Full report - ECE - Cornell University
Contents
1. SOFTWARE FOR THE ANALYSIS OF 3D BIOLOGICAL DATA SETS A Design Project Report Presented to the Engineering Division of the Graduate School of Cornell University in Partial Fulfillment of the Requirements for the Degree of Master of Engineering Electrical and Computer by Adam McCann Project Advisor Bruce Land Degree Date May 2011 EXECUTIVE SUMMARY This project is the creation of a software interface and data processing toolbox tailored to the specific needs of a neurobiology research group The project scope and requirements were set in collaboration between Cornell University s Electrical and Computer Engineering Department and Cornell s Department of Neurobiology and Behavior The software created is called receptorToolsGUI and is run within MATLAB utilizing its image processing user interface and 3D rendering functionalities Professor Ron Harris Warrick s lab is investigating the cellular consequences of spinal cord injury SCI As part of their research they have begun to analyze the SCI induced changes in sensitivity to serotonin 5 HT by identifying the presence of serotonin receptors The serotonin receptors can be visualized in transverse image sections of a spinal cord through the use of a confocal microscope The images are spaced closely together to create a three dimensional representation of a short segment of the spinal cord In characterizing the presence of the serotonin receptors the researchers wanted to ide2. 800 voxels are labeled first Once they re labeled the label data is passed to the isosurface function along with the label value This function looks for points in the data with the specified label value and tries to connect then with patches These patches are calculated as vertices and NU receptorToolsGUI User s Manual by Adam McCann edges and are then passed to the patch renderer function with instructions to make them green The same process happens for medium receptors 800 volume 3000 that are colored yellow and large receptors volume 3000 voxels which are colored red These rendered patch surfaces are displayed on a graph with dimensional aspect ratio adjusted to account for the physical distance between confocal images and the pixel to pixel distance For most cases in data I ve seen the following values determine the aspect ratio pixSize xy 91 55e 9 pixSize z 1 18e 6 ZE actor pixelze 27 DlXOorZze xv 2 5 The factor of two in the denominator of the zFactor is due to the fact that the 3D view is rendered at half resolution 512x512 instead of 1024x1024 Some lighting and background color is also added to the plot to help the 3D effect The max projected image is also shown on the floor of the plot to help the researcher relate the 3D rendering with the 2D images CETUUITOeUTETEUETO uw F e dt Wes immi Toshk Dektep Window Help 0644 hk 150984 4 08 an 3DSet S6 P6 MLVIIIb intact amp 48 z000 chO1
3. 3D Blob Viewer Selecting Run SphereFinder gt will open up the user interactive threshold panel 1f auto thresh was not selected 10 receptorToolsGUI User s Manual by Adam McCann The user can adjust the threshold with the up or down buttons or by just typing it in the box and pressing enter When the selected threshold changes the left image in the Preview image set will show the resulting thresholded Max Z projection of the set while the right image shows the unchanged original The user may also want to use the image histogram provided at the bottom of the panel with a blue line to indicate the current threshold Selecting the Accept Thresh button will cause the program to finish and output all the selected options 5 3 Remove Ref Mean This tool is used to subtract the mean of one image from another Selecting Tools gt Math Operators gt Remove Ref Mean in the receptorToolsGUI figure will open the following dialog box Reference Image 8 Bit Subtract na primary_MAx_P3_MLWIlla_sc705 tif subtractor Image 12Bit S6 P amp MLVIlla sciqu 549 2000_ch01 tif The mean of the reference image will be subtracted from the subtractor image The resulting image is always in 8 Bit and will labeled so in the name of the resulting figure 5 4 Histogram This tool is used to show the histogram of a selected image Selecting Tools Histogram in the receptorToolsGUI figure will open an image select list Selecting Generate
4. Hda 5x59 a d D EJ m CDU p E un ai JOSet S6 PE MLVIH intact649_2000_ch01 3D Blob Vol Histogram seipsa po D de A A T pom Bl 3056 56 Pe vib tected 5000 cho 30 Ree S SE File Foe o Vates niei Tesh Drop Window Heip L I2 ids ik 093084 2 05 20 305er S6_PG_MLVElb_intact649_ r000_ch01 3D Blob Scatter Plot Member e 3D Blaba 200 400 5600 800 i000 1200 Mislum e Bl 20 Sct 56 Pe MLVES abaci 2000 OL JO Boba c E EE EZELIT 2000 ch01 Blob Aug A Average Irtengty amp 5 F E T s s gl F e Edu View Weee Tools Desktop Windes Hop LI Dedi k XTMXs wg eo ISa Be Pe Mum b intact 000 ch 3D Biot Ag inanity Histog He Edi Vere Inget Tes Desp Window Help D da jr 55098 4 S D5 20 h E SEPE ML VIR intacta 2000 chOt lob Arey Intensity Curesdlateve Hi E Mambar of 30 Blobs E Ra En um E J Currilaiw Histagrara E ES Output to Workspace When selected the software will place a copy of the 3xnumblobs characteristic data matrix in the workspace under the variable name sphere info If there is a variable in the workspace already call sphere info this will not append the new data to the old data but will instead write over the sphere info variable 3D Render When selected a 3D rendering of the identified receptors is generated using MATLAB s isosurface function Before rendering the identified receptors are given a size category number Small receptors volume
5. Histogram gt will produce the histogram for the image If the image is mostly black the histogram will let the zeroth bin of the histogram run off the chart and display a warning box ile Edit Wew Insert Tools Desktop Window Help E Jide h 9 9X Xl BOA mm 8 Bit Subtract no primary MAX P3 MLVIIla sc705 tif Histogram 3000 2500 2000 1500 1000 500 0 50 100 150 200 250 dius receptorToolsGUI User s Manual by Adam McCann 5 5 Max Z Project This tool is will compute the Maximum intensity for the each z projected pixel in a image sequence For all notation in the software and in this manual the image axes are x and y while the axis perpendicular to the plane of the image is the z plane Selecting Tools gt Histogram in the receptorToolsGUI figure will open an image select list Selecting multiple files of the same bit depth will compute the max z project and create a figure of the same bit depth with the annotation Z PROJECT in the figure name 6 Sample Work Flow In this example the goal 1s to use a 12 bit raw image and its 12 bit corresponding reference image to get a blob info data set Open both images with receptorToolsGUI Use the Remove Ref Mean to subtract the raw reference mean from the raw image Save image not the reference image in 8 bit Open the 8 bit image in ImageJ and apply deconvolution then save in 8 bit from ImageJ ImageJ has a plugin called Iterative Deconvolution Run this 10 it
6. Tools 5 1 2D BlobFinder The blobFinder tool is used to identify and characterize the blobs in a given image The output is a list of blobs and their corresponding area total intensity and average intensity values I ll use the 8 bit example image Subtract no primary MAX P3 MLVIIIa sc705 tif for the following walk through receptorToolsGUI User s Manual by Adam McCann File Edit ew Insert Tools Desktop Window Help w Ode k ASG L Selecting Tools gt Filters gt 2D BlobFinder in the receptorToolsGUI figure will bring up the following interface Select Image 8 Bit Subtract no primary_MAX_P3_MLVilla_sc705 tif Settings Figures to Display AutoThresh Only Original Image e Use Thresh Interface Size Restricted Image 2 Min Blob Radius RGB Overlay Blob Numbering Output Options uM XLS Output m Show Plots XLS File Name Browse Output to Workspace Run BlobFinder The Select Image drop down box will list all the currently open images I ve selected the example image shown above In the Settings group the AutoThresh Only will just disable the user interactive thresholding of the image and just use the value of the image mean intensity one standard deviation of the intensity The Min Blob Radius setting controls the radius of the minimum allowable blob The possible figures to display are Original Image When selected this will keep the orig
7. analysis This was particularly useful for the researchers in performing T Tests on their accumulated data The other display options are for viewing an image with the blob numbers shown or an RGB overlay that shows identified receptors in red overlaying the blue background image B uiBlobFinderGUI Select Image 12Bit S6 P6 MLVillb intact649 z009 ch01 tif Settings Figures to Display AutoThresh Only Y Original Image O Use Thresh Interface Size Restricted Image 2 Min Blob Radius 4 RGB Overlay Blob Numbering Output Options XLS Output V Show Plots XLS File Name Browse Output to Workspace Run BlobFinder Figure 4 2D Receptor ID Interface Bl uithreshGul Preview Image Threshold Controls Preview Histogram Accept Thresh A amp Bit 3ublract na primary Mix P3 NMLVIa 709 0 A m C Vr ow TIRE Fie Edit View Insert Tools Desktop Window Help O S Halk SS TDA um LC E3 Bit_Subtract no primary_MAX_P3_MLVIlla_sc705 tit Blob Area Hi Number of Bless 400 600 zg amp Bit 3ublract ne peirmary Nix P3 MLVIa 709 0 on A a Fie Edit View Insert Tools Desktop Window Help Haas Aspat alln m B ds Sottract no primary MAX P3 MLVllla Mt eS E File Edit View Insert Tools Desktop Window Help E d didBehs09ws iu 0H 40 B Bit Subbract no primar
8. at the National Institutes of Health While this program is adequate in many aspects it is not designed for the 12 bit images from the lab s confocal microscope and does not provide the user the necessary computational flexibility What the researchers needed most was a piece of software where the all the image processing subtleties would be known and they could define the core functionality The software was built is several stages with frequent input from the researchers that would later use it At the start much of the focus was on developing the image processing and denoising tools necessary to identify the serotonin receptors most accurately Once the deconvolution solution to the confocal microscope image distortions was found focus was switched to creating a basic I O and interface structure so that multiple images could be opened and displayed without overwhelming the memory load on MATLAB This basic infrastructure turned out to be a real challenge because the raw images are of significant size 1024x1024 12 bit images After setting up the infrastructure and means of passing the raw data from user selected plots to user selected tools all the tools were added into a file menu system and connected to their respective source scripts The last tool created was the 3D rendering of input data sets Finally a user s manual was written to facilitate the use of the program At all design stages the project technical requirements were of equal impor
9. in 2D except that is adds to direction check for the new dimension After the b w size restricted image has been component labeled a loop just collects the area and intensity values for the now segmented blobs As an output option the user can display the following plots Histograms for receptor size and average intensity Cumulative histograms for size and average intensity Scatter plot of receptor size vs avg intensity Additionally the data may be exported to Microsoft Excel or to the MATLAB workspace for further analysis This was particularly useful for the researchers in performing T Tests on their accumulated data The other display options are for viewing an image with the blob numbers shown or an RGB overlay that shows identified receptors in red overlaying the blue background image mn ipherrAnderGLl Sara Sule Thea Daly Use Tree hienat Mir presa Poach ee L PE LN Li EE T 58 3 c E EZ E unu reached 8 i MLVEh _ reacia HOT ck MLVEh _ reacia TOO ck 1 Fe ML WE madri oe 6501 13 IE J Culpa Options ALS Qutpul 7 hw Pics Y 30 Fender Mia File Haee Qutpet ta Warkssace Figure 8 3D Receptor ID Interface Be Ms duro Wee hemi doc mido Weie f Ome a AS Sea 2 NT i Bl ate Tear cdeseld Conia muta Tira Up Thea Figure 9 Threshold Interface 10 B niece 56 06 MLVED intacr d _s000 chdl 30 Blob 09 8 i B 3050 56 P6 MLE intacr d8_2000_ch01 3D Blo
10. is a basic feature underlying the functional properties of neurons The position of the receptor in the neuron affects their role in the transmission and integration of information by the neuron The visualization of the localized receptors that are being studied in this context is biologically meaningful at a resolution of 0 1 lum The technique that achieves this resolution is the combination of immunofluorescence and confocal microscopy Adding specific antibodies to the extracted spinal cord causes the 5 HT c receptors fluoresce under certain wavelengths of light The confocal microscope uses a laser to excite the fluorescently labeled proteins 2 1 Confocal Microscopes rotating mirrors ME screen with laser pinhole F 4 Ni e detector PMT s fluorescent sample Figure 1 Schematic diagram of confocal microscope The confocal microscope used to collect the samples analyzed by this project was a laser scanning confocal microscope In this method a laser is reflected off a dichroic mirror passes many colors of light but reflects a small range of colors then reflected off a set of moving mirrors before being directed as the fluorescent sample The sample is excited and gives off light of a different frequency than the laser emits This sample generated light is redirected through the same moving mirrors as the laser and passes through the dichroic mirror Before the fluoresced light reached the detector 1t p
11. not use the File Save function in the image figures to save the images This will save a screen shot of whatever is in the figure window so you will be saving an 8 bit half sized version of your original image TROUBLESHOOTING TIPS If you load in 1mages and they look darker than they are supposed to be it s probably because you got the wrong bit depth Typically images coming directly from the microscope are in 12 bit and deconvolved images are in 16 bit It is helpful to label saved images with their bit depth 4 2 Saving Images In order to save an image that is open in one of the figures select File gt Save Image on the receptorToolGUI figure Presumably you would want to do this after you ve performed a Max Z Project or a Reference Mean Subtraction The interface will bring up a list of currently open images for you to choose from along with a bit depth to be saved in and the file name Select Image 12Bit S6 P6 MLVIlla sciqu 649 z0D0O0 ch01 tif Select Bit Depth 12 File Name E Matlab Scripts 4 5 Mac 2 Bit_56_P6_MLVIlla_sciqu_649_ Browse Save WARNING Saving in 12 Bit format will really be in 16 Bit format but will leave the last four bit blank 4 3 Close All Images In the receptorToolsGUI figure selecting File gt Close All Images will close only the images with the Tag starting with the image or equivalently all the figures that aren t plots or interfaces 5
12. of data set Colored coded volume visualization based on receptor size Conversion of 3D data set to usable 2D data set via max z projection Image Histogram Basic image arithmetic operations Detailed User s Manual with instructions for expanding software capability A final draft of the software was finished in early April and was used in the receptor analysis of a paper submitted by an undergraduate researcher Gabrielle Van Patten as part of her research honors thesis The software should see good use as the lab continues receptor research and was made to be easily expandable should they want to add more functionality Software for the Analysis of 3D Biological Data Sets Adam McCann ABSTRACT This project is the creation of a software interface and data processing toolbox tailored to the specific needs of a neurobiology research group The project scope and requirements were set in collaboration between Cornell University s Electrical and Computer Engineering Department and Cornell s Department of Neurobiology and Behavior The data set and characterization process are essential to the research conclusions of the Harris Warrick group in Cornell s Neurobiology Department The data set is a sequence of greyscale 2D images taken from a confocal microscope at different depths The process of characterizing the data set involves segmentation according to intensity and geometric shape within a unique noise environment imposed by the confocal
13. primary MAX PA MV se 705 08 lo m J IPUR TE i Paste sg Bl a Mi Subtract mo primary MAX PN MUVIE Te bal Dg File Edk Wew Inet Teoh Destop Window Help He d View bant Took Desktop Window Help HWOaGae 24150304 4 08 20 A Jide 59v s SO ao al Bi Su tract no primary MAX P MLVIlla scT05 tif Mob Area Hestogr nc no primary MAX P3 MLVIIa sc705 M Elab Area Cumulative Hi pre 2 B in suben no primary MAX PI MLVIDa ct a uw Vue det Wes inset Took Desktop Window Help te ddalk X S Www mu B Eit Subtract no primary MAX P3 M VIa seT05 b Blob Scatter Plo 1 Idi 3 Bit Subien no primary MAX P3 MIVIa seTOSuse co d per z L A 2 F e EM View F t Tools Desktop Wed Help 0043 2453 3840 4 0 haci no primary MAX MAL Ville ecni tif Blob fug Intensity Cunami hie Eda Wew ieee Took Desktep Window Help Tie d ST e az 2 Oe ao t Subtract no primary MAX P3 MIL Villa scTOS bf Blob q intensity Hisi 2 PFTCTETTZSTTTISU EX washer Ciel peg belimasker img r 1 1 j m n label f not at the top Pe 1 13 5 3 gt 0 amp masker 1 1 m 3 abel masker imger 1 1 j m n label Output to Workspace When selected the software will place a copy of the 3xnumblobs characteristic data matrix in the workspace under the variable name blob info If there is a variable in the workspace already call blob info this will not append the new d
14. the images that are already open in receptorToolsGUI so I m going to implement a new GUI called randomizeGUI m and randomizeGUI fig Again type guide and this time select Blank GUI Add a pop up menu for image list and make tag for it imageList using the property editor for the pop up menu Next add a run button and give it the tag runRandomize Save the figure as randomizeGUI fig and it will generate the file randomizeGUI m In randomizeGUI m there should be a function called imageList_CreateFcn Paste the following code under it ihandles findobj regexp Tag image 1list cell 0 for i 1 size ihandles 1 ilist ilist cellstr get ihandles 1i Name end if isempty ilist ilist cellstr No Images Open end set hObject String ilist Next under the function runRandomize_Callback add the image passing code a call to the original function randomize and the output image passing to figure code Image Selection Data Passing imgList get handles imageList String listVal get handles imageList Value imgName imgList listVal if isequal imgName No Images Open errordlg No Images Selected make Hist Error return end imgHandle findobj Name imgName img get imgHandle UsSerData bitDepth str2double imgName 1 2 e receptorToolsGUI User s Manual by Adam McCann 00 we Call to Xxunctrion img 8 rando randomize img bitDepth Pass res
15. A E OE O E 11 5 4 o A II o 0 EA 11 5 5 hue ree ERE o E A 12 COMES cs RETE 12 Addo bit RENDER 12 receptorToolsGUI User s Manual by Adam McCann 1 About The software interface outlined in this document was created to the specific needs of Cornell s Harris Warrick lab in the department of neurobiology Last Updated May 2011 by Adam McCann ajm232 cornell edu 2 System Specifications All MATLAB scipts were created and tested in the MATLAB 2010b version with image processing toolbox included 3 Installation All software scripts and files are included in the zipped file receptorToolsGUI zip Including the following scripts blobFinder m makeHistGUI fig makeHistGUI m putvar m receptorToolsGUI fig receptorToolsGUI m removeRefMeanGUL fig removeRefMeanGUl m savelmageGUI fig savelmageGUI m sphereFinder m sphereFinderGUI fig sphereFinderGUI m uiBlobFinderGUL fig uiblobFinderGUI m uigetBitDepth fig uigetBitDepth m uilhreshGul fig uiThreshGUI m the following documents McCann MEng Report pdf McCann Meng Report User s Manual pdf as well as a folder containing a pair 2D sample images one 3D sample set and one PSF image To install and use the software place all the scripts above in the same directory change the current directory in MATLAB to this directory and run receptorToolsGUI m receptorToolsGUI User s Manual by Adam McCann 4 Input Output As mentioned above to start the program use the
16. arvey Xiaole Li R Luke W Harris Edward W Ballou Roberta Anelli Charles J Heckman Takashi Mashimo Romana Vavrek Leo Sanelli Monica A Gorassini David J Bennett and Karim Fouad Recovery of Motoneuron and Locomotor Function after Spinal Cord Injury Depends on Constitutive Activity in 5 HT_2C Receptors Nature America16 6 2010 694 701 Print 2 Hutcheon B L A Brown and M O Poulter Digital Analysis of Light Microscope Immunofluorescence High resolution Co localization of Synaptic Proteins in Cultured Neurons Journal of Neuroscience Methods 2000 1 9 Print 3 Prasad V D Semwogerere and Eric R Weeks Confocal Microscopy of Colloids Journal of Physics Condensed Matter 19 11 2007 113102 Print 4 Van Patten Gabrielle N and Ronald M Harris Warrick Immunohistochemical Quantification of 5HT2C Receptors and CaV 1 3 Channels after Spinal Cord Injury in the Upper Lumbar Mouse Spinal Cord Tech Print Suo APPENDIX deus receptorToolsGUI User s Manual by Adam McCann USER S MANUAL File Tools Help y Table of Contents MEME A e Po E EE RO 2 2 io io ose oai A OIM NIME QUUM eee ee d UM E NIU M SEMEN UE 2 O Room R 2 d DOO e 5 A E E 3 4 1 ODOT TA GES a E 3 4 2 SCRIP ERR A e no tase pcteee tose te tatecseetess 4 4 3 NOSE PaCS LLO 4 OS srs ccc E o o 4 5 1 2BD Mey VON I PERI E E E E E AEE E EE NE EEE E EE N E 4 32 IRB DDE A E E E E E E EE 8 5 3 A AAA OEA E E
17. asses through a pinhole placed in front of the detector photo multiplier tube so as to eliminate out of focus light The light that is passed through the pinhole is measured by the detector as one pixel in the resulting images The mirrors rotate to move the laser across the sample and generate the full image The resolution of the confocal microscope is limited by the diffraction of light When the fluoresced light is passed through the circular pinhole aperture it diffracts in a pattern that is referred to as the Point Spread Function PSF For confocal microscopes the point spread function is an Airy function in the lateral dimension but is a Sinc function in the axial dimension see Figure 2 In order to get the original undistorted image back out we must perform the deconvolution of the output image with the PSF seen in Figure 2 Deconvolution or the inverse transfer function of a convolution is extremely sensitive to the noise from the photo multiplier tube Due to the noise the deconvolution is typically performed statistically instead of inverse filtering MATLAB has a few statistical deconvolution techniques but I was not able to implement any in a way that was as successful as the method implemented in ImageJ For this reason deconvolution was not included as a tool in the resulting software Instead the instructions for using ImageJ s method are described in the User s Manual Figure 2 Confocal PSF 3 METHODS This section c
18. ata to the old data but will instead write over the blob info variable Selecting Run BlobFinder will open up the user interactive threshold panel if auto thresh was not selected Up Thrash Accept Thresh receptorToolsGUI User s Manual by Adam McCann The user can adjust the threshold with the up or down buttons or by just typing it in the box and pressing enter When the selected threshold changes the left image in the Preview image set will show the resulting thresholded image while the right image shows the unchanged original The user may also want to use the image histogram provided at the bottom of the panel with a blue line to indicate the current threshold Selecting the Accept Thresh button will cause the program to finish and output all the selected options 5 2 3D BlobFinder The 3D blobFinder tool is used to identify and characterize the blobs in a given image sequence The output is a list of 3D blobs and their corresponding volume total intensity and average intensity values Pll use the 12 bit example sequence in the folder 3D Test Set Selecting Tools gt Filters gt 3D BlobFinder in the receptorToolsGUI figure will prompt the user to select a group of files It is assumed that the files to be analyzed are all the same bit depth and are sorted in the file so that when they are loaded the stack is in the correct order This is simple to do if put the z parameter in the file name Upon selecting the images the
19. b File Edt View Inset Toole Desktop Window Help port ut EN rue h View deret Tool Desltop Window Help E 18d A ho Sg adsum XUH2 5 509w9zia adag 3OSet S6 P6 MLVIII intact 2090 ch 1 30 Blob Vol Histogram F E S Set S amp PS MV intact lS 2000 ct01 30 Blob Vel Cumulative Pista 1 de mn 305ek 5b P MN setae dL 30 thot so a 8 Fie Gee wer inset Tesi Desktop Window Hip x Dada 5509874 2 06 90 3DSe 56_PG_MUVIIb intaci 49_2000_ch01 3D Bish Scatter Plot i u r Li lll 3000 se Pe ium intecis i000 01 30 Bieb l Mil ESELI macies 2000 chop Blob avg EDO Average Irtengty amp amp 5 n amp 9 s w F Fie Edi View beee Tools Desktop Widow Help File Edi Vere Ingert Too Deep Window Help dius hog a08am ISa Edad Ut 30 Blob Asg Intensity Histo T 6 5 4 3 2 1 di Figure 10 Output receptor data plots 3 3 Three Dimensional Data Set Rendering The 3D rendering of the identified receptors is generated using MATLAB s isosurface function Before rendering the identified receptors are given a size category number Small receptors volume 800 voxels are labeled first Once they re labeled the label data is passed to the isosurface function along with the label value This function looks for points in the data with the specified label value and tries to connect then with patches These patches are calculated as vertices and edges and are then passed to the pat
20. ch renderer function with instructions to make them green The same process happens for medium receptors 800 volume 3000 that are colored yellow and large receptors volume 3000 voxels which are colored red These rendered patch surfaces are displayed on a graph with dimensional aspect ratio adjusted to account for the physical distance between confocal images and the pixel to pixel distance For most cases in data I ve seen the following values determine the aspect ratio pixSize xy 91 55e 9 pixSize z 1 18e 6 zFactor pixSize z pixSize xy 2 The factor of two in the denominator of the zFactor is due to the fact that the 3D view 1s rendered at half resolution 512x512 instead of 1024x1024 Some lighting and background color is also added to the plot to help the 3D effect The max projected image is also shown on the floor of the plot to help the researcher relate the 3D rendering with the 2D images edu a 3DSet S6 P6 MLVIIIb intact649 z000 ch01 3D Blob Viewer m File Edit View Insert Tools Desktop Window Help U hh uS oe9uss uw mg mem 3DSet S6 P6 MLVIIIb intact649 2000 ch01 3D Blob Viewer X pixels Figure 11 3D Rendering for a sample data set 4 TESTING AND SYSTEM QUALIFICATION In order to qualify the methods used above it was important to create a known data set for the 2D and 3D setups I wrote a script to populate a 3D volume with a specified number of spheres of a specified radi
21. erations Open the 8 bit deconvoled image with receptorToolsGUI Apply 2D BlobFinder and output data to workspace 7 Adding Future Functionality Adding new functionality to receptorToolsGUI can be very easy if you use the following steps and copy certain sections of existing code in order to get the I O right If for example I wanted to add a function that generated a random integer for every pixel in a specified image Pd call this function randomize and save it in a file called randomize m It is important the raw image and it s bit depth be input parameters into the function The script for randomize m might be something like this where I ve included the bitdepth conversion just to show function img 8 rando randomize raw_img bitDepth imgHi 2 bitDepth 1 imgLo 0 img 8 rando uint8 255 double img imgLo double imgHi imgLo a D S receptorToolsGUI User s Manual by Adam McCann for i l1 size img 8 1 for jJ l size img 8 1 IMG 33 Pando randi 299 end end To connect this function to receptorToolGUI start by typing the command gt gt guide Open receptorToolsGUI fig and Tools gt Menu Editor ll add randomize to Tools and give it the Label Randomize along with the Tag menu_tools_randomize Save receptorToolsGUI fig and open receptorToolsGUI m In the receptorToolsGUI m file there should be a new function listing called menu tools randomize Callback Now I want to use randomize on
22. following command gt gt run receptorToolsGUI For processing 2D images open the images first then select the desired process from the menu bar on the figure titled receptorToolsGUI For 3D data sets image sequences select the desired process from menu bar on the figure titled receptorToolsGUI and it will prompt you to open a sequence of images 4 1 Opening Images In the receptorToolsGUI figure under the file menu select Open Image or use the shortcut CTRL 0 when the receptorToolsGUI is active You will be prompted to select a file Look in Q Matlab Scripts 4 5 Mac L E Example Images 3 ot ron You may select tif images of bit depth 8 12 or 16 After selecting open you will be prompted to enter the bit depth of the image that s just been selected What is the bit depth of this image 12 Bit The software will then open a figure with the properties Tag image Name bitDepth filename User Data raw image data in raw bit depth The image that is displayed inside the figure is the selected raw image converted to 8 bit and resized to be half of its original size We convert to 8 bit for all displayed images because the human visual system can really only distinguish on the order of 256 different grey levels The resizing just help the 1 0 to run quickly and allows the user to keep many images open at once with minimal memory load receptorToolsGUI User s Manual by Adam McCann WARNING Do
23. he image at half size Displaying in this way makes the I O quick allows the user to display many images without slowing down MATLAB and doesn t heavily degrade the quality of the displayed version of the image In creating the figure for the image the software stores the raw image data in the User Data property of the figure and labels the Tag property on the image in the format image where starts at O for the first image and counts up for each subsequent image that is opened When 2D processes are opened they scan the open figures for the Tag that starts image Note that the Tag is different than Name which is displayed on the figure s menu bar A figure s Tag is only seen by background functions looking for it UNE SERO ce um m T 1 EVI RR S Ra a 5 m L Ga 55 Ph MUI miaii SO chii S uus n 12 amp 56 PE VI tet EOD ciet al Lun x m He Ed View keet Took Desten Window Hep A oden see emos nr read Grita Sed F e Edt View met Took Dedtop Window Hep Je d mini mmu BO ao Yue s 5neeeas 42 O8l eo oro Toa A Fie Teel Help Open image Cube Save image Chee All image Cir t Figure 3 Open Images Screen Shot 3 2 Receptor Identification Receptor identification in this software assumes that the image has already been deconvolved The user selects the image to look in then minimum sized blob that will be considered a receptor and finally the output options 3 2 1 Two Dime
24. ibody C Image B after subtraction chemistry original image control of no primary antibody D Binary output after E Mask for setting intensity threshold image C Ss Figure 7 Matlab analysis to determine the size and intensity of 5HT c receptor clusters in intact SCl saline and SCI quipazine mice First the average intensity of the no primary antibody control B is subtracted from the original image A to adjust for tissue auto fluorescence and non specific binding of the secondary antibody The resultant image C then has an intensity threshold applied in Matlab D and a minimum cluster radius is set for identifying regions of interest The program uses a morphological close operation to remove regions under the cluster radius restriction and then segments them to give a labeling mask structure for the deconvolved no primary subtracted image E Reproduced with Permission from e a The result of the paper is also reproduced with permission below We found that after SCI the number and average size of 5HT2C receptor clusters increases as well as the area and mean intensity of CaV 1 3 channels Hayashi et al 2010 found that SERT is down regulated after severe spinal cord contusions and Murray et al 2011 found that there are more constitutively active 5HT2C receptors after Likely many more receptor subtypes and channels are affected in different ways by SCI and only by using
25. inal image from closing Size Restricted Image When selected the software will show the black and white thresholded and blob size restricted image receptorToolsGUI User s Manual by Adam McCann RGB Overlay When selected the software will show the identified blobs in red overlaying the original image in blue and green 1 Baza Cerrar ot Dri Subtract no primary MAX P1 MU ec EL s Fd Wer bnhan Tech Dese i Blob Numbering When selected the software will show the identified blobs and if the user selects the Data Cursor in the figure window they can click on a blob the label number will be shown Ha EN em Fam Pooh Dedos Mindo Hei Dls b SSS e a DEB ad Fo te The possible output options are XLS Output When selected this will save a copy of the blob characteristic data to an XLS file of the name specified by the user in the text dialog box below the XLS Output button You cannot use this receptorToolsGUI User s Manual by Adam McCann function to append the results onto an existing XLS spreadsheet because the data values will always be written in the first three columns of the first sheet Show Plots When selected the software will display the following 5 plots regarding the blob characteristic data Histograms for receptor area and average intensity Cumulative histograms for area and average intensity Scatter plot of receptor area vs avg intensity Wl En sae e
26. mages in the order they will stacked As with the 2D case the user is also give the option to display the same 5 data plots with volume instead of area and the option to export to Excel or MATLAB workspace Additionally the user has a 3D render option Once the images and settings are confirmed the software will convert the data matrix to 8 bit and open a threshold interface This interface is exactly the same as with 2D except that the preview image a max projection of the stacked image set onto one image Upon accepting the threshold the software will convert the data matrix to black and white b w The b w image is then operated on to restrict the size of the 3D blobs to be those that are greater in radius than the user specified parameter The size restriction operation is a morphological open as in the 2D case but using a ball structuring element of radius equal to the minimum allowable radius The morphological operators extend to 3D in a very straightforward fashion The overall operation has the effect of slightly rounding the edges of all the blobs and removing those whose radius 1s smaller than the user specified value Following the size restriction the data matrix 1s still b w and is passed to a connected component analysis This will use a 6 connection criterion for labeling all the pixels in a connected blob with the blob number This is done recursively starting from the top left corner of the data matrix using the same algorithm as
27. microscope Prior to my project s completion the research group was only able to perform this analysis manually in 2D The application of image processing techniques to this biological context provides value to the researcher by improving the precision and accuracy of measurements as well reducing the time and frustration encountered through manual data characterization 1 INTRODUCTION Professor Ron Harris Warrick s lab is investigating the cellular consequences of spinal cord injury SCT As part of their research they have begun to analyze the SCI induced changes in sensitivity to serotonin 5 HT by identifying the presence of serotonin receptors The serotonin receptors can be visualized in transverse image sections of a spinal cord through the use of a confocal microscope The images are spaced closely together to create a three dimensional representation of a short segment of the spinal cord In characterizing the presence of the serotonin receptors the researchers wanted to identify potential receptors in the noise environment imposed by the confocal microscope Once potential receptors were identified their volume and average intensity also needed to be determined The lab also requested that the 3D data set be rendered in such a way that the visualization might help to characterize the data set Analysis of the spinal cord image sequences was previously done in 2D using ImageJ an open source Java based image processing program developed
28. multiple approaches such as immunohistochemistry and electrophysiology can we hope to elucidate all the effects of spinal cord injury and how we can mediate those effects to increase quality of life It is evident from these two sections that the software analysis tools were indeed useful in confirming the paper s conclusion that more in area and intensity receptors were seen in SCI mice I expect that the tools provided by this project will be useful in further confirming the lab s receptor sensitivity hypothesis in future publications 6 FUTURE CONSIDERATIONS In an effort to make the software created for this project most usable for the neurobiology in future work I made the infrastructure modular so implementing new functionality is simple and comes with instructions in the User s Manual If for example they wanted to add functionality for characterizing the web like Calcium channels they could simply write the script for it connect it to the menu and implement the function I O as per my instructions in the User s Manual 7 ACKNOWLEDGEMENTS I would like to thank Bruce Land Ron Harris Warrick and Gabrielle Van Patten for their help with this project They have helped to gain a better understanding of basic neurobiology image processing techniques and especially the process of creating user interfacing tools dd a 8 REFERENCES 1 Murray Katherine C Aya Nakae Marilee J Stephens Michelle Rank Jessica D Amico Philip J H
29. nsion Identification Once the user has selected in the image and desired settings the software will convert the input image to 8 bit and open a threshold interface This interface allows the user to set a hardline number if they wish or to adjust the threshold interactively using the updating preview image and histogram to help select a desired level Upon accepting the threshold the software will convert the image to black and white b w The b w image is then operated on to restrict the size of the blobs to be those that are greater in radius than the user specified parameter The size restriction operation is a morphological open using a circular structuring element of radius equal to the minimum allowable radius The morphological open is implemented by first applying a morphological erosion with the disc structuring element then applying a morphological dilation with that same structuring element A morphological erosion will iterate through the input image and check at each non zero pixel to see if the structuring element centered at that pixel would fit Fit means that when the structuring element is centered on the current image pixel and overlaid if there are any pixels that don t match in the overlay the current pixel is changed to zero A morphological dilation will iterate through the input image and at overlay the centered structure element on the current pixel The overall operation has the effect of slightly rounding the edges
30. ntify potential receptors in the noise environment imposed by the confocal microscope Once potential receptors were identified their volume and average intensity also needed to be determined The lab also requested that the 3D data set be rendered in such a way that the visualization might help to characterize the data set Analysis of the spinal cord image sequences was previously done in 2D using ImageJ an open source Java based image processing program developed at the National Institutes of Health While this program is adequate in many aspects it is not designed for the 12 bit images from the lab s confocal microscope and does not provide the user the necessary computational flexibility The researchers needed a piece of software where the all the image processing subtleties would be known and they could define the core functionality At all design stages the project technical requirements were of equal importance to the user interface usability for the neurobiology researchers The software resulting from this project is a user friendly packaging of the following tools that can operate on 8 12 or 16 bit images Image I O with image sequence handling Serotonin receptor identification in 2D or 3D Full data output to MATLAB workspace and or Microsoft Excel Annotated output images for corresponding receptor id data to image Auto generates receptor characterizing plots A Allows auto or interactive user thresholding 3D rendering
31. of all the blobs and removing those whose radius 1s smaller than the user specified value Following the size restriction the image is still b w and is passed to a connected component analysis This will use a 4 connection criterion for labeling all the pixels in a connected blob with the blob number This is done recursively starting from the top left corner of the image using the following algorithm setlabel position label make value pixel at current position equal to label look at pixel to the right of current pixel if right pixel is unlabeled and nonzero call setlabel right of position label look at pixel to the left of current pixel if left pixel is unlabeled and nonzero call setlabel left of position label look at pixel below the current pixel if pixel below is unlabeled and nonzero call setlabel below position label look at pixel above the current pixel if pixel above is unlabeled and nonzero call setlabel above position label After the b w size restricted image has been component labeled a loop just collects the area and intensity values for the now segmented blobs As an output option the user can display the following plots Histograms for receptor size and average intensity Cumulative histograms for size and average intensity Scatter plot of receptor size vs avg intensity Additionally the data may be exported to Microsoft Excel or to the MATLAB workspace for further
32. overs the software s infrastructure and core functionalities For operation details and instructions the User s Manual is available in the Appendix 3 1 Infrastructure The key issues that the infrastructure had to address are the I O setup the way in which users will pass image information to functions and the way in which images can be displayed to without slowing down MATLAB It seemed that use of ImageJ is quite common with cell level image work so I modeled my infrastructure and I O after ImageJ The software is setup so that the I O is different for opening 2D vs 3D data sets If the user is using a 2D data set they open the individual raw images then open the process and finally select the desired image from a drop list in the process interface In this way each 2D process will compile a list of open images upon creation of its interface If the user is using a 3D data set they open the process and that process will prompt them to open a set of input files This works because no function in the software is for the manipulation of 3D data sets 1 e all 3D tools operate on image sequences from source files and output data not 3D data sets Similarly the user can save or display any 2D image in a variety of bit depths 8 12 or 16 but cannot save or display image sequences When images are opened the user is prompted to give the bit depth for the image to be opened The software then creates a figure that displays an 8 bit version of t
33. tance to the user interface usability for the neurobiology researchers Approaching the software design with this mentality led to an overall system architecture that is modular and easily expandable 2 BACKGROUND This section covers the background work in different fields that are being used in this project in creating the data set for which the software is designed for A brief explanation of the neurobiological goals and methods for the project is provided as well as a discussion of confocal microscopes and the noise environment they induce 2 1 Neurobiological Goals and Methods Muscle paralysis after spinal cord injury is partly caused by a loss of brainstem derived serotonin 5 HT which normally maintains motoneuron excitability by regulating crucial persistent calcium currents It has been observed that one of the long term effects of the lost serotonin is that the motoneurons compensate by becoming more sensitive to serotonin to regain excitability In an effort to better characterize the increase in motoneuron sensitivity the Harris Warrick lab is monitoring the long term changes in overall number size and intensity of 5 HT c receptor sites in healthy mice and mice with a spinal transection An increase in receptor activity restores the calcium channels in motoneurons to enable the recovery of motor function in the absence of serotonin but without regulation from the brain The subcellular distribution of neurotransmitter receptors
34. ulting image to figure figure Name num2str bitDepth Bit Ref Remove imgName NumberTitle Off Tag imageRefRemove set gcf UserData img 8 rando imshow img 8 rando delete handles randomizeGUI Finally in receptorToolsGUI m under the function menu tools randomize Callback put randomizeGUI sl s
35. us and intensity gradient in order to qualify the techniques and 3D rendering with a known data set and dimensional aspect ratio Slices of this populated data set were used to test the 2D tools The approach to use spheres with linear intensity gradient by radius was chosen to best mimic the actual data sets in way that was easy to test accuracy To choose a test radius and gradient factor I just tried a few options until the test set visually looked like the real data set The most important thing to set in the artificial data set was the mean intensity This value was quite low in the actual data sets and so the values needed to match to properly test the display and identification capabilities The figure below shows a slice of one the test data sets ex 2 Ah dm L El Ac wel ial L in cl a Pes Ti F wu A m a Figuee 510 I ds b w hie Edi View Fami Tea Deskep Weie Hep coGds 85039024 0 05 ao Figure 12 Artificial Data Set Slice 5 RESULTS AND DISCUSSION The end product of this project is fully functional to the requirements asked by the neurobiology lab and has surpassed their expectations More importantly the software has already aided in the research conclusions of an undergraduate researcher Gabrielle Van Patten as part of her Research Honors program thesis The following extract from her paper was generated using the receptorToolGUI Reproduced with Permssion from un A 5HT cimmunohisto B No primary ant
36. y MAX P3 MLVIlIa sc705 t Blob Scatter Plo 190 TOO yx Average ntens tw ul ao uim no primary MAX P3 MLVIIla scT05 ti Blob Avg Intensity Hisl Humber ef Blobs 120 140 160 180 Average Intensity 300 400 500 600 Area P CCFL Em A amp amp PICADA abel masker imgrr 1 1 m n 1abe1 nor ar rhe top Pr i1 1 m 0 amp amp masker C1 1 m 1 gpbelimasker imgPr i 1 m n label pul la maskfr signed char imger M Figure 6 Output receptor data plots J gb s Bit Subtract ne primary MAX PI MLVIBa se POS n 8 File Edt Wiew Insert Took Desktop Window Help E i 08gdsh sseoesenz Jiamsumun Subtract no primary MAX P3 MLVIIIa sc705 te Blob Area Cumulative Hi 300 Cumulate e H ste gram 100 200 300 400 Area Bl e Git Subtract no primary MAX P3 File Edit View Insert Tools Desktop Window Help UGiea ATDA a Eag ract no primary_MAX_P3_MLVilla_s 705 titBlob Avg Intensity Cumulati 3n Cumulative Hislagram 2 5 E a LA fa B lacini Overiey B Bis Subiract no primary MAX FI MLVE crt ees Fde Eds Wew Ir Took Daite Window Help 00d hk See a gag Figure 7 Output RGB Overlay 3 2 2 Three Dimension Identification For the 3D case the user is prompted to select a group of images These images are displayed as they are loaded in Once loaded in the user is given the same minimum size settings and is provided with a list of the i
37. y will be displayed as they are loaded in along with their filename in text receptorToolsGUI User s Manual by Adam McCann When clicked the Image List will list all the image loaded into the sequence in the order that they will be stacked In the Settings group the AutoThresh Only will just disable the user interactive thresholding of the image and just use the value of the image mean intensity one standard deviation of the intensity The Min Blob Radius setting controls the radius of the minimum allowable 3D blob The possible output options are XLS Output When selected this will save a copy of the 3D blob characteristic data to an XLS file of the name specified by the user in the text dialog box below the XLS Output button You cannot use this function to append the results onto an existing XLS spreadsheet because the data values will always be written in the first three columns of the first sheet Show Plots When selected the software will display the following 5 plots regarding the blob characteristic data Histograms for receptor volume and average intensity Cumulative histograms for volume and average intensity Scatter plot of receptor volume vs avg intensity lg ine se Pe vm intscr d S000 cll 3D Blob EZ d mm B 05ec 56 66 Me intacisti z000 ENL 3D Blob A Fla di View lapet Toole Dasktap Window Help em E i Page Fle idt View laser Tools Desktop Window Help OUSGEs e be az a 0 aD E 8
Download Pdf Manuals
Related Search