results for patient file P12PD avg patients Missclassified 34 7500 25 5515 Classified 101 0000 74 2647 Avg Error 26 5699 Avg Error Class 10 8085 Avg Error Miss 68 5398 70 accurate 86 7500 63 7868 Table 5 13 KNN results for Left leg words less than 30 error rate Next are the average error rates per class The first column specifies the average number of patients missclassified taking in consideration that there were a total of 100 Chapter 5 Data Acquisition and Analysis avg patients Missclassified 20 5000 15 0735 Classified 115 5000 84 9265 Avg Error 26 3211 Avg Error Class 20 5500 Avg Error Miss 54 3915 70 accurate 62 8333 46 2010 Table 5 14 KNN results for Right leg 132 healthy patients and 8 with the injury however out of those 8 patients 4 had the injury on the left leg as the other 4 on the right For this reason previous tables had a total of patients of 136 The metric avg class error is the avg error rate that was present in the correctly classified tests and avg miss error is the same for the ones that were missclassified patients avg miss patients miss avg class error avg miss error leg H 34 75 26 3257 0 1096 0 6854 left ACL 0 0 0000 0 0720 0 0000 left H 23 6 17 8787 0 2473 0 6616 right ACL 1 25 0000 0 2256 0 5109 right T
7. as valid metrics towards giving objective nature to the test results The work showed in this thesis represent the effort of the design and implemen tation of a portable electronic prototype capable of reproducing acceleration and rotational metrics out of the pivot shift The ideal instrumented clinical examina tion test should be noninvasive portable and applicable in both operating room and office environments 9 11 Chapter 2 The Pivot Shift Prototype Hardware Design This chapter describes in detail the hardware specifics of the previously designed prototype including an overview of the microcontroller used the sensors and their specifications as well as the communication that takes place between the embedded system and the PC 2 1 Arduino Arduino is an open source and open hardware computing platform based on sim ple input output board and a development environment that is based on the Processing 10 programming language Hardware wise Arduino consist of a microcontroller with a set of peripherals like digital input output pins analog input pins likely to be used when working with analog sensors LED s for power and user feedback and support for communication 12 Chapter 2 The Pivot Shift Prototype Hardware Design protocols like Serial I C 11 SPI and Bluetooth among others Arduino offers more than a dozen of different development boards varied in dimen sion features and applicat
8. e log exercise activity on professional athletes 16 Practically speaking a MEMS is a silicon chip which contains mechanical mi crostructures microelectronics microsensors and microactuators 15 Chapter 2 The Pivot Shift Prototype Hardware Design 4 MicroActuators y Microstructures Microelectronics y Figure 2 2 MEMS components diagram The microsensors detect changes in the environment measuring mechanical ther mal magnetic chemical or electromagnetic information while the microelectronics take this information to signal the microactuators in order to react and create changes to the environment Nowadays power management in portable electronics is a must to extend battery life a good thing about MEMS sensors is that they can sense when a device is not being used and put it into sleep mode The following section describes the functionality and characteristics of the MEMS acceleromer and gyroscope used in this project 2 2 1 Accelerometers A MEMS accelerometer is different from a integrated circuits in the fact that a proof mass is machined into the silicon Then when acceleration a is applied the mass m displaces according to Newton s second law F ma which is detected by the sensor This mass proof disturbs the capacitance of a nearby node that change is 16 Chapter 2 The Pivot Shift Prototype Hardware Design measured and then filtered One of the most important specifications to have i
9. 2 0625 3 9879 0 0037 HZ L 0 0647 0 4623 0 0101 1 8810 1 9214 3 8024 0 0048 HX R 0 2447 0 5172 0 2440 2 0383 2 0504 4 0887 0 0054 HY R 0 2263 0 3893 0 1069 1 8609 1 8770 3 7379 0 0041 HZ R 0 3285 0 4703 0 2762 1 9657 2 0544 4 0202 0 0049 ACL L X 0 5411 0 2660 0 5988 1 2319 0 7601 1 9919 0 0119 ACL L Y 0 1291 0 3266 0 0544 1 0900 0 9300 2 0200 0 0146 ACL L Z 0 2728 0 3223 0 1900 1 3327 0 4214 1 7540 0 0144 ACL R X 0 1234 0 3550 0 0948 1 3044 1 2359 2 5403 0 0156 ACL R Y 0 1690 0 4184 0 0060 1 7077 1 8730 3 5806 0 0184 ACL R Z 0 3833 0 4401 0 3770 1 5665 1 7440 3 3105 0 0194 Table 5 7 XYZ summary for patients by groups left and right legs 88 Chapter 5 Data Acquisition and Analysis And finally the data summarizing the average maximum and minimum accelera tion values per leg by healthy and injured groups variable mean sd median min max range se HXmin L 0 7663 0 5299 0 6935 2 0948 0 6915 2 7863 0 0530 HXmax L 0 7722 0 6167 0 7742 0 5665 2 7802 3 3468 0 0617 HYmin L 0 6651 0 4028 0 5464 1 9214 0 0706 1 9919 0 0403 HYmax L 0 2760 0 2030 0 2319 0 0343 1 0544 1 0887 0 0203 HZmin L 0 9551 0 2837 0 9597 1 7278 0 1714 1 5565 0 0
10. and Carlos Vega M D at the Christus Muguerza del Parque Hospital between May 2013 and May 2014 1 1 How the project got started Part of the work showed in this Thesis backs from early 2013 before my arrival to UNM Everything started with a call from a friend MD Carlos Vega that at that moment was working on some orthopedical research with regards of the ACL anterior cruciate ligament knee injury In that call he mentioned got an idea of an electronics project that i may could possibly be interested in He started talking about how he was working on this routine test called pivot shift and the problems it had with the equipment that he was using to measure the results Then he finally Chapter 1 Introduction said could be a way to measure the movement or the force used during the test something to see the results on a PC After that I contacted my friend also an electronics engineer Jose Diaz got together bought some microcontrollers and sensors and started playing with the project 1 2 The ACL Injury Knee injuries comprise about 55 of all sports injuries Out of those ACL anterior cruciate ligament tear represents one of the most common ones Athletes involved in high demand disciplines like basketball soccer and football are more likely to develop this injury Briefly analyzing the knee anatomy the main three bones being part of the knee e Femur e Patella e Tibia These bones connect to each other by li
11. b Y axis Z Axis e e e e Frequency 6000 0 2000 acceleration c Z axis Figure 5 9 Histograms for complete dataset a X axis b Y axis c Z axis From the histograms it can be show how for the X axis the majority of the data samples are between the Og and 1g values Y axis in the Og to 0 5g range while the 83 Chapter 5 Data Acquisition and Analysis Z axis does the same in the 0g 0 5g range axis mean sd median min max range se ALLX 0 1448 0 5939 0 1835 3 2883 2 8085 6 0967 0 0035 ALLY 0 1857 0 3759 0 0746 1 9294 2 0625 3 9919 0 0022 ALLZ 0 1708 0 5095 0 1835 1 9738 2 0544 4 0282 0 0030 Table 5 2 Summary statistics for X Y and Z data for all patients variable mean sd median min max range se HX 0 1561 0 5967 0 2077 2 5988 2 8085 5 4073 0 0044 HY 0 1845 0 3721 0 0827 1 9254 2 0625 3 9879 0 0028 HZ 0 1311 0 5060 0 1391 1 9657 2 0544 4 0202 0 0038 OX 0 1392 0 6146 0 1754 3 2883 2 0262 5 3145 0 0073 OY 0 2083 0 4002 0 0665 1 9294 1 5060 3 4355 0 0047 OZ 0 2468 0 5092 0 2923 1 9738 2 0464 4 0202 0 0060 ACLX 0 0728 0 4931 0 0827 2 3448 1 5020 3 8468 0 0103 ACLY 0 1246 0 3153 0 0262 1 7399 1 8730 3 6129 0 0066 ACLZ 0 2499 0 5017 0 3246
12. of code myPort write 0 60 Chapter 3 The Pivot Shift Prototype Software Design when the Arduino receives the string stops sending sensing data and stays in an idle state waiting for the next test to start sending a string 1 enables that At this point the user can save the test data on an csv file and a png image in this way the graphs can be easily reproduced using an spreadsheet software and the data can be further analyzed Additionally to storing the acceleration and rotation metrics from each of the axes the software calculates the magnitude of both of the metrics to provide an additional graph in which the three axes are combined into a single signal e g to calculate acceleration s magnitude accelerationmag y acceleration acceleration acceleration 3 2 dev ttyUSBO START 1ACC XYZ ACC AND POSITION 0 999 0 868 105 600 118 800 118 800 Figure 3 10 Sample test showing axes signals individually where accelerationx accelerationy and accelerationz are acceleration samples for each of the axes Figure 3 11 illustrates the magnitude for the acceleration and 61 Chapter 3 The Pivot Shift Prototype Software Design rotation signals of the test coming from image 3 10 dev ttyUSBO Save Image Open Previous Graph START 1ACC VMAGNITUDE ACC AND POSITION 1 143 1 109 1 074 1 040 79 200 105 600 118 800 92 400 105 600 118 800 Figure 3 11 Sample test s
13. 3 5 3 6 3 7 3 8 3 9 3 10 3 11 4 1 4 2 5 1 5 2 5 3 5 4 5 9 5 6 5 7 User interface for the pivot shift tester Pivot Shift UI dual layout presentation Sample acceleration graph x axis in red y axis green and z axis in LSM9DSO module directions for acceleration and angular rates Sample test using angular velocity css Sample test using angular displacement Sample test showing axes signals individually Sample test showing the magnitude for acceleration and angular dis Placement o a aene des Aud ri Ae ae geli See a Eat ipsae GE be Prototype used during pivot shift maneuver Graph showing acceleration output per axis from pivot shift test Pivot shift plot analyser nx se dart X SE eee a ee ey Patient graphs left leg left side right leg right side Patient s side to side acceleration comparison for each axis Side to side comparison for X axis in four different patients The pivot shift movement step by step bottom describes positive directions for acceleration on each of the axis 0 02 Interpretation of the maneuver a Compensation of internal tibial rotation for X axis readings xli List of Figures 5 8 5 9 5 10 5 11 5 12 5 13 5 14 5 15 5 16 5 17 5 18 5 19 6 1 6 2 6 3 Side to side X axi
14. 30 35 40 as Test Duration Test Duration Figure 5 2 Patient graphs left leg left side right leg right side From this point on for all upcoming XYZ acceleration graphs the color scheme used for each axis is e X axis red e Y axis green 69 Chapter 5 Data Acquisition and Analysis e Z axis blue Quickly looking at figure 5 2 is difficult to assess something at first glance The recreation of the three consecutive pivot shift maneuvers is spotted without much effort as each of the axis presents 3 peaks or maximum points with similar values To compare left leg with right leg accelerations the following plots were done via the Python application Pivot Shift Test Beta Tue Feb 24 13 50 54 2015 Pivot Shift Test Beta Tue Feb 24 13 51 44 2015 art93 p002pi csv art93 p002pd csv 150 art93 p002pi csv art93 p002pd csv T T T T T T T T T I xLeftleg yLeftleg xRight leg 100 y Right leg 300 200 100 o Acceleration L ph o S Acceleration I a 3 200 100 300 150 400 200 500 i i i i i 250 i i i i i 0 10 20 30 40 50 60 0 10 20 30 40 50 60 Test Duration Test Duration a X axis b Y axis Pivot Shift Test Beta Tue Feb 24 13 51 56 2015 art93 p002pi csv art93 p002pd csv 7 Leftleg ZzRightleg 400 300 200 100 0 Acceleration 100 200 300 400 i i i i i 0 10 20 3
15. 5 8 Maximum and minimum values for patients by groups left and right legs 5 7 Classifying patients from data After showing the statistic properties of the data from patient trails this section focuses on seeing up to what degree is possible to tell from a healthy patient and 89 Chapter 5 Data Acquisition and Analysis injured patient based on the metrics provided by the data presented in the previous section Different approaches were taken in consideration from taking the mean minimum maximum or a combination of the three as main parameter to perform the classification The following experiments that we will call metric classifiers make use of the magnitude of the acceleration 28 can be define as follows MaGace VX Y Z2 5 3 which combines the three acceleration components into one 5 7 1 Metric classifiers The first attempt takes as main parameter the mean value of the acceleration pre sented on each of the patient groups variable mean sd median avg min avg max kurtosis se Mag H L 0 7603 0 4640 0 6664 0 2739 1 5847 2 0334 0 0049 Mag H R 0 8060 0 4577 0 7215 0 2666 1 7663 0 1010 0 0048 Mag_ACL_L 0 7878 0 2098 0 7700 0 3478 1 1516 0 5318 0 0094 Mag_ACL_R 0 6655 0 4921 0 5348 0 1698 1 5837 0 0985 0 0216 Table 5 9 Magnitude metrics summary for healthy and ACL patients For example Mag_H_L represents magnitud
16. Chapter 2 The Pivot Shift Prototype Hardware Design C plates proot4nass Spot Magn Det WD 34 20 um 1010kV 30 650x SE 128 ST Micro LIS331DLH CW TB Figure 2 4 Three Axis accelerometer LIS331DLHXY used in iPhone 4 in detail micromachined proof mass interleaved with capacitive sensors In this kind of gyroscope MEMS the resulting physical displacement is read using capacitive sensors These gyroscopes use a tuning fork configuration in which two masses are placed side to side they oscillate and move constantly in opposite directions When angular motion is applied the Coriolis force on each mass acts in opposite direction The difference between their capacitance is then measured to produce the rotation metric As counter example when linear acceleration is applied to the two masses both move towards the same direction this produces a differential capacitance of zero which makes the measurements of the gyroscopes not to be affected by acceleration forces Similarly to the accelerometers MEMS gyroscopes can be found in single axis or three axis presentations Since they measure the angular rate or velocity the output 19 Chapter 2 The Pivot Shift Prototype Hardware Design dt Fe zoriolts 2mO No Figure 2 5 MEMS gyroscope submitted to Coriolis effect unit is often dps degrees per second RPMs revolutions per minute and radians per second 2 2 3 MMA7361LC Three Axis Low
17. acceleration 6 0 183 mg LSB Linear acceleration measurement range Linear acceleration RENA Linear acceleration 8 0 244 Linear acceleration 16 0 732 Linear acceleration sensitiv From 40 C to 85 C 1 5 ity change vs temperature Linear acceleration typical 60 mg zero g level offset accuracy Operating temperature 40 to 85 C range Table 2 2 LSM9DSO Accelerometer characteristics Before continuing lets define some of the terminology showed on the previous tables When talking about sensors it is common to see terms like range sensitivity and zero levels 2f Chapter 2 The Pivot Shift Prototype Hardware Design Parameter Test conditions Specification Unit 245 Angular rate et range zn ape 2000 RA Angular rate 245 8 75 T senate Angular rate 500 17 50 mdps digit Angular rate 2000 70 Angular rate sensitivity From 40 C to 85 C 2 change vs temperature 245 dps 10 500 dps E25 Table 2 3 LSM9DSO Gyroscope characteristics The range represents the levels the sensor s output signal can achieve maximum and minimum values that in regards of the accelerometer is expressed in g with option of five different ranges to choose from For instance if the accelerometer is working in the 2g range this means that if a 4g force is applied it will not be properly displayed by t
18. and considering that by this point two major operations have been perform to the orginal or raw dataset e X axis compensation due internal tibial rotation e Conversion from ADC units to g s The complete dataset consist of the right left pair of csv files containing the tests results for 140 patients As mentioned in chapter 4 each of the patients was clini cally evaluated by the medical staff and classified the patients in three main groups healthy with observations and ACL patients were patients with observations refer to those presenting some leg or knee previous condition unrelated to a torn ACL Third group consisted on patients with ACL injury The final distribution of the patients was the following ALL Healthy Observation ACL 140 100 32 8 Table 5 1 Patients distribution leaving a 71 496 of healthy patients 22 patients with observations and 5 7 patients with ACL injury The overall total of acceleration samples produced by the prototype during the 140 tests came to 27605 XYZ triplets 18188 form healthy patients 7128 patients with observations and 2289 from ACL patients 82 Chapter 5 Data Acquisition and Analysis X Axis Y Axis ce e S e LO pem e o q LO 2 e S e e o eH DU m c e c 1 o n o e o o o o e Ls bh O e LO I T T 1 I I I l 4 3 2 1 0 1 2 3 2 1 0 1 2 acceleration acceleration a X axis
19. and implements a wide variety of functions and libraries aimed for computer graphics techniques like vector raster drawing color models image processing network communication and object ori ented programming Custom made libraries can extend the functionality to generate sounds send and receive data in different formats and enhance 2D and 3D file for mats All programs created on Processing are a subclass of the PApplet Java class additional classes are treated as inner classes when the code is translated to Java code before compilation Processing allows two main modes the Java and Javascript mode applications in Java are executed as a standalone window applet whereas the Javascript mode translates the code to work and run in a browser Applications can be exported to Linux Mac and Windows operating systems with the option to embed Java runtime which is recommended to ensure the same performance between platforms Core libraries and code used in the exported application and applets are licensed under GNU Lesser General Public License allowing the programmer to release their code with license of choice 21 There is extensive community support and resources to learn this programming language recommended sites for this purpose are https processing org and http www openprocessing org 42 Chapter 3 The Pivot Shift Prototype Software Design 3 2 The Basics Here we describe the particularities of the programs c
20. classification results of future patient trials 107 Appendix A Arduino and Processing Code Code repository https maespinozas bitbucket org maespinozas pivotshift git 108 References 1 V Duthon C Barea S Abrassart J Fasel D Fritschy and J M n trey Anatomy of the anterior cruciate ligament Knee surgery sports traumatology arthroscopy vol 14 no 3 pp 204 213 2006 Online Available http link springer com article 10 1007 s00167 005 0679 9 J H Mink T Levy and J Crues 3rd Tears of the anterior cruciate ligament and menisci of the knee MR imaging evaluation Radiology vol 167 no 3 pp 769 774 1988 Online Available http pubs rsna org doi abs 10 1148 radiology 167 3 3363138 W H A Ng J F Griffith E H Y Hung B Paunipagar B K Y Law and P S H Yung Imaging of the anterior cruciate ligament World journal of orthopedics vol 2 no 8 p 75 2011 Online Available http www ncbi nlm nih gov pmc articles PMC3302044 S Arneja and J Leith Review article Validity of the KT 1000 knee ligament arthrometer Journal of Orthopaedic Surgery vol 17 no 1 2009 Online Available http www josonline org index php JOS article view 432 http www josonline org index php JOS article view 432 375 S Wiertsema H Van Hooff L Migchelsen and M Steultjens Reliability of the KT1000 arthrometer and the Lachman test in pat
21. elsef myPort write a 1 accelerometer mode myPort write 1 50 Chapter 3 The Pivot Shift Prototype Software Design This code is executed when the event on the Start toggle is activated when pressed it checks the state of the other toggle called OpMode right next to Start toggle to see if the user chose one accelerometer or two A boolean value true for OpMode means using two accelerometers as false specifies to use only one Now the embedded system expects two strings the first one indicates the mode a for 1 accelerometer and b for two and a start stop flag used to begin or finalize the test 1 starts as 0 terminates 3 3 3 During the test Once the embedded system starts sending data from the sensing modules the impor tance resides in properly display it and store it for future analysis The strings the Arduino sends including the sensor s readings have an specific format illustrated by the following examples for acceleration these are valid strings e r 1 26 e y 0 98 e z 1 08 and for rotation e gx 25 50 e gy 15 00 e gz 85 96 where x y and z represent acceleration and gx gy and gz rotational metrics from the gyroscope Therefore the string is formatted in a metric space value manner 51 Chapter 3 The Pivot Shift Prototype Software Design The strings are parsed during the serial port event in the application dividing them in two sections the firs
22. g Accelerometer The MMA7361LC was the accelerometer used in the first pivot shift prototype and from which came many of the test results to be analyzed in following chapters Since this is an analog device the output signals are voltages proportional to the measured acceleration The main features of the device are selectable sensitivity 1 59 6g acceleration sensing for X Y and Z axis 2 2 to 5 supply voltage range e low current consumption 400uA 20 Chapter 2 The Pivot Shift Prototype Hardware Design Ci Fcoriolis F coriolis AC 0 Figure 2 6 MEMS gyroscope submitted to angular motion Fac Ac 0 Figure 2 7 MEMS gyroscope submitted to linear acceleration e 800 mV g sensitivity at 1 5g range e low cost The interaction between the accelerometer and the Arduino is as follows The accelerometer is powered up to 3 3 volts provided by the Arduino therefore reference voltage is 3 3v Out of the box the MMA7361LC works on sleep mode reducing the operating current to 3uA which is great for power saving purposes but turns off the output signals In order to disable sleep mode and turn on the x y and z outputs sleep mode pin 7 should be set high which was achieved using a pull up resistor 21 Chapter 2 The Pivot Shift Prototype Hardware Design Figure 2 8 MMA7361LC analog accelerometer breakout board showing respective directions for each axis The three axis outputs are then c
23. groups Statistical metrics were computed to compare results among groups Later on some of these metrics were implemented as classification parameters Over all error rate was calculated for each of the different tries Restricting the use of only X and Z axis data backed up in the idea that X reports the internal tibial rotation and Z takes care of the anterior posterior tibial translation with respect of the fe mur became the approach leading to the minimum error rate Optimal results out of these classifiers had an average error rate of 3096 approximately for both left and right legs After performing these tests and with the intention of implementing a non para metric classification method pattern recognition k nearest neighbors algorithm was used as classification rule Reported results coming from this approach proved to be 106 Chapter 6 Conclusion and Future Work more effective reaching 25 and 15 error rates for left and right legs respectively Data is a precious resource and in a time period little over than a year data results from over 150 patients were collected Getting data of injured patients proved to be challenging but still possible Nevertheless given the results and the fact that those were purely based on acceleration readings from the very first prototype grows the curiosity of seeing how the addition of rotational data from the gyroscopes can or cannot add more significant features towards improving the
24. repeats for this test four times as seen in the graph The image showed for this example was a left leg test The same visual interpre tation can be assumed for the right leg results taking in consideration the obvious difference between the internal tibial rotation reflected on the X axis from one leg to the other they have opposite direction As showed in previous side to side re 75 Chapter 5 Data Acquisition and Analysis sults in figure 5 4 left side rotation is measured as a X acceleration while its right counterpart as X acceleration force 5 3 Adjusting Left leg X axis readings So far it has been stated that the most significant difference visually in the results of a pivot shift test between a left and right leg relates to difference found in the X axis results which are produced by the direction of the internal tibial rotation that each knee has during the test Considering that many of the statistical calculations to be presented in following sections involve side to side comparisons of tests results e g looking at the difference in maximum and minimum acceleration readings between legs or getting the mean value of the acceleration for all patients X axis both legs If those calculations were to be made taking the data as it is it would be no problem in directly comparing side to side values taken from Y and Z axis since they tend to generate the same movement from back and forth leg movement and lifting
25. scheme in 96 Chapter 5 Data Acquisition and Analysis Healthy Right vs ACL Right healthy o acl N EN o N Figure 5 16 XZ datasets corresponding to right leg which the object to classify calculates the k nearest samples to itself The called or resulting class will be the one presenting the highest number of occurrences in those k neighbors Figure 5 18 illustrates an example in which one patient is being classified right leg ACL patient The upper left image shows the training sets for both classes healthy and injured upper right image adds to it the testing set corresponding to the patient to be classified blue The image at the bottom contains a portion of the upper right figure zoomed Here we can see a testing point in blue If we were to apply knn with k parameter equal to 3 the resulting class would be the ACL or injured class this because out of the 3 nearest neighbors to the sample point 2 belong to the injured class and only one to the healthy class therefore ACL is the class with more occurrences in the neighbors Due this voting nature of the algorithm is recommended to use only odd 97 Chapter 5 Data Acquisition and Analysis Healthy Right vs ACL Right Healthy Right vs ACL Right healthy ad o healthy 2 healthy ancl ad esl ad T T T T T T T T T 2 1 0 1 2 2 1 0 4 2 Figure 5 17 Creating subse
26. the sense that no ionizing radiation x rays is used Instead MRI makes use of powerful magnetic field and radio frequency pulses to produce pictures of tissues organs and bones Protons hydrogen atoms in tissues containing water molecules are used to create a signal that after being processed forms an image of the body Energy from the magnetic field at a certain resonant frequency is applied to the patient Then the excited protons emit a radio frequency signal that is measured by a receiver coil This radio signals can be used to encode position information by changing the magnetic field using gradient coils The contrast between different tissues is determined by the rate at which the excited atoms go back to an equilibrium state Similarly to the CT scan contrast agents may be used to provide better image results MRI results in better quality images of soft tissues like the ACL having a high rate on ACL tear diagnosis with accuracy sensitivity and specificity of more than 90 2 As well MRI is also best for detecting concomitant mensiceal ligamentous or chondral injuries 3 Chapter 1 Introduction Figure 1 4 MRI Acute anterior ligament tear KT 1000 This test developed by Dale Daniel MD back in the 1980 s has been widely used over the years for the diagnose and follow up on patients with the ACL injury This arthrometer is an objective instrument for the ACL reconstruction which measures the anterior tibial
27. 0 40 50 60 Test Duration c Z axis Figure 5 3 Patient s side to side acceleration comparison for each axis These group of plots show the acceleration differences on each of the axis X Y Z between left and right leg of the patient Although one limitation of the data produced during the trails is noticed left and right leg tests will not exactly match in the sense that do not equal in number of samples This is caused due the fact 70 Chapter 5 Data Acquisition and Analysis that is practically impossible for the the physician to execute both maneuvers pivot shift for right and left leg with the exact same timing that is why an offset between the two tests can be seen as well Despite this limitation it will be shown how this overlapping plots still are useful for the analysis After reviewing this new graphs over several patients showing side to side com parison for each of the axis separately one thing became quite noticeable Recalling the pivot shift technique we can consider its two main components as the rotation and valgus force applied to the knee This rotation is called internal tibial rotation which makes reference to the direction in which the movement occurs and can be described as a rotation towards the direction of the other knee That being said e g the internal tibial rotation in the left leg is seen as rotating the leg towards the right leg Now that internal tibial rotation was defined we can go f
28. 12 to 1023 reflect positive forces while ADC values from 0 to 511 negative accelerations To make this negative and positive ac celerations ADC units representation more straightforward the following calculation was performed for each sample ADC out m ADC out 512 5 2 this simple subtraction makes negative accelerations be in the 0 to 512 range and positive accelerations in the 0 to 512 range This explanation provides context to understand the reason behind the accel eration values shown so far however in order to provide meaningful results to the calculations and metrics to be presented in the following sections it is required to step away from the ADC units and make use of an standard unit of acceleration In this case the complete dataset was transformed to g s remembering that 1g 9 8m s Next snippet shows the function created to convert the complete dataset to g s using the R programming language 79 E w o CON Oo O Chapter 5 Data Acquisition and Analysis HHHCHANGE FROM ADC VALUES TO G S FOR MMA ACCELEROMETER call example datahealthyG togs testdata healthy togs lt function dataset out dataset for i in 1 length dataset for each test file ADCVAL 512 VREF 1023 ZEROV SENSIBILITY out L ill 11 2ssppiy Cdataset illLl c C1 unction i t Y values 14512 3 3 1023 1 65 8 out il 2 sapply dataset illl c 2 function i 4t Y values i 512 3 3 1023 1 6
29. 283 0 2016 0 1673 1 0948 1 2621 0 0136 ALLZmin 0 6659 0 4306 0 6391 1 7278 0 3649 2 0927 0 0257 ALLZmax 0 9818 0 5261 0 9940 0 3085 2 0181 2 3266 0 0314 Table 5 4 Maximum and minimum acceleration metrics These calculations came from summarizing information like the one presented in the following table which shows the X Y and Z maximum and minimum acceleration values per patient xmin xmax ymin ymax zmin zmax patient 1 0222 0 3851 0 8327 0 0020 0 4859 2 0181 p0O1pd csv 0 9052 0 3891 0 8488 0 2923 0 8810 0 6794 p00lpi csv 0 7923 0 9294 0 4012 0 2964 0 3367 1 6714 p002pd csv 0 7601 0 5383 0 1794 0 2399 0 7157 0 4657 p002pi csv 1 4778 0 8609 1 3165 0 0423 0 3690 1 5665 p003pd csv 1 3488 0 4536 0 5948 1 0544 1 2117 0 4617 p003pi csv 0 8810 1 2762 0 6956 0 0585 0 3851 1 8367 p004pd csv 0 7762 0 4133 0 3609 0 3972 0 8448 0 8004 p004pi csv 1 0060 0 7117 0 6754 0 5141 0 4133 1 5060 p005pd csv 0 8407 0 1875 0 1794 0 1956 0 9133 0 1069 p005pi csv Table 5 5 Example table showing maximum and minimum values for 5 patients 5 6 1 Left and Right legs statistics Unlike the test using the KT 1000 arthrometer that requires to be performed in both right en l
30. 284 HZmax L 0 5715 0 3778 0 5101 0 3085 1 5746 1 8831 0 0378 HXmin R 1 0793 0 3295 1 0282 1 8327 0 3246 1 5081 0 0329 HXmax R 0 8812 0 4036 0 8589 0 3931 1 9778 2 3710 0 0404 HY min R 0 8871 0 3879 0 7843 1 8165 0 3327 1 4839 0 0388 HYmax R 0 2474 0 2239 0 1875 0 0706 0 9496 1 0202 0 0224 HZmin R 0 4163 0 3365 0 3790 1 6149 0 3649 1 9798 0 0336 HZmax R 1 8745 0 2965 1 4234 0 7560 1 9536 1 1976 0 0296 ACLXmin L 0 9753 0 5346 0 9172 2 0867 0 3528 1 7339 0 1890 ACLXmax L 0 3891 0 7380 0 3448 0 8206 1 4214 2 2419 0 2609 ACLYmin L 0 5599 0 3876 0 4980 1 1593 0 1593 1 0000 0 1370 ACLYmax L 0 2927 0 2866 0 1573 0 0948 0 8800 0 7852 0 1013 ACLZmin L 0 9959 0 3639 1 0442 1 5343 0 4133 1 1210 0 1287 ACLZmax L 0 3778 0 2829 0 2774 0 0867 0 7278 0 6411 0 1000 ACLXmin R 0 8458 0 2633 0 8649 1 2238 0 4093 0 8145 0 0931 ACLXmax R 0 8982 0 3452 0 9315 0 3891 1 4335 1 0444 0 1220 ACLYmin R 0 9526 0 4404 0 9254 1 6351 0 3931 1 2419 0 1557 ACLYmax R 0 3407 0 2695 0 2238 0 0907 0 8931 0 8024 0 0953 ACLZmin R 0 2273 0 2995 0 2621 0 6633 0 1673 0 8306 0 1059 ACLZmax R 1 4068 0 2524 1 4315 1 0746 1 8770 0 8024 0 0892 Table
31. 40 A valgus torque and internal rotation are applied to the leg A positive pivot shift test is defined as a forward subluxation Chapter 1 Introduction Figure 1 7 Pivot Shift test of the tibia during the change of direction This clinical test tries to reproduce the event when an ACL tear occurs when the knee gives way due the loss of the ACL 6 The pivot shift is a relatively complex test For this reason and the lack of a quantitatively accepted method to measure the results 7 is evaluated only on the basis of the examiner s experience 8 Still with this limitations the test is considered one of top three applied techniques towards the diagnosis of an ACL condition part of this has to do with the fact that contrary to the Lachman and Drawer tests the Pivot Shift is the only of the three that analyses the dynamic and rotational stability of the anterior cruciate ligament which more accurately relates to the knee function The complexity of the test using rotational and valgus forces makes the quanti tative analysis more challenging in comparison with uniplanar stress testing like the one present in the Lachman test that can be measured with the KT 1000 For this reason several studies 7 9 have tried to decompose the pivot shift movement into quantifiable parameters suggesting e anterior posterior translation e anterior posterior acceleration 10 Chapter 1 Introduction e anterior posterior rotation
32. 5 8 out ill 3112sa8ppiy dataset LiJ c 3 function i Z values 1 512 3 3 1023 1 65 8 This code applies the already covered procedure to get g s from chapter s 2 equa tion 2 3 5 5 Tools and Packages used The list of software packages and tools used to perform the majority of the calcula tions were the following e Python 2 7 9 python2 matplotlib 1 4 2 3 25 python2 numpy 1 9 1 1 26 e RStudio 0 98 1091 r 3 1 2 1 80 Chapter 5 Data Acquisition and Analysis pysch 1 5 1 library 27 Qt 4 8 6 e libreoffice 4 3 5 2 with hardware specifications mrhyde arch76 lscpu Architecture x86_64 CPU op mode s 32 bit 64 bit Byte Order Little Endian CPU s 8 On line CPU s list f Thread s per core Core s per socket Socket s NUMA node s Vendor ID enuineIntel CPU family Model 58 Model name Intel R Core TM i7 3740QM CPU 2 70 GHz Stepping CPU MHz CPU max MHz CPU min MHz BogoMIPS Virtualization Lid cache Lii cache L2 cache L3 cache NUMA nodeO CPU s Memory running a Linux distribution Linux arch76 3 18 6 1 ARCH 1 SMP PREEMPT x86 64 GNU Linux LSB Version 1 4 Distributor ID Arch Description Arch Linux 8l Chapter 5 Data Acquisition and Analysis Release rolling 5 6 Data Statistics This section starts to statistically described the data obtained from patients trials
33. Marco Antonio Espinoza Sanchez Candidate Electric and Computer Engineering Department This thesis is approved and it is acceptable in quality and form for publication Approved by the Thesis Committee Wei Wennie Shu Chairperson Thomas P Caudell Marios S Pattichis Design amp Implementation of a Pivot Shift Prototype for Quantitative Analysis by Marco Antonio Espinoza Sanchez B S Chihuahua Institute of Technology 2011 THESIS Submitted in Partial Fulfillment of the Requirements for the Degree of Master of Science Computer Engineering The University of New Mexico Albuquerque New Mexico May 2015 2015 Marco Antonio Espinoza Sanchez ill Dedication To my family and friends which always supported me through the University years I probably wouldn t be writing this thesis without their help lv Acknowledgments I would like to thank and sincerely acknowledge my Thesis advisor Dr Wei Wennie Shu for the guidance and support provided during the realization of this work I would also like to thank Dr Thomas P Caudell and Dr Marios S Pattichis for being part of the thesis evaluation committee As well I would like to express my gratitude to CONACYT Consejo Nacional de Ciencia y Tecnolog a for all the support to make this happen Special thanks to my friend Engineer Jose Diaz for the help with the design and support of the prototype and to the medical staff leaded by Edmundo Beru
34. Selectable axis the user can display acceleration of three axis X Y and Z or choose which ones to use e Ability to overlap two graphs useful to compare left and right legs in a same patient e Graphs can be zoomed and saved to png svg and pdf formats e Load all csv files from a given directory to easily plot different files e Works both for graphs produced by the MMA7361LC prototype only acceler ation and LSM9DSO prototype acceleration gyro 68 Chapter 5 Data Acquisition and Analysis Pivot Shift test Plot Analyzer so Acceleratio Angulation Overlap 2 Grap Axis Axis EX w Left knee art3a p002pi BY P Right knee art3a p002pd c BZ Plot Working directory home mrhyde ownCloud ece Thesis Pivot Shift te art3a p001pd csv art3a p001pi csv art3a p002 pd csv art3a p002pi csv art3a p003pd csv art3a p003pi csv art3a p004pd csv art3a p004pi csv art3a p002pd csv Figure 5 1 Pivot shift plot analyzer with this side to side comparison was done in patients to spot if obvious patterns were present Next figure shows a patient left and right resulting acceleration graphs Pivot Shift Test Beta Tue Feb 24 13 46 17 2015 Pivot Shift Test Beta Tue Feb 24 13 46 13 2015 200 art93 p002pi csv p art93 p002pd csv X x 30D y 300 mt i z z 200 o 100 100 E d UR H 200 2 t 100 300 200 400 ESO Es 10 20 30 40 50 60 ions 5 10 15 20 25
35. The Pivot Shift Prototype Hardware Design TxO Vin RXI GND RST RST GND Vcc 2 A3 3 A2 4 A1 9 AO 6 13 T 12 8 11 9 10 Figure 2 1 Arduino Mini Pro Low voltage board 3 3V Thin PCB 0 8mm 3 3V regulator e Max 150mA output DC input 3 3V up to 12V e Power and Status LEDs Analog Pins 8 Digital I Os 14 The main and most important component is the Atmega328p microcontroller which is a 8 bit AVR RISC based microcontroller 15 that combines 32 KB ISP flash 14 Chapter 2 The Pivot Shift Prototype Hardware Design memory 2 KB SRAM 1 KB EEPROM 23 general purpose I O lines 32 general purpose working registers internal and external interrupts support three timers a Serial programmable USART SPI and IC port a 10 bit A D converter All with low power requirements from 2 7 5 5 volts 2 2 MEMS Accelerometer and Gyroscope MEMS Microelectromechanical systems is a process technology to bundle together mechanical and electrical components in tiny integrated devices or systems They are fabricated using integrated circuit IC batch techniques and can range in size from a few micrometers to millimeters MEMS increased their popularity in the last few years and now can be seen in multiple areas such as automotive medical electronic communication etc Exam ples are e accelerometers for airbag sensors e inkjet printer heads e hard drives read write heads e microvalves e blood pressure sensors
36. able 5 15 KNN missclassifications rates by groups The average error found in the correct classified patients was of 16 36 including both legs and classes Average error in the incorrectly classified patients was of 46 44 for both classes and legs Regarding the ACL patients all of the left legs were correctly classified reason why avg miss error is zero as the right counterpart missclassified 1 out of the 4 patients 101 Chapter 6 Conclusion and Future Work 6 1 Limitations and Future Work In the course of developing this project and specially at the moment of starting the patient trials and data analysis some limitations were found The fact that this project involves both medical and engineering staff and that the prototype was finally implemented by the end user medical team really helped up the project towards future development since the staff came up with several improvement ideas that would not have been identified if the prototype had not been implemented Some of them involved minor hardware changes in the prototype like a reset button or the use of an LED to display battery status to little more complicated like going wireless since the prototype needed to be directly connected to the pe via an USB cable This was resolved by adding a pair of XBEE modules 31 based on the IEEE 802 15 4 which are fairly ease to use As well there were software related observations like making the application able to generate the g
37. ansition on the data line SDA while the clock line stays high 2 2 6 Getting data from the LSM9DSO This section shows the required connections to be made between the LSM9DSO and the Arduino board in order to communicate The following schematic shows a configuration of two LSM9DSO connected to the Arduino Mini board using the 3l Chapter 2 The Pivot Shift Prototype Hardware Design previously introduced I C communication protocol Figure 2 13 Schematic diagram of two LSM9DSO s connected to an Arduino Mini board under IC Compared to the connection of the MMA7361 back in figure 2 9 this one looks pretty straightforward considering that includes two LSM9DSO modules In this configuration the Arduino acts as master for the I C leaving the LSM9DSO s as slaves The two LSM9DSO boards share the bus for the clock and data signals SCL and SDA The clock signal is controlled and generated by the Arduino while the SDA can be used to send and receive data from the accelerometers Initially both LSM9DSO boards have the same address which makes impossible to the master to differentiate from each other To solve this the least significant bit of the address of one of the modules is changed by setting pins SDOG and SDOXM to a low state Arduino board and inertial modules are powered up by 3 3 Volts 32 Nec a dee ENC Ap Qoo DAU p o N N N e OO N WN o pe C 0000 Q0 C0 Ww C o C N Chapter 2 The Pivot S
38. avity produces a downward deflection on the proof mass that appears equivalent to an upward acceleration of the sensor at lg Free fall shows a similar situation in which F F and there is no displacement of the proof mass therefore no acceleration is measured output of the accelerometer is zero despite the fact that the device is accelerating at 9 8 m s due the gravity This phenomenon is a consequence of the statement from Newton s law that the sum of the gravitational and inertial forces equals to zero on an body in free fall From the previous we can conclude the following e when the accelerometer lies at rest it will output an a 1g due gravity e when the device is on free fall the acceleration outputs a 0 2 2 2 Gyroscope A gyroscope is a device used to measure rotation and detect inertial angular motion There are many kind of gyroscopes which can work based on different principles for example the mechanical kind of gyroscope that measures the Coriolis force applied to a body in a rotating frame MEMS gyroscopes typically use vibrating structures because of the size constrain that makes difficult to incorporate micro machined rotating parts with the required mass 17 MEMS gyroscopes use the Coriolis effect Taking as example figure 2 5 consider a mass moving in v direction When an angular motion is applied red arrow the Coriolis effect makes the mass experience a force in the direction of the yellow arrow 18
39. aximum values in the original space in this case if the metric is acceleration then low1 should be the minimum value of all the acceleration samples stored so far as high1 works in the same way taking the maximum The arguments low2 and high2 represent the new range in which the value will be mapped into In short the function takes a numeric value and transforms it from the scale low1 high1 to a new value in the range from low2 high2 For example if the current value for acceleration is 1 25 g s then map function has to be called twice to get the corresponding X and Y position in pixels of the sample inside the plotting area xplot map 1 25 1 arraytime size plotx1 plotx2 X position in pixels yplot map 1 25 minval maxval ploty2 ploty1 Y position in pixels For the X axis the original boundaries are between 1 the minimum number of samples and the total number of samples The new range is defined by the width of the plotting area that is between plotx1 plotx2 The Y axis goes from the range defined by the minimum and maximum acceleration values to the range between the points ploty2 ploty1 To finally reproduce the graph of the three axes in real time a function called Plotarray arraytime arraymetric strokecolor is defined The first argument ar raytime is the array holding the values for the horizontal axis arraymetric can be one of six choices arrayz arrayy arrayz arraygx arraygy or arr
40. aygz storing the 95 Chapter 3 The Pivot Shift Prototype Software Design acceleration or rotation of one of the axes The strokecolor parameter specifies the color of the line to be drawn Therefore in order to generate the continuous graph of the acceleration metric the following piece of code needs to be called inside the draw loop PlotArray arraytime arrayx R PlotArray arraytime arrayy G PlotArray arraytime arrayz B leading to a graph like this Figure 3 6 Sample acceleration graph x axis in red y axis green and z axis in blue since the Plotarray function is called during the main loop it automatically scales the graph to the values stored in the data arrays The image shows the resulting acceleration in g s signals for each of the axes during the test 3 3 4 From angular velocity to angular displacement Recalling the features of the sensing module LSM9DSO from the previous chapter we know that incorporates both an accelerometer and a gyroscope The acceleration is presented as g s while the angular velocity is presented in dps degrees per sec ond Angular velocity is defined as the rate of change in angular displacement while 56 Chapter 3 The Pivot Shift Prototype Software Design angular displacement is the angle through which a body is rotated in a specific sense direction and axis This section shows the procedure taken to convert gyroscope output from angular ve
41. celeration values registered during the pivot shift movement As mentioned earlier for each of the tests the pivot shift movement was reproduced three times in a row reason why all the graphs present three similar spikes in the signal in order to asses repeatability of the acceleration readings Taken that into account there were some cases in which one of the three maximum or minimum values was significantly different in comparison of the other two Pivot Shift Test Beta art3a p006pd csv ZMaxMedian Acceleration Test Duration Figure 5 12 Taking the median of the maximum values For this reason as shown in 5 12 it was decided to take the median of the maximum and minimum acceleration values while computing these metrics e g while getting the maximum Z acceleration for the patient file used in this figure art3a p006pd csv the reported value will be the median represented in this case by Zmaz2 This prevents outline readings like Zmax3 to affect statistics in further calculations 86 Chapter 5 Data Acquisition and Analysis The metrics computed based in the maximum and minimum acceleration values per axis are displayed in table 5 4 variable mean sd median min max range se ALLXmin 0 9307 0 4683 0 8952 2 2722 0 6915 2 9637 0 0280 ALLXmax 0 8023 0 5497 0 8044 1 4335 2 7802 4 2137 0 0329 ALLYmin 0 8184 0 4252 0 6976 1 9214 0 0706 1 9919 0 0254 ALLYmax 0 2622 0 2
42. ch makes the application not to depend on specific hardware In this 52 Chapter 3 The Pivot Shift Prototype Software Design application it is set to have a size of about 85 of the screen width and height Then after size is defined two variables are automatically set width and height as we can see in the code size round displayWidthx 85 round displayHeight 85 Single graph layout plotX1 120 plotx2 width 120 plotY1 100 70 plotY2 height 85 Dual layout plotY1 1 plotYi1 plotY2 1 plotYi plotY2 40 end of first plot area plotYi1 2 plotY2 1450 start of second area for plot plotY2 2 plotY1 2 plotY2 40 end of second plot area the size function takes two parameters being the width and height What the code does is setting variables that represent corners of the rectangle enclosing the plotting area As the application is set to show acceleration or acceleration and rotation two layouts are predefined This code is found inside the setup function called at the beginning of the sketch and having those dimensions already defined facilitates switching layout during runtime The next screen shot illustrates the position of the rectangle s vertices for the plotting areas Function rect a b c d is used to draw the rectangles each parameter describing e a x coordinate of rectangle upper left corner by default b y coordinate of rectangle e c width of the rectangle d height of the recta
43. d 82 5 6 1 Left and Right legs statistics dod x RR ae oe ga 87 5 7 Classifying patients from data 2l 89 Dora Metric classifiers Seu bestem eue m ee ee Ra A 90 5 7 2 Non parametric classification K Nearest Neighbors 95 6 Conclusion and Future Work 102 6 1 Limitations and Future Work 4 uk gave sa quaeso Vp eee due ds 102 02 Conclusion sus tue ee sur oue eos Rhe doas x Aou do seit cubus mock Mb dd 106 A Arduino and Processing Code 108 References 109 List of Figures 1 1 1 2 1 3 1 4 1 5 1 6 1 7 2 1 22 2 3 2 4 2 5 Normal knee anatomy front view 2l ls Frontal view of a knee with ACL injury Arthroscopy equipment and setup on patients knee MRI Acute anterior ligament tear 2 4 KT 1000 test measuring anterior posterior knee translation Clinical assessments for the ACL diagnosis Pivot Shitt testis ps qe Bord BE ee A eon wed en oe ed Arduino ND 2D BO 1 be eat 42 33 Meu Sn Me uf RR oes eee MEMS components diagram Piezoelectric accelerometer spring moves as acceleration forces act upon the sensor producing a voltage proportional to that force Three Axis accelerometer LIS331DLHXY used in iPhone 4 in detail micromachined proof mass interleaved with capacitive sensors MEMS gyroscope submitted to Coriolis effect 10 14 16 17 19 Li
44. e containing all the data generated during the test Open previous graph button allows the user to open an csv file from a previous test and display its graph 3 3 2 Beginning the test Once the application is launched and the visual environment is setup then the serial communication between the embedded system and the PC begins Serial myPort 2 tryt 49 Chapter 3 The Pivot Shift Prototype Software Design myPort new Serial this portName 19200 myPort bufferUntil n println Starting test catch Exception e println No serial port available This sample code initializes the serial port using a baud rate of 19200 bps The variable portName holds the name of the serial port selected from the dropdown menu located in the upper left part of the UI At this point the embedded system waits to receive some initialization strings from the serial port indicating how many accelerometers to use and when to start sending sensor data if theEvent controller name start Start the test if start true timestart millis if first_time true println Selected serial port dropDownItemName portName dropDownItemName myPort new Serial this portName 19200 Cleargraph clear data arrays to start new test first time false myPort bufferUntil n if CopMode true opMode Operation mode myPort write b 2 accelerometers mode myPort write 1
45. e is transmitting or receiving from the slave while a 9th bit is used as NACK ACK bit When this address is sent then all devices compare it with their own to check if they 30 Chapter 2 The Pivot Shift Prototype Hardware Design are being addressed by the master Acknowledge is mandatory for this protocol both for the data and address frames Once the first 8 bits address and r w are sent the receiving device is given control over the SDA line To successfully generate the acknowledge the receiver needs to pull low the SDA line prior the 9th clock pulse iit xi ci TAA MU Wee start conditio a E Mast is equ uasi ng data ACK NACK A 1 in this position Stop condition SDA goes lo iw ba fore SCL indicates anita iddre SDA goes high after SCL d not respi id ores unable process the request SDA Figure 2 12 I C signals showing the start condition address and data frames and stop condition Data transmission can start after the address frame has been sent During data transmission the master continues the with the control and generation of the clock pulses while the data is placed in SDA line by the master or the slave depending the value of the r w bit 8th bit of address frame The number of data frames is arbi trary and most slaves will auto increment the internal registers to allow subsequent reads or writes The operation of the protocol end with a stop condition defined by a low to high tr
46. e metrics for left leg healthy patients The first metric first column in the table is the result of taking the mean value of the acceleration in each of the patients in this case the total hundred healthy patients and taking the average of that quantity 90 Chapter 5 Data Acquisition and Analysis The following snippet shows the coded function Python meanClassifier which was applied to all patient files which takes as an input the patient tests file csv file generated from the software described in chapter 3 and the specification of if it is a right or left leg test file def meanClassifier testfile leg left mag xyz print filename testfile data getdataMag testfile mag datamean np mean data meanaclL 0 7878379 meanaclR 0 6654715 meanHL 0 7603489 meanHR 0 8060259 evaluate the test if leg left delta acl abs datamean meanaclL delta h abs datamean meanHL if delta acl lt delta h print DataMean t str datamean t MeanACL_L t str meanaclL t Delta_acl t str delta_acl t ACL else print DataMean t str datamean t MeanH_L t str meanHL t Delta_h t str delta_h t Healthy else delta acl abs datamean meanaclR delta h abs datamean meanHR if delta acl lt delta h print DataMean t str datamean t MeanACL_R t str meanaclR t Delta_acl t str delta_acl t ACL else print DataMean t str datamean t MeanH_R t str meanHR t Delta_
47. e pivot shift test e KT 1000 test The examiners group was small consisting of 2 physicians and the lead surgeon Physician were instructed by lead surgeon on the execution of the pivot shift tech nique in order to perform the test as similarly as possible The pivot shift shift test was perform three times in each knee for every patient with the intention of assessing test repeatability Patients with a previously con firmed ACL diagnosis where submitted to special treatment and had the test done under anesthesia For each of the patients clinical evaluation of the pivot shift maneuver was de termined manually by the physician and documented to keep track on patients with positive results to later compare with prototype results As for the prototype pa tient s results consisted in two images and two csv files corresponding to each leg acceleration readings The KT 1000 test was perform in both of the patient s legs documenting the indi vidual result for each one as well as the absolute difference between legs as described in chapter 1 This results were manually captured and stored on a spreadsheet file Figure 4 2 shows the resulting graph of the acceleration readings coming from the pivot shift test The graph shows three noticeable maximum and minimum values 64 Chapter 4 Clinical Trials for each of the axis These values are related to each of the three consecutive tests performed in the patient as described p
48. ected in the serial port After the device is found then a pairing routine is executed This routine establishes the the mode of operation for the test which consist of whether the user wants to use both sensing modules to calculate differential metrics or work with only one accelerometer Once operation mode is defined a message is sent back to the embedded device to properly setup and initialize the sensors Concluded the setup the Arduino sends back a string specifying that is ready to start the test at this point is in control of the Processing application to start the data acquisition The user begins the test via the user interface and starts to poll data from the sensors While the test runs the acquired data is showed both in text format and by plotting a graph Finally when the user stops the test an csv file containing all the readings from the test is saved along with a png image of the graph 3 3 1 The Coding Before starting coding the application few libraries are imported e processing serial e controlP5 23 e java io processing serial enables support for the serial port class controlP5 is GUI and controller library that incorporates elements like sliders buttons toggles knobs 46 Chapter 3 The Pivot Shift Prototype Software Design textfields among others to ease user interaction This library is very convenient because Processing don t provide GUI elements out of the box although they can be manually
49. eft legs to produce a diagnosis the pivot shift test is able to asses the ACL 87 Chapter 5 Data Acquisition and Analysis knee injury without testing both patient legs for this reason further analysis shows the results from each legs separately Data is displayed for all patients as well as divided by healthy and injured groups variable mean sd median min max range se ALLX L 0 0582 0 6646 0 1310 3 2883 2 8085 6 0968 0 0057 ALLY L 0 1534 0 3501 0 0625 1 9294 2 0625 3 9919 0 0030 ALLZ L 0 0273 0 4797 0 0343 1 9738 1 9214 3 8952 0 0041 ALLX R 0 2300 0 5007 0 2198 2 0544 2 0504 4 1048 0 0042 ALLY R 0 2174 0 3970 0 0907 1 8609 1 8770 3 7379 0 0034 ALLZ R 0 3655 0 4606 0 3206 1 9657 2 0544 4 0202 0 0039 Table 5 6 Summary statistics for X Y and Z in left and right legs One thing worth noticing is that even after performing the fix on left legs X axis data compensating the rotation due contrary anterior posterior tibial rotation in section 5 3 still there is a significant difference in the mean values from left to right leg which backs up the idea of performing the analysis in left and right legs separately variable mean sd median min max range se HX L 0 0682 0 6545 0 1593 2 5988 2 8085 5 4073 0 0068 HY L 0 1430 0 3494 0 0665 1 9254
50. el acceleration x Serial print accely Serial println accel acceleration y Serial print accelz Serial println accel acceleration z reading the gyroscope Serial print gyrox Serial println gyro gyro x Serial print gyroy Serial println gyro gyro y Serial print gyroz Serial println gyro gyro z if breaktrans 0 break stop transmission This code provides the same functionality shown in the program for the MMA7361 with the difference of addition of angular velocity provided by the gyroscope Two main libraries are imported Wire which allows communication with I C TWI de vices and the Adafruit library 20 that incorporates the class Adafruit LSM9DSO that includes useful functions to easily configure the device to the different ranges as well to get the outputs from each of the axis and sensors available accelerometer gyroscope and magnetometer without coding to a register level The function setupAccel lets you choose the range of operation for the accelerom eter from the following options e LSM9DSO_ACCELRANGE_2G 34 Chapter 2 The Pivot Shift Prototype Hardware Design e L5M9DS0 ACCELRANGE 4G e L5M9DS0 ACCELRANGE 8G e LSM9DSO_ACCELRANGE_16G whereas setupGyro allows setting the angular velocity range to e L5M9DS0 GYROSCALE 245DPS e L5M9DS0 GYROSCALE 500DPS e L5M9DS0 GYROSCALE 2000DPS sensor event 1 is a structure that provides a single sensor events in a com
51. elz 1 02 gyrox 1 20 gyroy 0 14 gyroz 0 23 accelx 0 08 0 4 Acceleration 0 2 accely 0 02 accelz 1 02 gyrox 1 26 gyroy 0 52 gyroz 0 38 0 25 i i i i i 50 100 150 200 250 300 Autoscroll No line ending 19200 baud Test Duration a Arduino IDE serial port monitor b Output for x y and z axis Figure 2 14 Reading the output of the LSM9DSO accelerometer resting at horizontal position performing the same test but adding the angular velocity readings from the gyro the results were an average of 290 samples for each variable 2 3 The Prototype This section describes the idea behind the pivot shift tester prototype and how it was implemented Going back to what was introduced in the first chapter we saw the different medical tests and tools commonly used in the diagnose of the injury of the anterior posterior ligament We were able to identify them in two main groups e imaging techniques e routine tests The problem with the first group imaging resides in the fact that uses equipment considerably expensive in such a way that the patient often needs to visit a clinic or 36 Chapter 2 The Pivot Shift Prototype Hardware Design hospital to take test It is highly unlikely to find them in the specialist s office and both CT scans and MRI typically have a cost superior to 1000 US dollars Aside from the economic factor other disadvantage that the imaging techniques have is that b
52. es Newnes 2011 Adafruit LSM9DSO Library Adafruit Online Available https github com adafruit Adafruit LSM9DSO Library GNU Lesser General Public License The Free Software Foundation Online Available https www gnu org licenses lgpl html C Reas and B Fry Processing a programming handbook for visual designers and artists Mit Press 2007 controlP5 A GUI graphical user interface library for processing Andreas Schlegel Online Available http www sojamo de libraries controlP5 M Summerfield Rapid GUI Programming with Python and Qt the Definitive Guide to PyQt Programming 2007 matplotlib python plotting library John Hunter Online Available http matplotlib org NumPy library Scipy org Online Available http www numpy org psych Procedures for Psychological Psychometric and Personality Research William Revelle Online Available http cran r project org web packages psych index html L A Kelly D G McMillan A Anderson M Fippinger G Fillerup and J Rider Validity of actigraphs uniaxial and triaxial accelerometers for as sessment of physical activity in adults in laboratory conditions BMC medical physics vol 13 no 1 p 5 2013 L Engebretsen C A Wijdicks C J Anderson B Wester haus and R F LaPrade Evaluation of a simulated pivot shift test a biomechanical study Knee Surgery Sports Traumatol ogy Arthroscopy vol 20 no 4
53. gaments to keep knee stability and protection limiting the movement These ligaments can be described as e Collateral ligament These ligaments can be found on the sides of the knees The lateral collateral is located on the outside while the medial collateral on the inside Their function consist on controlling the knee sideways motion and prevent unusual movement e Cruciate ligaments Ligaments that are found inside the knee joint The pos terior cruciate ligament being located on the back while the anterior posterior Chapter 1 Introduction lies on the frontal part of the knee They are responsible of the back and forth dynamic of the knee Patella Figure 1 1 Normal knee anatomy front view Figure 1 1 shows the anatomy of a healthy knee where it can be seen how the ACL runs in the middle of the knee in diagonal direction because of this prevents the tibia from sliding or moving out in front of the femur 1 In the other hand figure 1 2 exposes a knee with the ACL injury which among other things can be caused by e Direct impact or collision on one side of the knee Sudden stop of movement Overextend the knee joint Quick stop of motion and change direction while turning landing from a jump or running Slowing down while running Landing from a jump incorrectly Chapter 1 Introduction i Acomplete tearoftheACL A o Figure 1 2 Frontal view of a knee with ACL injury 1 3 D
54. h t str delta_h t Healthy The idea is simple the function already contains the calculated mean values for healthy and ACL patients The input files contain the X Y and Z acceleration readings for that particular pivot shift test so the first thing is to calculate the 91 Chapter 5 Data Acquisition and Analysis magnitude array to get magnitude of the acceleration that is handled by the get dataMag function The function contains as fixed parameters the calculated mean values for the ACL and healthy class seen in the previous table Then two variables are calculated delta_ACL and delta_h which are the absolute difference between the mean of the testing file patient to classify and the ACL and healthy means respec tively If delta_h is less than delta ACL then the patient is classified as healthy that is if the patient s mean is closer to a healthy patient mean Leg Missclassified Error Classifier Left 55 40 41 MeanClassifier Right 52 38 23 MeanClassifier Left 94 69 11 MaxClassifier Right 87 63 7 MaxClassifier Left 91 66 91 MinMaxClassifier Right 86 63 23 MinMaxClassifier Left 90 66 17 MeanMaxClassifier Right 90 66 17 Mean MaxClassifier Table 5 10 Results of different metric based classification approaches Previous table show the results of trying to classify patient data based on e mean value e maximum value e minimum and maximum val
55. he output since the maximum is set to 2g For this reason is important to know the application in which the device will be used on to select the appropriate output range From the previous statement would be very tempting to say that the best solution is to select the broadest range however this is were the sensitivity comes into play The sensitivity is the ratio of change between the input and output signal in the sensor The metric is specified at a particular operating voltage typically in mV unit for analog devices and LSB unit or unit LSB for digital output devices For the accelerometer it s seen in mg LSB where LSB stands for least significant bit LSB relates to accuracy of the digital representation of the measured unit in this case acceleration For example from table 2 2 we see that when using range 2g there is a sensitivity equal to 0 061mg LS B this means that when the lowest order bit in the output changes the acceleration had a change of 0 061mg s or 0 0005978m s 28 Chapter 2 The Pivot Shift Prototype Hardware Design in other words 0 0005978m s is the tinniest change in acceleration the sensor can have between readings Looking at the specification can be seen how for a broader range the sensitivity value increments as well which makes the sensor less accurate Additional to the convenience of providing both acceleration and gyroscope met rics in the same chip the LSM9DSO output signals deliver digital data in
56. he program started To provide the difference in time between samples two variables are used one 59 Chapter 3 The Pivot Shift Prototype Software Design representing sample 7 current sample and sample i 1 previous sample Once the difference between the samples is known angular displacement can be described as timestamp timestampi 4 A Re 1000 x AngularV el 3 1 where both timestamps are provided in milliseconds by the function millis and AngularV el is the current angular rate dps provided by the gyroscope and Angle represents the angular displacement between sample and sample i 1 Next table shows an example of how angular displacement is calculated in three consecutive samples First column represents the timestamp difference between the two samples in seconds the second column is the current angular rate coming from the gyroscope in dps and the third column shows the resulting angular displacement degrees between consecutive samples tgx tgx 1 s Angular Vel dps Angular displacement 0 03 1 71 0 0513 0 03 2 23 0 0669 0 029 1 78 0 05162 Table 3 1 Conversion from angular velocity to angular displacement in X axis 3 3 5 Ending the test and storing data Once the Pivot Shift maneuver has been performed the test can be terminated to do so it s a matter of sending a string 0 to the Arduino executing the following line
57. hift Prototype Hardware Design The next code snippet shows the Arduino data adquisition from the LSM9DSO inertial module include lt Wire h gt include lt Adafruit_LSM9DSO h gt include lt Adafruit Sensor h gt i2c arduino pins Arduino analog input 5 I2C SCL Arduino analog input 4 I2C SDA wf Adafruit LSM9DSO lsm Adafruit LSM9DSO char breaktrans void setupSensor 1 Set the accelerometer range lsm setupAccel lsm LSM9DSO_ACCELRANGE_2G 2 Setup the gyroscope lsm setupGyro lsm LSM9DSO_GYROSCALE_245DPS void setup 1 while Serial Serial begin 19200 Serial println Send a 1 to start the test while Serial available lt 0 waiting for some string breaktrans 1 to start the test Try to initialise the device if lsm begin O Serial println Unable to initialize the LSM9DSO Check your wiring while 1 Serial println Found LSM9DSO 9DOF Serial println void loop 1 33 Chapter 2 The Pivot Shift Prototype Hardware Design breaktrans Serial read if breaktrans 1i read data fron sensor while 1 breaktrans Serial read continue to read to catch Stop condition getting the sensor event sensors event t accel mag gyro temp from adafruit sensor master library lsm getEvent amp accel amp mag amp gyro amp temp reading the accelerometer Serial print accelx Serial println acc
58. his section describes the routine tests and imaging techniques used in the diagnosis of the ACL injury all of the following are considered non invasive tests Computer Tomography Commonly known as CT or CT scans During the test the scanner sends X ray pulses to the patient s body These pulses act during less than a second taking pictures of a thin slice of the area or organ of interest In some cases the scanner is able to tilt its position allowing three dimensional CT This kind of scans are typically used to study an organ but can be used to examine blood vessels bones and the spinal cord In order to make the image analysis easier an iodine dye constrast material is used to provide more clear pictures on the organs or area of study This dye can be put in a vein IV or can be orally administrated Chapter 1 Introduction CT can be used to visualize the ACL however its visibility is not the best when haemarthrosis bleeding into joint spaces is present situation where MRI has more significant results Therefore CT scan is used in cases when the patient cannot be exposed to an MRI for example when the patient has a peacemaker brain aneurysm clip or cochlear implant ear implant Magnetic Resonance Imaging Magnetic Resonance Imaging MRI is a medical imaging technique used in radiology to help physicians diagnose and treat medical conditions One of the highlights of this technique is its non invasive nature in
59. howing the magnitude for acceleration and angular dis placement 62 Chapter 4 Clinical Trials This section explains the origin of the previously collected data set used for the statistical analysis in this thesis Once the hardware and software were implemented a completely functional pro totype based in the first accelerometer MMA7361LC was handed off to a medical staff leaded by the orthopedist Edmundo Berumen Nafarrete M D and Carlos Vega Najera M D from the the Christus Murguerza hospital located in Chihuahua Mexico with the goal of testing the prototype in clinical trials The inclusion criteria for this trials consisted on patients who attended orthopedic examination with the specialist and volunteer to be part of the study no gender age or specific medical conditions were restricted to the study The study group included a total of 156 patients with mean age 24 range 11 76 years 37 women 119 men All patients participated voluntarily in the study by reading and signing a consent form approved by Christus Murguerza hospital ethics committee Clinial trials took place from May 2013 to May 2014 The study group can be further classified in three groups 63 Chapter 4 Clinical Trials e healthy patients e patients with previous knee or leg condition unrelated to an ACL injury e patients with the ACL anterior cruciate ligament injury For this trails each patient was submitted to two clinical tests
60. iagnosis Once there is the suspicion of an ACL tear Doctors often use several techniques to determine if in fact an injury exist There are invasive techniques like the knee arthroscopy and multiple non invasive that can identified in two main groups one being imaging tests like Radiography CT scans and MRI and the ones called routine tests or medical assessments including lachman pivot shift and drawer tests 1 3 1 Knee Arthroscopy Knee arthroscopy is a minimally invasive technique that allows orthopedic surgeons to assess and treat several conditions affecting the knee joint The procedure consist of a small incision in the vicinity of the affected joint Then a tiny camera around size of a pencil is inserted along with fiber optics to provide light source The camera transmits the images in real time to a monitor in the operating room this way the specialist can gain multiple views of the joint area The surgeon can use this technique to assess repair or remove damaged tissue to do so other small surgical equipment is inserted via a secondary incision around Chapter 1 Introduction cannula arthroscopic instrument o Em arthroscope camera amp light source Figure 1 3 Arthroscopy equipment and setup on patient s knee the knee Aside from the invasive nature of this technique it is the most effective one to determine the existence and severity of an ACL injury 1 3 2 Non Invasive Techniques T
61. ical one 95 Chapter 5 Data Acquisition and Analysis In this context now we have two main XZ datasets one containing the samples for healthy patients and a second one for the injured class One major thing to consider is that the number of samples from healthy patients is clearly higher than the available on patients with the injury and in order to accurately make use of the knn k nearest neighbor algorithm training data must be of equal size for all classes this case healthy and ACL patients Pivot Shift Test Beta Sun Mar 15 22 06 05 2015 qxicapi csv X e e xzqxicapi csv z 0 5 0 0 e gt i 8 iy e s e Acceleration a du o 20 40 60 80 100 120 140 160 180 1 5 Test Duration x Figure 5 15 From acceleration time to X Z sample points Figure 5 16 shows all the data available for the right leg Red points are data for injured patient as black dots correspond to healthy ones The total of points per class are e ACL 517 e healthy 9054 To overcome this difference in the total number of samples per class the approach taken was to produce different subsets of the healthy dataset with equal number of samples to the ACL class an example is shown in figure 5 17 When classifying using the knn algorithm the output results in defining to what class each testing sample belongs to This class is a result of a vote
62. ients with an ACL rupture The Knee vol 15 no 2 pp 107 110 2008 Online Available http www sciencedirect com science article pii S0968016008000082 M H Lam D T Fong P S Yung E P Ho W Y Chan and K M Chan Knee stability assessment on anterior cruciate ligament injury Clinical and biomechanical approaches BMC Sports Science Medicine and Rehabilitation vol 1 no 1 p 20 2009 Online Available http www biomedcentral com 1758 2555 1 20 109 References 7 10 11 12 13 14 15 16 17 P H Araujo M Ahlden Y Hoshino B Muller G Moloney F H Fu and V Musahl Comparison of three non invasive quantitative measurement systems for the pivot shift test Knee Surgery Sports Traumatology Arthroscopy vol 20 no 4 pp 692 697 2012 Online Available http link springer com article 10 1007 800167 011 1862 9 http link springer com article 10 1007 s00167 011 1862 9 fulltext html F R Noyes E S Grood J F Cummings and R R Wroble An analysis of the pivot shift phenomenon The knee motions and subluxations induced by different examiners The American journal of sports medicine vol 19 no 2 pp 148 155 1991 Online Available http ajs sagepub com content 19 2 148 short V Musahl A D Pearle and R F Warren Measurement Validation of Navigation During the Pivot Shift Test Operative Techniques in Orthopaedics
63. implemented The java io library is used to manage filesystem capabilities and have access to local files and be able to save test results to disk LISTBOH MULTILIST ITEM O LEVELI ITEM 1 LEVEL2 LEVEL2 ITEMI LEVEL2 ITEM2 LEVEL2 ITEM4 LEVEL2 ITEMS LEVEL2 ITEM ITEM 2 ITEM 3 ITEM 4 ITEM 5 ITEM 6 ITEM 7 RADIOBUTTON 50 100 113 40 hello world BUTTON SLIDER SLIDER TEHTFIELD BUTTON CHECKBOH y 0 50 KHOE KNOB SLIDER SLIDER SLIDER Figure 3 3 A selection of controllers available with ControlP5 library As was mentioned previously the core structure of a Processing sketch is found in the functions setup and draw Global variables are allowed as well and they need to be declared outside the scope of setup and draw functions Many global variables and classes are defined to setup the visual environment among other things including initialize serial port class initialize ControlP5 class setup window size and color scheme define text labels and plotting area position and defaults 47 Chapter 3 The Pivot Shift Prototype Software Design e setting default operation mode 1 accelerometer 2 accelerometers e plot mode acceleration only accelerometer and gyroscope e declare acceleration and rotation arrays After global variables and classes are defined setup is called and draws the initial user interface positioning text labels plotting area and GUI elements like butt
64. ino Mini Pro 328 side a pt wed bx E Ng 2 0 MEMS Accelerometer and Gyroscope lll 2 2 34 Accelerometers sn vu ix xiii Contents 2 2 2 GYTOSCOP Cs sd woh dk we ee ae wa ERE en 2 2 3 MMAT361LC Three Axis Low g Accelerometer 2 2 4 LSM9DSO 3D Accelerometer Gyroscope module 295 PO Protocol s es au Se doeet at hh ee ee A 2 2 6 Getting data from the LSM9DSO 2 2 23r Jhe PIOPOLVBOS se eio mop Ges Se bh se ee Be eS 3 The Pivot Shift Prototype Software Design Jul C PYOSBSSITB HOS ci ate Al ent dique bl reve Oque neris 32 Phe BASTOS M ee ops nl ten od ks are ad ay hc a eg te ae A as 3 3 The Pivot Shift Software uuo hei ear eei emus Ea od heeding xu tos psd e aac uc ES 3 3 2 Beginning the test i r des des oae RA E SS pn S 3 3 3 During the test tto de Ed EA SUR Ud s 3 3 4 From angular velocity to angular displacement 3 3 5 Ending the test and storing data 4 Clinical Trials 5 Data Acquisition and Analysis 5 1 First glance at graphs and PyPlotter 5 2 Interpretation of Graphical Data ase Ge aces pue viii 41 41 43 45 46 49 5l 56 60 63 67 Contents 5 3 Adjusting Left leg X axisreadings 000 4 76 oo Eranstorm ng DOO S auem ay pe e E pe Se eo es e d 78 5 5 Tools and Packages used us Oe Oe cea eee RS ELE ws 80 56 Data St tistics o custos cuisse o uec ede Mox de Dub ME uoto go tte r
65. ion However they share the same programming language which facilitates porting applications from one board to another The board connects to a computer via USB and communicates using standard serial port and can run both standalone or connected to another device or computer Both the Arduino software and the boards itself are cross platform running on Windows Linux and Mac OSX This was a very important feature taken in consid eration to choose this embedded solution 12 The Arduino IDE provides the functionality of code editor serial monitor and lastly and more important it is used to translate the Arduino code which is based on Wiring an open source programming framework for microcontrollers into C code then handed over to the avr gcc compiler that makes the final translation producing the program that gets flashed in the microcontroller Besides the use of the Arduino programming language extended functionality and low level programming can be achieved creating custom C C libraries 13 A thing worth mentioning is the big role the community plays on the success and continue development the Arduino platform this due contributions in code and projects from enthusiasts and hobbyists alike 2 1 1 Arduino Mini Pro 328 This board takes a small and minimal design of the Arduino which made it more suitable for the project These are the main features of the board e Microcontroller Atmega328 14 at 8MHZ 13 Chapter 2
66. l begin 19200 Serial println Send a 1 to start transmission while Serial available lt 0 waiting for some string breaktrans 1 te start the test void loop breaktrans Serial read if breaktrans 1 read data from sensor while 1 breaktrans Serial read to be able to stop valx analogRead A0 valy analogRead A1 valz analogRead A2 Serial print x Serial println valx 23 Chapter 2 The Pivot Shift Prototype Hardware Design Serial print y Serial println valy Serial print z Serial println valz if breaktrans 0 break stop transmission This code sketch initializes the serial port at a baud rate of 19200bps then in order to start the transmission the Arduino waits until receiving a string 1 can come from another embedded system or computer connected to the device Transmission is ended after receiving a string equal to 0 This pairing routine to start and end transmission is important when the system is designed to run on battery The analogRead function takes care of the ADC sampling of the pins used as analog inputs in this case the constants A0 Al and A2 mapped to pins 23 24 and 25 of the Atmega328 After uploading the code to the Arduino board we can see how the voltages are received and coded to integer values between 0 and 1023 due the 10 bit analog to digital converter Looking at figure 2 14b it s clear how the outp
67. leg towards flexion whether the tests was from one leg or the other Conversely this represents a problem with the readings from the X axis since they will have the tendency to cancel each other due their opposite values For this reason this section shows the method used to transform all the results taken from the X axis in left leg tests in order to compensate the opposite rotation between legs Once this done it s possible to perform accurate side to side data comparisons involving X axis as well The function applied to perform this transformation is described as X t X i 2Xinitiat 5 1 where X represents the array or vector containing al X axis samples and X initial 76 Chapter 5 Data Acquisition and Analysis stands for the first element or sample value in the array This operation is done to all the samples in the X axis array with the exception of the first one this due that the initial position for the device accelerometer in both legs left and right still is the same this operation only tries to compensate the rotation it does not mirrors the entire function horizontally which can be easily performed by doing X i as it will be showed The results of applying this operation have the form Pivot Shift Test Beta Thu Feb 26 18 34 33 2015 P39PI csv P39lflip csv xLeft leg Original data 3007 x Left leg Transformed 200 100r Acceleration 200 300 400
68. locity to angular displacement Directions for acceleration and angular rate are presented in the following image Z DIRECTION OF DETECTABLE ACCELERATIONS DIRECTION OF DETECTABLE ANGULAR RATES Figure 3 7 LSM9DSO module directions for acceleration and angular rates the directions for rotation of each of the axes are presented which are commonly known in flight dynamics like e roll rotation in X axis e pitch rotation in Y axis e yaw rotation in Z axis For this the Pivot Shift application it was considered to be more helpful to con sider the angular displacement rather than the angular velocity Using angular dis 57 Chapter 3 The Pivot Shift Prototype Software Design placement allows to clearly identify the directions which the prototype was rotated during test time as it will be presented shortly Expressed in other terms as we recall that the Pivot Shift maneuver takes both a valgus and rotational force it is assumed that we can track the valgus force via acceleration while for rotation angular displacement is needed rather than velocity Now it will be compared the results of using angular velocity and angular dis placement dev ttyUSBO START 1ACC XYZ ACC AND POSITION 1 091 0 952 0 813 0 674 150 300 150 300 Figure 3 8 Sample test using angular velocity Both of the tests showed by the graphs describe a similar motion as it can be seen from the acceleration plot up
69. men M D and Carlos Vega M D collaborating on the patient trials Design amp Implementation of a Pivot Shift Prototype for Quantitative Analysis by Marco Antonio Espinoza Sanchez B S Chihuahua Institute of Technology 2011 M S Computer Engineering University of New Mexico 2015 Abstract This thesis presents the utilization of a portable medical device intended to help in the diagnosis of the Anterior Cruciate Ligament ACL knee injury The prototype consists of an embedded system integrated with various sensors including accelerome ters and gyroscopes to provide force orientation and acceleration measurement The prototype has been used to quantify the results of a medical test called pivot shift which tests the dynamic stability of the patient s knee With the initial prototype built limited clinical trials were conducted Two schemes metric based classification and k nearest neighbors have been applied to the data set to empirically learn and judge ACL diagnosis vi Contents List of Figures List of Tables 1 Introduction 1 1 How the project got started Yeas eX pa de PA 64484 ESS es 1 2 The ACC Injury S ne hue cde out o de d os dee ot A t ree L3 JDUISSH SIBC que ERR TP eg le ge ee ey Us 1 3 1 Knee Arthroscopy 6 9 4 4 x SEE ww Xx GRO D E E 1 3 2 Non Invasive Techniques 426 p aos rss pase oe eos 2 The Pivot Shift Prototype Hardware Design 2e Ardn eie ee ee AR RR x Rc AC x a ee ae Re ae 2 1 1 Ardu
70. mon format In this way each sensor creates its own object from which it can retrieve its respective data from Calling getEvent grabs sensors new data so in order to continuously get new readings it needs to be executed recursively Similarly to the test done with the analog accelerometer these are the results of flashing the Arduino board with the previous code and getting some data from the Sensor As expected the graph of the three axis acceleration at rest horizontal postition is pretty similar of what was presented on 2 14b with the difference that in this case the outputs are in g s already With the goal of comparing the sampling rate between each of the chips previous Arduino programs were slightly modified to transmit data during a period of 10 seconds In the results the MMA7361C delivered an average of 165 triplets Xout Yout Zout as the LSM9DSO delivered an average of 650 triplets This difference is caused in some proportion to the overhead caused by reading the analog inputs and performing the analog to digital conversion when using analog sensors When 35 Chapter 2 The Pivot Shift Prototype Hardware Design eo dev ttyUSBO EAE x Pivot Shift Test Beta Mon Feb 9 13 13 43 2015 ae LMA test csv Send Send a 1 to start the test E Found LSM9DSO 9DOF 1 0F NCC x accelx 0 09 accely 0 02 0 8 accelz 1 02 gyrox 1 33 gyroy 0 31 gyroz 0 39 accelx 0 08 0 6 accely 0 02 acc
71. n 1 0 E 10 20 30 40 50 60 70 80 0 5 10 15 20 25 30 35 Test Duration Test Duration a patient left leg b patient right leg Figure 6 2 Disparity in sample size and beginning of test START 2ACC XYZ ACC AND POSITION 173 600 195 300 Figure 6 3 Recent test after software changes 3 14 15 105 Chapter 6 Conclusion and Future Work 6 2 Conclusion The project in this thesis introduces the design and implementation of an electronic device able to store quantifiable data out of the clinical pivot shift test This test is considered a very valuable asset in the diagnose of the ACL anterior cruciate ligament injury and in the evaluation of its reconstruction surgery however is often questioned over its subjective interpretation 32 This debate over the subjectiveness of the test highlights the importance of hav ing a device that produces quantifiable data out of this screening test The successful implementation of the proposed prototype tries to fill that gap providing numeri cal data in the form of acceleration and angular displacement that can be further analyzed Clinical trials were conducted in collaboration with the Christus Muguerza del Parque Hospital located in Chihuahua Mexico with the goal of validating the proto type and gathering patient samples Data produced in patient trials was reviewed and analyzed with the goal of determining typical values out of the healthy and in jured
72. n mind when looking for an accelerometer is the number of axes MEMS proof mass can measure one parameter in each available axis that being said a one axis accelerometer can sense the g force g 9 85 in a single direction Now is getting more common to use three axis devices which returns the acceleration on X Y and Z directions The output of a three axis accelerometer can be calculated as A eode 2 1 dm where a represents the acceleration m the mass of the body F the sum of all the forces actuating on the body gravity included F is the gravity force As mentioned before the accelerometer can be constructed attaching a proof mass to a spring An upward or downward acceleration in the sensor causes the proof mass to displace which can be measured to calculate the acceleration This acceleration can Spring 7 a b be accounted to the term F in equation 2 1 BORK RH Oe DAO LS Figure 2 3 Piezoelectric accelerometer spring moves as acceleration forces act upon the sensor producing a voltage proportional to that force 17 Chapter 2 The Pivot Shift Prototype Hardware Design The second term F is part of the model of the accelerometer because the force of the gravity not only accelerates the sensor body it also produces the displacement of the proof mass For example if the accelerometer stays still in horizontal position on a table is not accelerating and F 0 but still gr
73. ngle Therefore in order to draw rectangles for acceleration and rotation rect needs to be called with parameters 53 Chapter 3 The Pivot Shift Prototype Software Design rect plotX1 plotY1 1 plotX2 plotY2_1 upper rectangle acceleration rect plotX1 plotY1_2 plotX2 plotY2 2 bottom rotation dev ttyUSBO Save Image Open Previous Graph Pivot Shift Test START 1ACC XYZ ACC AND POSITION Figure 3 5 Pivot Shift UI dual layout presentation Now that the area to display the graphs is properly defined it is time to map the values obtained from the sensors acceleration and rotation into the area enclosed by the rectangles The final graph for each of the metrics has as the Y axis the values for acceleration and rotation as the X axis represents the number of samples for the test To draw this line graph sample is connected with sample 1 by a line To achieve this the first thing is to transform the value of the sample acceleration or rotation value to its corresponding pixel value inside the coordinate system created by the plotting rectangle This is done using the map function 54 Chapter 3 The Pivot Shift Prototype Software Design float map float value float lowi float highi float low2 float high2 return high2 high2 low2 value lowi hight lowi where value stands for the number to map or transform low1 and high1 are the minimum and m
74. ns with the movements during the pivot shift it can be established that the negative bumps coming from the X and Y axis are consequence of the knee rotation during the test First there is the negative bump followed by a period of stabilization to similar values prior the rotation and this has to do with the fact that close to the end of the test frames 5 and 6 the knee reverts the rotation and extends to its natural position 73 Chapter 5 Data Acquisition and Analysis Figure 5 5 The pivot shift movement step by step bottom describes positive direc tions for acceleration on each of the axis 74 Chapter 5 Data Acquisition and Analysis Pivot Shift Test Beta Thu Feb 26 12 43 43 2015 rotura lca pi csv 200 p 100 Acceleration l po o o 200 300 180 Test Duration Flexing knee lifting while applyin Internal tibial rotation g SU See PP Figure 5 6 Interpretation of the maneuver Regarding the Z axis the initial values correspond to the leg in full extension up to sample 12 approximately after this internal rotation happens frame 2 which slightly generates a negative acceleration on the Z axis towards the floor that can be seen in the graph by the minimum values seen between samples 13 to 20 approximately Finally once rotation is in place then the flexion happens lifting the knee generating the peaks around sample number 30 This Z axis dynamic
75. nsformation prior to any side to side data comparison or metric calculation involving the entire dataset three axis both legs 5 4 Transforming to G s The resulting graphs presented up to this point share something in common that is all of them have their acceleration values expressed in ADC analog to digital converter units that is why the magnitudes can have a value of hundreds of units Taking a step back down to chapter 2 of this work it was defined that when using a 10 bit ADC like the one used in this prototype based on the MMA7361LC ac celerometer the range of the output signal will have a value from 0 to 1023 Now the reason why the results shown so far contain negative values and there is none 78 Chapter 5 Data Acquisition and Analysis reaching this maximum 1023 is that since it is required to differentiate between posi tive and negative accelerations on each axis a posterior calculation was made which will be explained shortly The way the output of the accelerometer analog operates is that if no input acceleration is applied the output will reach a value equal to half the ADC range in this case 1024 2 512 In short with no acceleration the expected output in ADC units should be 512 for each axis Other important fact is that from the ADC range half of all possible values are used to represent negative acceleration while the second handles positive accelera tion Therefore by default values from 5
76. onnected to analog inputs in the Arduino de noted by AO Al and A2 pins Capacitors are connected to the outputs to minimize clock noise and on VDD to decouple the power source ARDUINO Duemilanove Diecimila UNO R1 R2 HLELLLEL LELLLLEI Figure 2 9 MMA7361LC and Arduino connection diagram The Arduino reads these analog voltages from the accelerometer and then digital izes them using the internal analog to digital converter In this model the analog to 22 Chapter 2 The Pivot Shift Prototype Hardware Design digital converter has a 10 bit resolution which means that it will map input voltages from 0 to 3 3 volts into integer values between 0 to 1023 219 1 This gives a reso lution of 3 3V 1024 or 3 2mV between readings The maximum number of samples that can be read in a second is 10 000 it takes about 100 microseconds to read an analog input The device can measure both positive and negative acceleration depending on the direction of the force being applied With no acceleration the output should be at midsupply For positive acceleration the output voltage will increase above Vpp 2 as for negative will go decrease below Vpp 2 The following is a simpe Arduino code to read the three outputs from the ac celerometer and print the values via the serial port 18 MMA7631 ANALOG 3axis accelerometer test char breaktrans float valx valy valz void setup 1 while Serial Seria
77. ons and toggles as well as looking for a connected device in the serial port At the end of setup execution we end up with this pivotinteractive dev ttyUSBO Shift Test START 1ACC XYZ ACCELERATION Figure 3 4 User interface for the pivot shift tester 48 Chapter 3 The Pivot Shift Prototype Software Design Figure 3 4 shows the initial state of the UI user interface at launch At that point the UI can be defined in 5 blocks showed as yellow rectangles the first block indicates the area where the graph will be drawn block number two is the area where text labels are placed to indicate the current metric e g for acceleration can be x 1 25 y 2 00 2 98 to specify current g values in each axis Block three is the area where y axis acceleration or rotation and x axis number of samples labels are set for the graph Fourth block contains four toggle buttons Start Stop Used to begin and end the test Operation mode Sets up the test to use one or two accelerometers e Enable Magnitude Switches the graph to display each axis as its own curve or enables a single curve showing the magnitude of the three components Graph Mode Toggles view mode to display acceleration only or acceleration and rotation in a single view finally block five contains a dropdown menu to select serial port in case more than one device is using the port a Save image button that saves both an image of the test and an csv fil
78. oth tests the patient s leg and knee in a static environment reason why in some cases is difficult to assess a definitive diagnose if the information provided by the image is not completely evident Here we provide an example the first image MRI shows a complete tear of the ACL ligament In an image like this the specialist is able to confirm the ACL injury without much hesitation because the ligament simply is not there anymore Figure 2 15 MRI complete tear of ACL ligament Compared to the first image this second one looks not that straightforward The MRI shows an incomplete tear of the ACL ligament that still represents a problem however is difficult to grade the severity of the damage by the image itself Is in this situation where routine tests like the lachman and pivot shift test help the specialist providing information about the stability of the knee in motion From what was shown in chapter 1 it can be established that the results of the lachman test can be related to the results of the test using the KT 1000 arthrometer 5 Therefore can be stated that the KT 1000 helps to provide quantitative anterior posterior 37 Chapter 2 The Pivot Shift Prototype Hardware Design Figure 2 16 MRI partial tear of ACL ligament translation in millimeters information to the lachman test Unlike the lachman test at this point there is not a definitive instrument or tool used for the measurement of the pivot shift The wo
79. per one Initial state of the device is at rest in horizontal position then it is rotated in an angle between 20 and 30 in the X axis take figure 3 7 as reference Analyzing the first image showing angular velocity can be seen how the angular speed picks up at the very moment that the object accelerates on the Y axis after 58 Chapter 3 The Pivot Shift Prototype Software Design pivotinteractive dev ttyUSBO Time 5332 0ms START 1ACC XYZ ACC AND POSITION 0 979 0 858 0 737 159 300 147 600 159 300 Figure 3 9 Sample test using angular displacement that point starts to lose velocity de accelerating to return to its initial state While this still is useful it cannot provide information of the magnitude of the rotation which is clearly seen in the subsequent image Looking at the graph using angular displacement the movement done during the test can be reconstructed in some way it can be said that while the device was positively rotated rolled about 23 it reached an acceleration on the Y axis about 0 374 g s The method used to transform the angular velocity to angular displacement in volved adding the timestamp of each of the angular velocity samples got from the embedded device in this way is known the difference in time between the current sam ple and the one received previously To have that information the function millis was used which returns the number of milliseconds since t
80. pp 698 702 2012 Online Available http link springer com article 10 1007 s00167 011 1744 1 http link springer com article 10 1007 s00167 011 1744 1 fulltext html R O Duda D G Stork and P E Hart Pattern classification New York Chichester Wiley 2000 111 References 31 R Faludi Building wireless sensor networks with ZigBee XBee arduino and processing O Reilly Media Inc 2010 32 N Lopomo S Zaffagnini S Bignozzi A Visani and M Marcacci Pivot shift test Analysis and quantification of knee laxity parameters using a navigation system Journal of Orthopaedic Research no 2 pp 164 169 2010 112
81. pplication using Processing Now that we showed the premise of programming in processing can go further and talk about the design of the pivot shift user interface software 44 Chapter 3 The Pivot Shift Prototype Software Design 3 3 The Pivot Shift Software The following diagram describes in high level the functionality of Processing s pivot shift application start Q Plot acceleration p D And Or gyroscope data n Start communication with Arduino Send end transmission message to Arduino tM m End the test Generate csv File with test data and Save graph image Print current Angular velocity Add to angular velocity array Setup accelerometers And start the test Is test Running Perform continuous data acquisition E Parse string from serial port Print current acceleration Is it acceleration Add to acceleration Data array Figure 3 2 High level algorithm for pivot shift software 45 Chapter 3 The Pivot Shift Prototype Software Design All starts establishing the communication with the embedded device this is done via serial port When using the Arduino board there are two options from which the serial port can be accessed the first one via the digital pins 0 RX and 1 TX and the second one and the used in this project via USB First there is a routine to check for the presence of a device conn
82. raph in real time to small UI details to make it more intuitive and 102 Chapter 6 Conclusion and Future Work Figure 6 1 XBee module implementation easy to use While starting looking at the resulting data from patient trials some design flaws were found Two of the main ones were e The number of samples per tests was not the same e Start of the actual test varied form patient to patient These two observations became pretty evident at the begging of the data analysis figure 6 2 shows tests results from left and right leg for the same patient To begin with it can be seen how one test has a total of samples for each axis of about 85 while the other leg has about 35 There were two factors that caused this the first one was related to a software constrain in the application designed to capture the readings The initial configuration of the application required the examiner to manually start and end the test using a button from the user interface or a key shorcut which automatically makes this disparity in sample size logical it is highly improbable for the examiner to stop the test meticulously after an specific number of seconds every time The second factor has to do with the actual execution of the 103 Chapter 6 Conclusion and Future Work pivot shift maneuver From examiner to examiner the technique may be the same but the timings or speed at which they perform it varies of samples left leg
83. rd sends the readings using the serial port interface which makes the data accessible to other embedded system or computer 39 Chapter 2 The Pivot Shift Prototype Hardware Design The next chapter describes in detail how the data is presented to the user in real time while the test happens via an user interface created in the Processing programming language 40 Chapter 3 The Pivot Shift Prototype Software Design In this chapter we explore the design and implementation of the software created to interact with the pivot shift prototype previously introduced The application visually displays the readings from both the accelerometers and gyroscopes while the test takes place The application was coded in the java based programming language Processing 3 1 Processing Processing is an open source cross platform programming language and integrated development environment initiated back in 2001 by Casey Reas and Benjamin Fry from the MIT Media Lab Initially intended as a introduction to programming for visual artists and design ers now has evolved to a development tool for professionals being used by students 41 Chapter 3 The Pivot Shift Prototype Software Design researchers and hobbyists alike Processing takes software concepts and transforms it to principles of visual form motion and interaction Within this visual context Processing is set to work as a software sketchbook The language is text based
84. reated in this programming language and a basic example which will help to assimilate more easily the main application to be introduced later on 22 In Processing all sketches have two main functions e setup e draw The setup function is called once during the execution of the sketch and it is used to initialize variables and run setup routines The draw function is the main loop of the application and it runs automatically after setup function it should not be called explicitly The function executes the lines of code contained inside its block until the application stops or noLoop is called The functions used to control the behavior of the main loop are e noLoop e redraw e loop If the main loop is stopped by noLoop it can be reestablished using loop or in the other hand if redraw is called then the code inside draw will be re executed only one more time The following code snippet is an equivalent of a hello world for processing This sketch will paint a point in the current position of the mouse cursor 43 Chapter 3 The Pivot Shift Prototype Software Design Hello world 2 void setupO 3 size 400 400 6 stroke 200 22 22 strokeWeight 5 background 200 200 200 9 fill 240 20 20 10 textSize 25 11 text Hello World 50 50 T 14 void draw 1 m point mouseX mouseY These is the result sketch_150211a Hello World Figure 3 1 Simple a
85. reviously The specifications of the PC used with the prototype were the following CPU DETAILS HFHHFHHHHHHEHHHHEHEH Architecture 1686 CPU op mode s 32 bit 64 bit Byte Order Little Endian CPU s 2 On line CPU s list Thread s per core Core s per socket Socket s Vendor ID GenuineIntel CPU family 6 Model 15 Stepping 6 CPU MHz 2000 000 BogoMIPS 3990 22 Virtualization VT x Lid cache 32K Lii cache 32K L2 cache 4096K HEMEMORY HET EIER G IE E EE GE E EIER GE E E VE E ERE EE ER V total used free Mem 2014 1610 403 buffers cache 933 1081 Swap 2044 0 2044 1 0 1 2 1 Running Linux distribution 3 11 0 12 generic 19 Ubuntu SMP Wed Oct 9 16 12 00 UTC 2013 i686 i686 i686 GNU Linux Distributor ID LinuxMint Description Linux Mint 16 Petra Release 16 Codename petra Next chapter introduces the results from testing the prototype on over a hundred 65 Chapter 4 Clinical Trials patients performing a data analysis and final thoughts about the results Aceleracion 0 20 40 60 80 100 120 Duracion de la prueba Figure 4 2 Graph showing acceleration output per axis from pivot shift test 66 Chapter 5 Data Acquisition and Analysis This chapter takes the results of testing the pivot shift prototype with a study group of 156 patients and analyzes the data files and graphs generated during these trials The data generated from the trails consi
86. right leg mean 96 18 97 43 std 54 79 55 26 median 75 83 Table 6 1 Variability in sample size from test to test The second observation it s clearly shown in 6 2a the initial 35 samples ap proximately are acceleration readings prior of making the pivot shift test and are expected to exist in the results but for some files were considerably high Since this initial samples do not represent acceleration readings caused by the actual medical test they had to be manually removed from each file which added considerable time to the analysis To reduce the impact of these two findings in future additions to the software were made e added a countdown prior of starting capturing data e limited the test duration with a 10 second timer The implementation of the countdown helped to reduce the number of initial ization samples before the actual movement as automatically ending the test after 10 seconds helped to impose the examiner to a particular rhythm of execution and helped as well to ensure that all the tests have similar sample size After the addition of the timer the average of samples per test is 215 samples 4 samples fact that will help to facilitate further data analysis 104 Chapter 6 Conclusion and Future Work Acceleration Pivot Shift Test Beta Tue Mar 24 16 53 53 2015 Pivot Shift Test Beta Tue Mar 24 16 53 37 2015 art93 p004pd csv art93 p004pi csv 20 Acceleratio
87. rk shown in this thesis project tries to fill that gap by introducing a non invasive prototype aimed to be used during the pivot shift maneuver to collect data about the test Recalling image 1 7 it is known that what makes particular this test is the ap plication of valgus and rotational forces which will be tracked using accelerometer and gyroscopes modules The concept is to implement two sensing modules one placed in the patient s tibia and the other in the bottom of the femur With this it is possible to take one point as reference and get differential acceleration readings representing the anterior posterior translation force applied during the test Both acceleration modules are connected to the Arduino board which handles the data acquisition and performs the following calculation for each new sample 38 Chapter 2 The Pivot Shift Prototype Hardware Design Figure 2 17 The pivot shift tester one sensing module located in the tibia and the second one in the femur Aces A CO femur Accxtibia Acc Acey femur ACCy tibia 2 4 Acc Acez femur ACCZtibia where Accx femur is the acceleration output registered by the rightmost module on image 2 17 whereas ACCXtibia the one on the left side Therefore Acc is the differential acceleration between femur and tibia during the test The prototype is powered up to a polymer lithium ion LiPO battery of 3 7V While the test takes place the Arduino boa
88. s comparison left side with original data right side compensating rotation 2 2 2 2 sl sh Histograms for complete dataset a X axis b Y axis c Z axis Mean median and sd for complete data set by axis Mean median and sd by healthy and ACL patients Taking the median of the maximum values Left legs mean median and sd for ACL and healthy patients Right legs mean median and sd for ACL and healthy patients From acceleration time to X Z sample points XZ datasets corresponding to right leg Creating subsets of the healthy class with equal size of ACL class Testing and training set examples Classification results for patient file PI2PD XBee module implementation a ln Disparity in sample size and beginning of test Recent test after software changes 3 14 15 xiii 83 85 85 86 93 94 96 97 98 List of Tables 2 1 22 2 3 2 4 3 1 5 1 5 2 5 3 5 4 5 5 5 6 5 7 5 8 Zero g voltage taken from data sheet and device testing 25 LSM9DSO Accelerometer characteristics ls 27 LSM9DSO Gyroscope characteristics 2 2 a 28 FG terminology us uia Is es ne eh Dec Rea ay ric hte ahah a 30 Conversion from angular velocity to angular displacement in X axis 60 Patients distribution edere o
89. s into extension 72 Chapter 5 Data Acquisition and Analysis 2 while extension application of tibial internal rotation 3 application of valgus force 4 flexion of the knee while applying both valgus and internal rotation 5 going back to knee joint extension Figure 5 5 shows each of the movements produced during the pivot shift maneuver just listed above The bottom of the image works as reference to show the initial position and orientation that the accelerometer has when the test is about to start This is critically important because keeps consistency on each axis among the tests The way each of the axis is presented bottom of the image makes them relate to particular movements during the maneuver in this case X axis handles the internal tibial rotation Y axis the back and forth movement of the leg and lastly the Z axis registers the knee flexion and extension Due the importance of this the prototype was marked to indicate to the physicians the correct way to align the accelerometers while conducting the tests Next to it the graph in figure 5 6 is the resulting left leg plot generated from the patient test shown in figure 5 5 Looking at it can be seen how there is a similar pattern in the X and Y axis in portions of the graph both axis have a negative bump followed by an increase in the acceleration on the Z axis This pattern is clearly seen for each time the test was repeated in that graph 4 times Relating this patter
90. slation registered its laxity is defined as grade I for 1 mm 5 mm grade II for 6 mm 10 mm and grade III for anterior translation over 10 mm This test can be complemented with the use of the KT 1000 in order to determine the anterior translation in millimeters 5 Chapter 1 Introduction a Lachman test b Drawer test Figure 1 6 Clinical assessments for the ACL diagnosis Drawer test The drawer test is a clinical test used in the initial evaluation when there is suspicion of an ACL tear injury Similarly to the Lachman test the patient stands with the hips flexed 45 and knees flexed to 90 with the feet flat on the table With the knee flexed verification of the relaxation is done by hamstring palpation this is done in order to prevent false negatives The thumbs are placed along the joint line on both sides of the patellar tendon After setup of the grip an anterior force is then applied to the proximal tibial with gentle to and fro back and forth movement to assess for increased translation between knees The test is considered to be positive if the tibia pulls in forward or backward direction more than normal Pivot Shift test This test is the combination of internal rotation and valgus force applied to the leg which helps to determine the dynamic instability of the knee The test is performed with the patient lying in supine position starting from the knee in full extension and then gently flexing to approximately
91. st of Figures 2 6 2 2 8 2 9 2 10 2 11 2 12 2 13 2 14 2 15 2 16 2 17 3 1 3 2 3 3 MEMS gyroscope submitted to angular motion 21 MEMS gyroscope submitted to linear acceleration 21 MMA7361LC analog accelerometer breakout board showing respec tive directions for each axis loui e ee Se Bh E SAS d 22 MMA7361LC and Arduino connection diagram 22 Reading the output of the MMA7361LC accelerometer resting at horizontal POSTON a a me esae ce ow Wut a ge hs et ee 25 The effect of gravity on accelerometer readings 26 PC signals showing the start condition address and data frames and stop condition Ve er ew tuc ee So eee te HEC wee d 31 Schematic diagram of two LSM9DS0 s connected to an Arduino Mini board nder P Se xcu tb Ere E Be ew Sieg 32 Reading the output of the LSM9DSO accelerometer resting at hori Bona positio sea ud 203 919 d a uo m udo ea RE LACK b eee 36 MRI complete tear of ACL ligament 37 MRI partial tear of ACL ligament 0 38 The pivot shift tester one sensing module located in the tibia and the second one in the femur ooa aa xo o o ons 39 Simple application using Processing 44 High level algorithm for pivot shift software Ls 45 A selection of controllers available with ControlP5 library 4T xi List of Figures 3 4
92. sted in the following e 280 csv files 140 for left leg 140 for right leg e 280 graphs 140 for left leg 140 for right leg e a spreadsheet file containing patient data like age weight KT 1000 test re sults and patient observations The total number of patients during the trials was 156 however inconsistencies where found in some of them like missing tests results csv file for one of the legs 67 Chapter 5 Data Acquisition and Analysis user typos while naming the output file making the test impossible to relate to an specific patient Due this complications some of the results had to be excluded in the following data analysis leaving a total of 140 patients 5 1 First glance at graphs and PyPlotter At the time the data from patients became available the first thing was to visually examine the group of graphs in order to get familiar with how real pivot shift tests acceleration graphs looked like As well a particularly interesting thing was to identify noticeable patterns or differences between the tests coming from left and right patient s legs This is important because as was stated previously given the direction of the movement accelerometers can register both negative and positive acceleration for any of the given axis A Python Qt 24 application was developed to ease this process of visual exami nation of the tests featuring Ability to reproduce graphs from csv files e
93. t identifying the metric and axis that belongs to and the last part is converted to a float specifying the magnitude if sdata indexOf x gt 0 x axis sdata2 split sdata arrayx add float sdata2 1 j In this snippet sdata is the string containing the newest line of data from the serial port The string then is separated into numerical and metric part Six global arrays are defined arrayx arrayy arrayz arraygr arraygy arraygz each one storing the data for its corresponding metric and axis This is how the application continuously stores the data coming from the embedded device the next step once each of the metrics has at least one sample is to proceed with the display of the graph An additional array arraytime is used This array keeps track of the minimum total of samples that all of the metrics have at a given point The main reason for this is to know up to what number of sample the graph can be displayed The following output displays the number of samples that each array haves at 4 seconds of elapsed time x 142 y 142 z 142 gx 142 gy 141 gz 142 t elapsed test time 4 291s For this example arraytime should have a value of 141 As was showed the plotting area is defined to use the majority of the UI space In Processing the size of the application window is specified explicitly in most cases using pixels In other approach it can be set to have a size based on the screen resolution whi
94. the axis outputs when no acceleration 0g acceleration is being applied This is typically provided in the device data sheet although is easy and recommended to calculate practically because can vary from chip to chip 380 Pivot Shift Test Beta Sat Feb 7 22 57 52 2015 MMA test csv T T Nx x i i l i i h 50 100 150 200 250 300 350 Test Duration b Output for x y and z axis Data Sheet From device testing Xout 1 65 V 1 591 V Yout 1 65 V 1 705 V Zout 2 45 V 2 205 V Table 2 1 Zero g voltage taken from data sheet and device testing 25 The Zero Chapter 2 The Pivot Shift Prototype Hardware Design Pivot Shift Test Beta Sat Feb 7 23 42 26 2015 Pivot Shift Test Beta Sat Feb 7 23 42 54 2015 A X CSV B 1 o 500 nee Arn e PN o eae SP PA gr Mite eder A Nx x Nx x B a o 450r o B t o A N o Acceleration Acceleration B o eo w a o w eo o T w a o 300 FRNA NIRA ARINC AGO A pio AAA KIA o POM oP cs i i i i i i i i 50 100 150 200 0 50 100 150 200 250 300 Test Duration Test Duration w S o 250 0 a Gravity acting on Y axis b Gravity acting on X axis Figure 2 11 The effect of gravity on accelerometer readings From the table above it can be seen a slight difference between the typical values provided from the data sheet and the ac
95. these two particular axis XZ the following try consisted in repeating the same metric classification methods but with the main difference of discarding Y axis In this context the new magnitude was taken as follows MagX Zac VX 22 5 4 93 Chapter 5 Data Acquisition and Analysis 0 6 0 5 0 4 acceleration g s 0 3 ACL Right legs vs Healthy Right legs 0 3 0 2 0 1 0 L R L R Y R 0 1 0 2 ACL R Z H R Z E mean sd E median Figure 5 14 Right legs mean median and sd for ACL and healthy patients Once done this was needed to recalculate the same metrics like in table 5 9 but this time based on XZ acceleration only leading to variable mean sd median avg min avg max se MagXZ H L 0 6916 0 4154 0 6176 0 2178 1 4774 0 0043 MagXZ H R 0 7038 0 4013 0 6283 0 2124 1 6750 0 0042 MagXZ ACL L 0 7051 0 2108 0 7043 0 2799 1 1495 0 0094 MagXZ ACL R 0 5721 0 3928 0 5199 0 1499 1 4425 0 0173 Table 5 11 MagnitudeXZ metrics summary for healthy and ACL patients And the results of implementing the same classifiers was the following As simple as the change was from using the information from three axis to going to only XZ data it s interesting to see how that effectively reduced the error rate almost in half for some of the proposed classifiers 94 Chapter 5 Data Acquisition and Analysis
96. translation in relation to the femur 4 between 20 and 30 degrees of knee flexion To perform the test the KT 1000 is firmly attached to the patient s leg by two bands after this the arthrometer is pulled to the tibia to provide an anterior force Audible feedback beeping to the examiner is noticed at 15 20 and 30 pounds of applied force The output of the test consist on the tibia translation with respect to the femur in millimeters mm With this data the laxity how loose the ligament is is calculated in side to side difference between the patient knees From extensive testing the literature states that an intact or partial ACL tear Chapter 1 Introduction Figure 1 5 KT 1000 test measuring anterior posterior knee translation have less than 3 mm of increased anterior translation during the test 4 Lachman test The Lachman test named after orthopedic surgeon John Lachman is a noninvasive medical test performed for the diagnose of the ACL injury known for its high accu racy In order to make the test the examiner ensures the tibia with one hand while securing the femur with the other one The knee is flexed at 30 while the patient lies supine The examiner stabilizes the femur and applies force in anterior direction on the tibia While a healthy ACL should prevent forward translational movement a positive result for an ACL injury will be a noticeable anterior translation of the tibia Depending on the anterior tran
97. ts of the healthy class with equal size of ACL class numbers as k parameter in order to avoid ties For this particular example can be seen how even setting k to 5 produces the same result classifying that sample as part of the injured class These are the classification results for that particular patient when k 3 That specific test file contains a total of 170 XZ samples out of which 17 were classified as healthy and 153 as ACL The result accurately classified the patient as injured patient P12 was indeed an injured patient having an error rate of 10 percent For this example k was set to 3 for illustrative purposes in theory knn algorithm is said to be optimal as k goes to infinity 30 but to do so infinite data would need to be available so in practice our selection on the k value is totally dependent on the amount of data available In either case for the following results our k value is bounded to half of the training set For each of the patients tested knn algorithm was run recursively starting from setting k equal to 1 up to half the 98 Chapter 5 Data Acquisition and Analysis T xzdataset aclR3 csv e e xzdataset aclR3 csv xzdataset HR 347 csv e e xzdataset HR 347 csv 2 xzP12PD csv E e e xzP12PD csv a Training sets b Training sets and testing set blue T T T r 0 32 e e xzdataset aclR3 csv e e xzdataset HR 347 csv e e xzP12PD csv
98. tual values got from testing the device Now finally to go from Volts to G s 1g 9 8m s the following calculation is performed for each of the axis Gz ADC xvoitage Zerox sensitivity Gy ADCyvoitage Zeroy sensitivity 2 3 G ADCZvoltage Zeroz sensitivity where Zerox Zeroy and Zeroz are 1 59V 1 705V and 2 205V respectively from table 2 1 The sensitivity value comes directly from the device specifications which can be 800mv g or 206 mv g for 1 5g and 6g operation ranges which in this case is set to operate at 1 5g 26 Chapter 2 The Pivot Shift Prototype Hardware Design 2 2 4 LSMO9DSO0 3D Accelerometer Gyroscope module The LSM9DSO is a system in package featuring a three axis accelerometer gyroscope and magnetometer also known as a nine degrees of freedom device 9DOF Each of the three sensors in the device supports programmable operating ranges The accelerometer has a full scale of 29 4g 8g 16 g the gyroscope an angular rate of 245 500 2000 dps degrees per second and the magnetometer a magnetic field of 2 4 8 12 gauss The device includes an PC serial bus supporting standard and fast mode 100 kHz and 400 kHz and an SPI serial interface Accelerometer and Gryoscope characteristics are found in the following tables Parameter Test conditions Specification Unit 2 4 6 g 8 16 Linear acceleration 2 0 061 Linear acceleration 4 0 122 Linear
99. u Beh S uos iura n caer oe 82 Summary statistics for X Y and Z data for all patients 84 Summary for X Y and Z data for patients by groups 84 Maximum and minimum acceleration metrics 87 Example table showing maximum and minimum values for 5 patients 87 Summary statistics for X Y and Z in left and right legs 88 XYZ summary for patients by groups left and right legs 88 Maximum and minimum values for patients by groups left and right xlv List of Tables 5 9 5 10 5 11 5 12 5 13 5 14 5 15 6 1 Magnitude metrics summary for healthy and ACL patients 90 Results of different metric based classification approaches 92 MagnitudeXZ metrics summary for healthy and ACL patients 94 XZ Magnitude metric classification results a 95 KNN results for Left lee ss ion ae ache E RT ee lp RA ee ode 100 KNN results for Right let us ug o X ER a SR VOA 101 KNN missclassifications rates by groups 0 101 Variability in sample size from test to test 2 2 2000 104 XV Chapter 1 Introduction This Thesis is based on the implementation of a previously design pivot shift pro totype and its tests results in patients trails The project consisted of an interdisci plinary collaboration between engineering and medical professionals The patient trials took place in Chihuahua Mexico with collaboration of Ed mundo Berumen M D
100. ues e mean and maximum values First column of the table specifies which leg the results are coming from the second one Missclassified indicates the average number of patients that were incor rectly classified and the third one shows the percentage that those missclassifications 92 Chapter 5 Data Acquisition and Analysis represent out of the total number of patients As can be seen clearly the error rate is too high so a different approach was taken after trying this method Going back to what was shown in figure 5 5 while describing the pivot shift movement was mentioned the importance that X axis has while taking care of the acceleration during the internal tibial rotation as the Z axis registers the anterior posterior translation acceleration Back and forth movement of the leg is assessed by the Y axis and could be considered as the least important feature of the three just mentioned during the pivot shift test 29 ACL Left legs vs Healthy Left legs 0 8 0 6 0 4 0 2 E mean sd 0 E median L H L L L Y L Y L H L Z 0 2 acceleration g s 0 4 0 6 0 8 Figure 5 13 Left legs mean median and sd for ACL and healthy patients Now following this hypothesis and looking at graphs 5 13 and 5 14 we can see how it seems to be a noticeable difference in the ACL and healthy groups specifically in the X axis most strong one and Z axis Seeing that and continuing with our emphazis of the importance of
101. urther and state the following there should be a noticeable difference between patient s right and left leg pivot shift tests the reason for this is that the rotation applied while performing the test on each side is opposite to its counter part Next graphs make clear this left to right side difference in rotation while the test takes place Figure 5 4 shows side to side acceleration graphs for the X axis on four different patients Red curves correspond to left leg while magenta to right Each of the graphs shows how right leg values tend to be considerably more positive than left ones This effect on the X axis is produced mainly by two factors e difference in the direction of the rotation while performing the test on each leg e the effect to the gravitational force going from the Z axis to the X axis while rotation happens 71 Chapter 5 Data Acquisition and Analysis xLeft leg FRE x Right leg RR SR SEEE DEPO hl ee IONE leg Figure 5 4 Side to side comparison for X axis in four different patients Next section tries to describe the pivot shift movement and translate it to sections on the already presented acceleration graphs 5 2 Interpretation of Graphical Data This section breaks down portions of the graphs and relates them to parts of the pivot shift movement to help make sense of the shape found on the data The pivot shift maneuver can be divided en the following frames 1 knee goe
102. useful units like G s and dps for acceleration and angular velocity respectively That being said when interfacing with digital sensors like the LSM9DSO there is no longer need of using the analog to digital converter of the Arduino board removing some over head of reading and sampling the analog inputs as well as making the conversion of ADC units to Voltage and then G s as seen previously using the MMA7361LC accelerometer Other fact worth mention is that the digital sensor eases the scalability of the system e g when using the MMA7361 three input lines are required to communicate the device with the Arduino Now if we were to implement a second accelerome ter or gyroscope in the system three additional lines would be required Xout Yout Zout Which equals the maximum number of analog inputs for the Atmega328 mi crocontroller that supports up to 6 without multiplexing 19 In the other hand to communicate two LSM9DSO0 s with the Arduino 2 accelerometers and 2 gyroscopes via I C requires only two lines as the following section will show 2 2 5 PC Protocol From the two protocols offered by the LSM9DSO PC was the one used to communi cate the accelerometer to the Arduino board The I C protocol requires two wires to connect a single master to a slave unlike SPI that requires four lines to achieve the same I C supports multi master system that allows more than one master to communicate with all slaves in the bus PC 29
103. ut registered by the z axis blue presents an offset from the other two axis This difference represents the gravitational force being applied to that axis Conversely if the device is rotated 90 degrees horizontally and vertically then the gravitational force will be spotted acting on the x and y axis Once the system is able to get and perform the quantization of the accelerometer outputs two things are helpful to give some context to the data obtained The first thing to is to convert the raw value returned by the ADC analog to digital converter to Volts To do so we do ADC voltage RawApc Vref 1023 2 2 24 Chapter 2 The Pivot Shift Prototype Hardware Design SO dev ttyUSBO KX SUI UU y 332 00 Z 435 00 x 311 00 y 333 00 z 435 00 x 311 00 y 335 00 z 432 00 x 309 00 y 331 00 z 431 00 x 308 00 y 332 00 z 433 00 x 309 00 y 333 00 z 434 00 va x Send 420 400 Acceleration 340 320 300 0 Autoscroll No line ending 19200 baud a Arduino IDE serial port monitor Figure 2 10 Reading the output of the MMA7361LC accelerometer resting at hori zontal position Where Raw apc is the integer from 0 to 1023 got from the ADC Vref is the reference voltage in this case 3 3V and the 1023 representing the levels of quantification or discrete values the ADC can provide After this the Zero g voltage is subtracted from the ADCYottage g voltage is the voltage that each of
104. vol 18 no 3 pp 181 184 2008 Online Available http www sciencedirect com science article pii S1048666609000044 C Reas and B Fry Getting Started with Processing O Reilly Media Inc 2010 Official I2C Specification NXP Semiconductors Online Available http www nxp com documents user manual UM10204 pdf M Banzi Getting Started with arduino O Reilly Media Inc 2009 Wiring an open source programming framework for microcontrollers Hernando Barragan Online Available http wiring org co ATmega48A PA 88A PA 168A PA 328 P Summary Atmel Online Avail able www atmel com images doc8161 pdf J L Hennessy and D A Patterson Computer Architecture A Quantitative Approach 5 ed Morgan Kaufmann 2012 A J Wixted D V Thiel A G Hahn C J Gore D B Pyne and D A James Measurement of energy expenditure in elite athletes using MEMS based triaxial accelerometers Sensors Journal IEEE vol 7 no 4 pp 481 488 2007 Online Available http ieeexplore ieee org xpls abs _all jsp arnumber 4105926 T R Hsu MEMS amp microsystems design manufacture and nanoscale engi neering John Wiley amp Sons 2008 110 References 18 19 20 21 22 23 24 25 26 27 28 B Evans Beginning Arduino Programming Apress 2011 D Ibrahim Advanced PIC microcontroller projects in C from USB to RTOS with the PIC 18F seri

thesisPivot - University of New Mexico

Contents

Download Pdf Manuals

Related Search

Related Contents