Home

Breath Pacing by Auditory and Visual Cues

1. Figure 8 3 A close up view of the waveform of the 220 Hz sound used to signify the beginning of exhalation The image displays approx 300 milliseconds of the total 500 milliseconds Ping s were positioned at 0 and 3 seconds and Pong s at 4 and 8 seconds See figure 8 3 for a close up of a single Pong and figure 8 4 for the complete 10 second sound clip The final sound was exported at 44 1 kHz sampling rate 16 bit resolution PingPong wav on the CD y Figure 8 4 Waveform of the PingPong sound file with a length of 10 seconds From left to right the red arrows mark positions of 3 seconds 4 seconds and 8 seconds respectively 35 8 Design and Implementation Test 2 8 3 Demo mode and Test mode 8 3 Demo mode and Test mode In this test the participants would only be performing breath pacing by listening to audio cues However it was decided that the demonstration given before the actual test would introduce the three different pieces of audio first with an audio visual pacer reusing the progress bar from test 1 and then through audio only This was done both to make instructions more similar between the two tests and because it was found to be a useful method for instructing participants 36 9 Testing Test 2 The purpose of test 2 was to investigate whether a difference in HRV exists between using two types of audio for breath pacing continuous and discrete The Waves and PingPong sounds describe
2. www eskodijk nl files so you can listen There are a few variants of each sample 2 during playback a mix was made of different samples periodically an Inhale then an Exhale sample and the total was mixed with a long several minutes calm sea sound which is not in the zip file at low amplitude to blend everything together 3 we tried samples also from Freesound org although not sure if these are the source of the material we used These are samples 31762 by Slanesh 48412 by Luftrum if you need more info let me know regards Esko Link removed by author of this report as the agreement was to include only email correspondence 97
3. 14 3 Appendix C 14 3 Appendix C Questionnaire from test 2 continues on the following pages Breath Pacer Test 2 Questionnaire For each of the three sounds please indicate how easy or difficult it was to follow the sound with your breathing Following the ACCORDION sound with my breathing was 12 34 5 SL very difficult 5 6 O6 very easy Comments Following the WAVES sound with my breathing was 2 365 7 very difficult very easy Comments Following the PING PONG sound with my breathing was 2 345 amp y very difficult 5 t s gt very easy Comments Back Continue 40 completed 53 14 Appendices 14 3 Appendix C For each of the three sounds please indicate how pleasant or annoying it was to listen to the sound throughout the test Listening to the ACCORDION sound throughout the test was 12S amp 5 6 f very annoying very pleasant Listening to the WAVES sound throughout the test was 123 5067 very annoying 0 very pleasant Listening to the PING PONG sound throughout the test was IZ bes very annoying O O very pleasant Back Continue 60 completed 54 14 Appendices 14 3 Appendix C For each of the three sounds please indicate how easy or difficult it was to remember how the sound should guide the breathing cycle Think about
4. Breath Pacer Remembering Hearing very easy 7r very easy 7r very difficult 1 very difficult 1 Accordion Waves PingPong Accordion Waves PingPong Breath Pacer Breath Pacer Figure 10 4 Bar plots of mean and standard deviation from questionnaire responses which were given on 7 point Likert scales Error bars represent 1 SD 44 10 Results Test 2 10 0 To the question about having previous practice with breath pacing 6 replied yes although for 5 participants the practice came from participating in test 1 at least three weeks earlier One participant mentioned sporadic practice over the past 10 years To the question about practice or experience with meditation 5 replied yes and 3 of those 5 noted that the experience was minimal the most recent being a year ago Out of the remaining 2 one mentioned 10 years of budo training a form of karate where each session has a short meditation at the beginning and end The other mentioned 5 years recent yoga and meditation practice 45 11 Discussion 11 1 Discussion of test 1 At the time of writing it is unclear what lies at the root of the statistically significant result found in test 1 of mean heart rate being higher during the AudioVisual breath pacer than during the Visual breath pacer Looking at the bar plots in figure 7 2 on page 29 the significance of the difference does not intuitively come to mind Also despite the ANOVA reporti
5. ECG electrodes are typically placed on the chest but a less intrusive option is on the arms or wrists 7 These more peripheral placements however increases interference with the ECG signal caused by electrical activity from muscle tissue other than the heart Such interference can also occur when electrodes are placed on the chest or lower rib area but is minimized by placement closer to the heart 3 Analysis 3 2 Heart Rate Variability ECG and PPG signals can exhibit interference from the 50 Hz AC power that supplies computers monitors and other equipment in the room This interference may affect PPG signals less since the sensor signal is based on light although potentially the connected sensor circuitry could be affected Filtering the signal with a 50 Hz notch filter can reduce this artefact Movement artefacts is also a concern for both ECG and PPG If the wires attached to the ECG electrodes are being pulled at contact between skin and electrode may be momentarily disrupted and cause distortions in the signal which then may be processed as an extra beat or the lack of one For PPG movement of the sensor is the main cause for distortion of the signal 7 3 2 Heart Rate Variability In the heart there is a group of cells called the sinoatrial node see figure 3 1 that gener ates electrical signals causing the contractions of heart tissue known as heart beats The frequency which these signals are generated by is regula
6. and disease In Energy medicine east and west A natural history of qi New York NY Elsevier 2011 cit on pp 2 3 Hye Sue Song and Paul M Lehrer The effects of specific respiratory rates on heart rate and heart rate variability In Applied psychophysiology and biofeedback 28 1 2003 pp 13 23 cit on pp 2 10 11 Richard Gevirtz The Promise of Heart Rate Variability Biofeedback Evidence Based Applications In Biofeedback 41 3 2013 pp 110 120 cit on p 3 Paul Lehrer et al Protocol for Heart Rate Variability Biofeedback Training In Biofeedback 41 3 2013 pp 98 109 cit on pp 3 11 HeartMath Inner Balance 3 1 User Manual v3 1 http cdn heartmath com manuals InnerBalanceManual v3 1 for i10S7 pdf Accessed 05 06 2014 cit on p 4 MyBrainSolutions MyCalmBeat https www mybrainsolutions com mycalmbeat Accessed 05 06 2014 cit on p 4 Fred Shaffer and Didier C Combatalade Don t Add or Miss a Beat A Guide to Cleaner Heart Rate Variability Recordings In Biofeedback 41 3 2013 pp 121 130 cit on pp 8 9 R Benjamin Knapp Jonghwa Kim and Elisabeth Andr Physiological signals and their use in augmenting emotion recognition for human machine interaction In Emotion Oriented Systems Springer 2011 pp 133 159 cit on p 8 Axel Sch fer and Jan Vagedes How accurate is pulse rate variability as an estimate of heart rate variability A
7. and resampled to 44 1 kHz at 16 bit resolution For the final result refer to figure 8 2 below and the file Waves wav on the included CD Figure 8 2 Waveform of the Waves sound file with a length of 10 seconds From left to right the red arrows mark positions of 3 seconds 4 seconds and 8 seconds respectively 34 8 Design and Implementation Test 2 8 3 Discrete audio PingPong 8 2 Discrete audio PingPong For the breath pacer made from discrete sounds it was decided to start with an empty track and use Audacity s tone generator to generate short segments of single frequencies Two segments of the tone Cs 523 2 Hz were generated to indicate the beginning and end of inhalation respectively A sound of this frequency will here be referred to as a Ping The Ping at the beginning has a duration of 500 milliseconds ms and the Ping at the end has a duration of 250 ms The amplitude of the end Ping was decreased by 6 decibel dB For both Ping s the first 10 ms were faded in linearly and the remaining either 240 ms or 490 ms were faded out exponetially Sounds to signify exhalation were generated as the tone Az 220 Hz referred to here as a Pong Again just as the Ping s the beginning Pong had a duration of 500 ms and the end Pong a duration of 250 ms The same amplitude adjustment and envelope as described above was applied d MAIN ALMA ARARAAAARAAA m VADER VAL Mn t
8. of responses were 3 for Audio 3 for Audio Visual 5 for Visual 4 for Neither In response to the question about having previous practice with breath pacing the re sponses were 3 Yes 12 No In response to the question about experience with medita 30 7 Results Test 1 7 0 tion 6 participants had a little to some experience and the time of the most recent experience was half a year ago and up to 6 years ago Among the comments there was a slight tendency to prefer the Audio or AudioVisual pacers with the argument that they allowed the participant to close his her eyes or simply not have to look at the screen However among those who preferred the Audio pacer as well as those who did not there was mention of the pacer sounding annoying 7 0 2 Remarks It should be noted that for the 15 participants included for statistical analysis the respiration sensor signal was not always strong enough to clearly indicate breathing cycles and in other cases it is debatable whether a signal should have been counted as one exhalation or two Therefore the statistics regarding respiration can not be considered valid It should also be noted that the large standard deviation seen in HRV peak frequency for the Audio and Visual breath pacers see figure 7 1 left comes exclusively from two of the fifteen participants One of these participants showed a HRV peak frequency for the Audio pacer at 0 003 Hz while showing the expec
9. review on studies comparing photoplethysmographic technology with an electrocardiogram In nternational journal of cardiology 166 1 2013 pp 15 29 cit on p 8 Marek Malik et al Heart rate variability Standards of measurement physiological interpretation and clinical use In European heart journal 17 3 1996 pp 354 381 cit on pp 8 10 13 Bibliography 13 0 Bibliography 11 12 13 14 15 16 17 18 19 Julian F Thayer et al Heart rate variability prefrontal neural function and cogni tive performance the neurovisceral integration perspective on self regulation adap tation and health In Annals of Behavioral Medicine 37 2 2009 pp 141 153 cit on p 9 Bradley M Appelhans and Linda J Luecken Heart rate variability as an index of regulated emotional responding In Review of general psychology 10 3 2006 p 229 cit on pp 9 10 J Philip Saul Beat to beat variations of heart rate reflect modulation of cardiac autonomic outflow In News in Physiological Sciences 5 1 1990 pp 32 37 cit on p 9 Pablo Laguna George B Moody and Roger G Mark Power spectral density of unevenly sampled data by least square analysis performance and application to heart rate signals In Biomedical Engineering IEEE Transactions on 45 6 1998 pp 698 715 cit on p 10 Anne Louise Smith Harry Owen and Karen J Reynolds Heart rate variability indices for
10. whether it was difficult or easy to remember which parts of the sounds was intended for breathing in breathing out and holding your breath Remembering the breathing cycle for the ACCORDION sound was T2 3 Xx 5 7 very difficult O O very easy Comments Remembering the breathing cycle for the WAVES sound was 102 3b Gs F very difficult O O very easy Comments Remembering the breathing cycle for the PING PONG sound was 2 DA very difficult O very easy Comments Back Continue 80 completed 55 14 Appendices 14 3 Appendix C For each of the three sounds please indicate how easy or difficult it was to hear to the sound Hearing the ACCORDION sound was T2325 6 Ff very difficult O very easy Hearing the WAVES sound was 12385 8 7 very difficult O O very easy Hearing the PING PONG sound was 1235807 very difficult amp O very easy Previous experience or practice Do you have any previous practice with paced breathing E Yes E No If yes please explain please include info about how when and for how long Do you have any practice or experience with meditation Yes No Ey mi f yes please explain please include info about how when and for how long Never submit passwords through Google Forms 100 You made it 56 14
11. 0 Results Test 2 10 0 e Following For each of the three sounds please indicate how easy or difficult it was to follow the sound with your breathing 1 very difficult 7 very easy e Listening For each of the three sounds please indicate how pleasant or annoying it was to listen to the sound throughout the test 1 very annoying 7 very pleasant e Remembering For each of the three sounds please indicate how easy or difficult it was to remember how the sound should guide the breathing cycle 1 very difficult 7 very easy e Hearing For each of the three sounds please indicate how easy or difficult it was to hear to the sound 1 very difficult 7 very easy Accordion Waves PingPong M SD M SD M SD Following 6 000 0 953 6 417 0 669 4 750 1 815 Listening 3 083 1 379 6 500 0 674 4 667 1 435 Remembering 6 417 0 996 6 333 0 888 5 083 1 832 Hearing 6 917 0 289 6 833 0 389 6 083 1 084 Table 10 4 Mean and standard deviation of questionnaire responses Values have been rounded to 3 decimal places A bar plot representation can be seen in figure 10 4 on page 44 A repeated measures ANOVA was performed for each question and the results can be seen in table 10 5 F 2 22 p value Following 6 941 0 005 Listening 20 767 0 000 Remembering 7 384 0 004 Hearing 6 217 0 0
12. 0 11 052239 67 510991 9 571247 67 164279 10 417123 Amplitude of HRV bpm 20 686512 7 021152 21 884004 7 150380 20 525908 6 644200 Respiration rate br min 6 020067 0 072352 6 020067 0 072352 6 000000 0 000000 Table 10 2 Mean values and standard deviation for features extracted from the sensor data arranged according to breath pacer A bar plot representation can be seen in figure 10 1 and 10 2 on page 41 and figure 10 3 on page 42 10 Results Test 2 Frequency Hz o o e HRV peak frequency 100 Accordion Figure 10 1 HRV peak frequency and power Waves Breath Pacer Trough heart rate PingPong 80r Heart rate bpm 100 Accordion Waves Breath Pacer Amplitude of HRV PingPong 80r Heart rate bpm Figure 10 2 Trough peak and mean heart rate and amplitude of HRV Error bars represent 4 SD 60r Accordion Waves Breath Pacer PingPong 41 140 120 100 60 Power s Hz 40 20 100 80 Heart rate bpm 10 0 HRV peak power 80 Waves Breath Pacer Accordion PingPong Error bars represent 1 SD Peak heart rate 0 Accordion Waves PingPong Breath Pacer Mean heart rate 100 r r 80r J P o 60 g E E 40 1 I o T 20r 1 0 Accordion Waves PingPong Breath Pacer m kh 10 Results Test 2 10 0 Resp
13. 07 Table 10 5 Repeated measures ANOVA for the questionnaire responses F and p values have been rounded to 3 decimal places The repeated measures ANOVA reported a significant effect in responses for all four questions as seen in table 10 5 where p lt 0 05 in each case Post hoc pairwise comparisons using Tukey s HSD test reported results as follows 43 10 Results Test 2 10 0 e Following the mean for the Waves M 6 417 SD 0 669 was significantly higher than the mean for the PingPong breath pacer M 4 750 SD 1 815 e Listening the mean for the Waves M 6 500 SD 0 674 was significantly higher than the mean for the Accordion breath pacer 3 083 SD 1 379 Also the mean for the Waves M 6 500 SD 0 674 was significantly higher than the mean for the PingPong breath pacer 4 667 SD 1 435 e Remembering the mean for the Accordion M 6 417 SD 0 996 was signifi cantly higher than the mean for the PingPong breath pacer 5 083 SD 1 832 Also the mean for the Waves M 6 333 SD 0 888 was significantly higher than the mean for the PingPong breath pacer 5 083 SD 1 832 e Hearing In this case the pairwise comparisons did not show any significant results at the p lt 0 5 level Following Listening very easy 7r very pleasant 7 2r 2 very difficult 1 z very annoying 1 z Accordion Waves PingPong Accordion Waves PingPong Breath Pacer
14. A noui uu ux Rege a aec ee So e X EE Ee mA 14 27 Appendix Bois E fe he eee Se wands quus qoe vct RS 14 3 Appendix Gis uu A A REOR IA SOS INIMA WO RPG GU LA 14 47 Appendix is q d em to de oe eer RC SA d Re E ede LAURO EEG RB Ud ili 26 28 30 31 33 33 35 36 37 37 38 40 42 46 46 46 47 48 49 Preface A CD is included with this report It contains a copy of the software that has been constructed as part of the project A digital version of this report and the raw test data from each test conducted throughout the project is also included on the CD I would like to thank Stefania Serafin for supervision Thanks also goes to Niels Christian Nilsson and Thomas Saaby Noer for inspiration and conversation on the subject matter 1 Introduction The human heart beats approximately 100 000 times each day 1 When looking closer at the heart s activity it is noticed that the time between beats vary from one beat to the next This variability is influenced by many factors One of them being our breathing which especially at slow rates can contribute to a significant amount of heart rate variability occuring at the same frequency as our breathing 2 The study at hand will center around what is called breath pacing a way of guiding a person s breathing to maintain a certain breathing frequency and obtain a high amplitude of heart rate variability while following the breath pacing 2 Pre analysis W
15. Aalborg University Copenhagen Semester MED 10 Title Breath Pacing by Auditory and Visual Cues Aalborg University Copenhagen Frederikskaj 12 DK 2450 Copenhagen SV Denmark Project Period Spring 2014 Semester Coordinator Secretary Semester Theme Master thesis Abstract Supervisor s Stefania Serafin This study investigates different modalities for breath pacing guided breathing and their effect on heart rate variability HRV Project group no Two tests have been performed In test 1 the breath pacers were auditory audio visual and visual In test 2 the breath pacers were all auditory and consisted of a continuous accordion like sound a Members TN continuous sound of ocean waves and a discrete Christian Toft ping pong type of sound A significant result was found in test 1 for the effect of breath pacer type on mean heart rate where the audio visual pacer resulted in a higher mean heart rate than the visual pacer The result should be approached with caution as the cause is unclear and should be investigated further In test 2 a significant result was found for the effect of breath pacer type on peak heart rate where the ping pong sound resulted in a lower peak heart rate than the waves and accordion Again the cause for the result is unclear and should be investigated further The waves sound was found to be significantly more pleasant to listen to than the other two sounds and
16. Appendices 14 4 Appendix D 14 4 Appendix D Personal correspondence from Esko Dijk co author of the Philips Research paper 19 mentioned in section 3 3 on page 11 Re Ocean waves as a breathing guide eskodijk gmail com on behalf of Esko Dijk eskoQieee org Sent Monday June 16 2014 21 34 To Christian Toft Hello Christian what we created for the setup is two audio samples of waves crashing on the shore one with the deep low rumble and one with the high noisy sound of the final phase of the wave The software we made plays these samples in the proper rhythm It can be set to any speed in terms of cycles per minute That is also used for the adaptive version which adapts to the user s breathing rate Compared to a real audio recording of the sea our approach can get a bit monotonous because the sampled sounds are always the same however for a short test period this is not an issue It could be for daily use etc I can tomorrow at work look up a bit more details about the setup see if I can tell you more on this by the end of tomorrow best regards Esko Re Ocean waves as a breathing guide eskodijk gmail com on behalf of Esko Dijk eskoQieee org Sent Wednesday June 18 2014 10 19 To Christian Toft Hello I looked at the code of our test setup and found some more info 1 samples used origin of the samples was not mentioned there but to give you an idea I have put the sample clips used on my website http
17. C to easily identify and extract values on the PC side The data was saved to the computer s hard drive in text files for later analysis n example of the data sent to the PC can be seen in figure 4 13 A1035319B797C330 A1035329B789C333 A1035340B779C329 A1035349B767C331 A1035359B752C328 Figure 4 13 An example of the data sent from the Arduino to the PC The timestamp is prefixed by an A and indicates the number of milliseconds passed since the Arduino started sending sensor data The pulse wave signal is prefixed by a B and finally the respiration signal prefixed by a C The complete code for the Arduino Uno can be seen in Appendix B on page 52 4 8 Monitoring of sensor signals In order to ensure that sensors had been placed properly and the signal output was acceptable a monitoring function was programmed as part of the test application This 21 4 Design and Implementation Test 1 4 8 Monitoring of sensor signals function plotted the pulse wave and respiration signals in real time on screen and allowed the experimenter to adjust sensor placement or signal gain if necessary before proceeding with the test see figure 4 14 The plot of the sensor signals was only intended to be seen by the experimenter and would not be displayed on screen for the test participants Pulse Wave Pacto Number Gender o reae Age Respiration Figure 4 14 The BreathPac
18. Pacer Figure 7 2 Trough peak and mean heart rate and amplitude of HRV Error bars represent 1 SD 29 7 Results Test 1 7 0 Respiration rate Respiration rate breaths min Audio AudioVisual Visual Breath Pacer Figure 7 3 Respiration rate Error bars represent 1 SD A repeated measures analysis of variance ANOVA was performed for each feature and the results can be seen in table 7 3 F 2 28 p value HRV peak frequency 0 473 0 628 HRV peak power 1 184 0 321 Mean heart rate 4 105 0 027 Peak heart rate 1 131 0 337 Trough heart rate 2 540 0 097 Amplitude of HRV 0 639 0 535 Respiration rate 0 074 0 929 Table 7 3 Repeated measures ANOVA F and p values rounded to 3 decimal places The repeated measures ANOVA reported a significant effect of breath pacer type on mean heart rate F 2 28 4 105 p 0 027 A post hoc pairwise comparison using Tukey s HSD test showed that the mean for the AudioVisual breath pacer M 77 014 SD 7 163 was significantly greater than for the Visual breath pacer M 75 407 SD 7 537 in respect to mean heart rate 7 0 1 Questionnaire responses In response to which breath pacer was found to be the easiest to follow the number of responses were 3 for Audio 7 for Audio Visual 3 for Visual and 2 for Neither In response to which breath pacer was found to be the the most difficult to follow the number
19. als could be calculated and used for statistical analysis A script written by Dmitry Savransky and made available on the Matlab File Exchange website was used for computing power spectral density of HRV with the Lomb Scargle method see Analysis of HRV on page 10 Pulse Wave and Respiration Heart Rate T T T4 T 1 1 1 Figure 6 1 Screen capture of the application that was developed in order to inspect the recorded sensor data and extract features for analysis The respiration sensor signal appeared to exhibit 50 Hz noise possibly from line power supplying other equipment connected to or close by the sensor circuitry To remedy the effect a 50 Hz notch filter was applied As previously mentioned see Test procedure on page 24 each breath pacer would run for 5 minutes The data that was considered for analysis was extracted from a window beginning 1 minute into the breath pacing and having a duration of 3 minutes and 50 seconds thus ending 10 seconds before the breath pacer stopped http www mathworks com matlabcentral fileexchange 20004 lomb lomb scargle periodogram 6 Processing of Sensor Data 6 0 The features that were extracted from the sensor data were as follows HRV peak frequency HRV peak power Mean heart rate e Peak heart rate Trough heart rate Amplitude of HRV Respiration rate HRV peak frequency and power and mean heart rate were calculated from the total 3 mi
20. athing The experimenter would then proceed along the following points e Explain that the test was about breath pacing and what was meant by that e Explain that there would be a visual audio visual and an audio pacer e That each pacer would last for 5 minutes preceded by a pause of 1 minute e Assist the participant in placing sensors and finding a comfortable position to rest the hand with the pulse sensor requesting that it be held still during the test e Turn off the monitor facing the participant and check sensor signals via the laptop adjusting sensor placement or gain if necessary e Clear the sensor signal plots from the screen and turn the monitor back on e Enter information about the participant number gender and age 24 5 Testing Test 1 5 3 Test procedure Place headphones on the participant Start the demo mode of the test application Explain that the experimenter would leave the room at the start of the test and come back in after 18 minutes the duration of the test Make sure that the participant was ready to begin the test Start the test mode of the test application and leave the room Come back to the test participant after 18 minutes and help remove sensors An A4 sheet of paper with screen images of the three breath pacing modes was used when first explaining how the cycle of inhalation hold exhalation hold would correspond to the progress bar It was explained that the order in
21. ating autonomic activity 17 18 Studies have found that each individual has a specific resonant breathing frequency which maximizes the HRV amplitude Resonant breathing frequencies were found to be in the range of 4 7 breathing cycles per minute among healthy individuals 2 17 18 Some researchers mention that for optimum effect breathing out should be slightly slower than breathing in Breathing should also be abdominal rather than thoracic meaning one should use the diaphragm muscles so that the abdomen expands and air is pulled into the lungs rather than using muscles in the chest area It should be a relaxed effort to not cause hyperventilation 4 17 In the CardioSense Trainer application from Complete Coherence see subsection 2 1 3 on page 5 the default 10 second breathing cycle 6 breaths per minute consists of breathing in for 3 seconds holding ones breath for 1 second breathing out for 4 seconds and holding for 2 seconds 3 3 Breath Pacing In research and commercial products related to HRV and respiration some form of breath pacing breathing guidance is used in order to achieve and maintain the desired rhythm In the reviewed literature concerning HRV biofeedback training there is briefly mention of a rising and falling bar on a computer monitor as a visual example of breath pacing or a tone that rises and falls in pitch as an auditory example After a number of training sessions using a pacer bar the c
22. ayers For the first 3 seconds representing inhalation two segments of incoming wave sounds were layered so that when one faded out the other was increasing its intensity effectively building a longer period of increasing intensity The beginning and end of both segments https www freesound org 8 Design and Implementation Test 2 8 2 Continuous audio Waves o T 4 Sess dise A mac as 3 5 M Length 00 0008 00k 00 abel ie 600 Figure 8 1 Screen capture from Audacity while editing the Waves sound were faded in and out respectively to transition into an underlying sound of gently trickling water representing the 1 second hold Next a less intense segment of waves rolling in was reversed effectively sounding like waves receding while distinguishing itself from other sound segments this part was used for the 4 seconds exhalation period Again the beginning of the segment was faded in and the end was faded out to blend with other layers For the 2 seconds hold another clip of gently trickling water was used In order to make the final sound clip loop seamlessly a segment of the trickling water extending beyond the 10 second mark was cut and moved to the beginning of a layer where it would blend and fade into the inhale wave The audio layers were finally combined to form a 10 second mono sound clip
23. d in Design and Implementation Test 2 on page 33 were used and additionally the Accordion sound from test 1 was included for reference The setup used in test 2 was identical to that of test 1 described in Test setup on page 23 The test application was slightly modified from test 1 to accommodate the procedure in test 2 described in the following sections 9 1 Questionnaire Questions similar to those in test 1 were included for test 2 In regards to any previous practice with breath pacing participants who recurred from the first test were asked to make note of this in their response Additional questions that were included in test 2 are described below To investigate preferences participants were asked to rate each of the three sounds on a 7 point scale in regards to how pleasant or annoying it was to listen to the sounds throughout the test Participants were also asked to indicate how easy or difficult it was to remember how the sounds should guide the breathing cycle Although it would be explained before the test started that breath pacing would always begin with inhalation this could be forgotten or overheard and could possibly influence results It was also asked how easy or difficult it was to hear each of the sounds This question was simply a verification to check that no sounds were perceived by the test participants to be too low Please refer to Appendix C on page 53 to view the questionnaire used in t
24. ded lab com 19 4 Design and Implementation Test 1 4 7 Data acquisition implemented in a belt around the chest or abdomen However due to time restrictions it was decided to use the e Health Sensor Platform airflow sensor see figure 4 10 to detect breathing The e Health Sensor Platform is designed to be used in tandem with e g an Arduino Uno microprocessor see Data acquisition on page 20 The airflow sensor is based on thermistors which change resistivity when temperature changes and is designed to be placed under the user s nose where it reacts to the cooling airflow during inhalation and warming during exhalation A limitation in the way this particular sensor is implemented however is that only exhalation is indicated in the signal output by the eHealth sensor circuitry inhalation and holding one s breath are both returned as the value zero whereas exhalation results in positive values Figure 4 10 eHealth airflow sensor connected to eHealth shield Image from http www cooking hacks com 4 7 Data acquisition An Arduino Uno microcontroller was used in combination with the e Health sensor shield see figure 4 12 The Arduino Uno features both digital and analog connection inputs and functions as an analog to digital converter with a resolution of 10 bits It samples the voltage output from a sensor and converts that signal to a value in the range of 0 1023 The values can then be transmitted to a computer
25. e bpm 85 764879 7 619335 86 805080 6 559240 86 217926 7 075486 Trough heart rate bpm 65 339128 8 664732 67 179210 8 452573 65 720184 8 799919 Amplitude of HRV bpm 20 425751 5 672465 19 625871 5 854586 20 497741 6 016019 Respiration rate br min 5 530435 1 251863 5 634783 1 414481 5 617391 1 344235 Table 7 2 Mean values and standard deviation for features extracted from the sensor data arranged according to breath pacer A bar plot representation can be seen in figure 7 1 and 7 2 on page 29 and figure 7 3 on page 30 7 Results Test 1 7 0 HRV peak frequency HRV peak power 0 14 r r 120 T r 0 12r 7 0 1F z 2 gt 0 081 an Q v 5 P 2 0 06 1 2 3 Ex a 0 047 1 0 027 1 0 Audio AudioVisual Visual Audio AudioVisual Visual Breath Pacer Breath Pacer Figure 7 1 HRV peak frequency and power Error bars represent 1 SD Trough heart rate Peak heart rate 100 r r 100 r r 80r 1 E E o Qa 2 60r 7 E g 40 4 oC oC o o X T 20r 1 0 Audio AudioVisual Visual Audio AudioVisual Visual Breath Pacer Breath Pacer Amplitude of HRV Mean heart rate 100 r r 100 T r sof 80 I A E Qa Qa 2 60r 7 2 60r o o g E 40 1 E 40 1 I I o v T T 20r q 20r 1 0 Audio AudioVisual Visual Audio AudioVisual Visual Breath Pacer Breath
26. e decreases 12 For adults normal respiratory rates are in the range of 12 20 breaths per minute This translates to 0 2 0 333 Hz in the HRV frequency spectrum thus normally placing RSA in the HF band 2 16 3 2 3 Resonant Breathing Frequency At respiratory frequencies that are slower than the normal adult rate other mechanisms add to the effect seen in HRV In the low frequency LF range the relatively slow influence on heart rate coming from the sympathetic nervous system plays a role Also 10 3 Analysis 3 3 Breath Pacing the baroreflex system which is part of a system that regulates blood pressure is thought to play a role in the LF range The response from the baroreflex system originates in baroreceptors sensors that react to stretching of blood vessels caused by increased blood pressure which in turn resulted from an increase in heart rate In summary an increased blood pressure causes the baroreflex to decrease heart rate and vice versa 16 17 It has been found that at respiratory frequencies around 0 1 Hz 6 breaths per minute there appears to be a resonance between different sources of HRV and the amplitude of heart rate oscillations become particularly high It is theorized that a beneficial side effect of the increased amplitude of oscillations in heart rate causing greater oscillations in blood pressure as well exercises the baroreflexes more and ultimately leads to greater efficiency in modul
27. eft side is from a participant while there was a pause before breath pacing The plot on the right side is from the same participant in the following minute where breath pacing took place 27 7 Results Test 1 Originally 20 people participated in the test Some test data was rejected In one case because the test participant was coughing during breath pacing severely affecting the pulse wave signal In some cases because of inaccurate pulse peak detection e g in two cases the pulse signal was clipped due to the sensor voltage exceeding the Arduino s maximum input range The test data considered below is for 15 participants 11 male 4 female age 23 41 years mean 28 4 standard deviation 4 896 In the following mean and standard deviation will be abbreviated as M and SD Breath pacer sequence participants Ist 2nd 3rd 5 Audio AudioVisual Visual 3 Audio Visual AudioVisual 1 AudioVisual Audio Visual 1 AudioVisual Visual Audio 2 Visual Audio AudioVisual 3 Visual AudioVisual Audio Table 7 1 The distribution of breath pacer sequences among the 15 participants Audio AudioVisual Visual M SD M SD M SD HRV peak frequency Hz 0 100474 0 037245 0 100354 0 000296 0 106923 0 025879 HRV peak power s Hz 19 992791 26 875805 72 043928 28 057212 80 303084 25 119667 Mean heart rate bpm 15 246157 8 091928 11 014328 7 163005 15 406837 7 536848 Peak heart rat
28. eight barHeight modAvPacer 3 0 set hPanelBPbarAV Position 0 0 1 if within 3 to 4 sec elseif modAvPacer 3 amp amp modAvPacer set bar at max height 1 set hPanelBPbarAV Position 0 0 1 if within 4 to 8 sec elseif modAvPacer 4 amp amp modAvPacer 8 set bar height barHeight 1 modAvPacer 4 set hPanelBPbarAV Position 0 0 1 if greater than 8 sec elseif modAvPacer 8 set bar at min height 0 001 set hPanelBPbarAV Position 0 0 1 end 4 3 Gapless audio playback units are relative within the bar frame barHeight 0 001 4 0 barHeight 0 001 since 0 is not allowed 0 0011 4 2 Gapless audio playback It was found that in order to achieve looped playback of the 10 second audio file without any gaps between repeats it was necessary to preload the audio and build an array con taining the required duration of looped audio This was implemented with the following Matlab code info audioinfo audio Accordion wav 9 loading audio Fs is sample rate a Fs audioread audio Accordion wav preallocate array for repeating the audio audioLoopedLength Fs max pacerDurationDemo pacerDurationTest 0 sooped audioLooped audioLoopedLength 1 integerRepeats floor audio ength info TotalSamples remainderSamples mod audioLoopedLength info TotalSamples fill the array with repetition
29. ened to just a few seconds and each breath pacing mode was demonstrated for 20 seconds equalling 2 respiration cycles In demo mode the sequence was fixed so the visual breath pacer would be presented first then the audio visual pacer and finally the audio pacer This sequence allowed the experimenter to explain how the intended inhale hold exhale hold breathing cycle corresponded with the progress bar and the audio 4 5 Randomization of test sequence The actual test sequence experienced by the participant was randomized at the start of the test The Matlab code below shows the implementation see figure 4 6 and 4 7 17 4 Design and Implementation Test 1 4 6 Sensors 9 seed random number generator based on current time when application starts rng shuffle matrix of all possible sequences 1 Audio 2 AudioVisual 3 Visual sequences 1 2 3 rer 1 3542 23516 2y lpas Se Los 3 152 li the default demo sequence Actual test sequence is randomized stSequence sequences 5 oe ct Figure 4 6 At application startup the random number generator is seeded based on date and time All possible test sequence combinations is held in a 6x3 matrix oe For test mode the sequence is randomly selected stSequence sequences randi 6 ct Figure 4 7 As part of the function that starts the test the test sequence is randomized For each partici
30. er test application while monitoring sensor signals The pulse wave is plotted in red in the upper plot window and respiration in blue in the lower plot window The sensor signals were not shown to test participants 22 5 Testing Test 1 The purpose of test 1 was to investigate whether a difference in HRV exists between three modalities of breath pacing audio audio visual and visual 5 1 Questionnaire As a supplement to the main data the pulse wave signal test participants would be asked to fill out a short questionnaire see Appendix B on page 52 The first questions were in regards to which if any of the three pacing modes was found to be the easiest or most difficult to follow and why Next participants were asked if they had any previous practice in paced breathing and if so elaborate on that The intention behind these questions were to highlight a possible cause in case some participants performed markedly more consistent across all pacing modes and others did not A question that was not included in the on screen questionnaire but asked orally was if participants had any experience with meditation In meditation it is not uncommon to focus on ones breathing and so this may possibly influence results 5 2 Test setup Visual and auditory breath pacing was delivered via a Philips 244E computer monitor 24 inches 1920x1080 pixels and Sennheiser HD202 headphones Wireless keyboard and mouse was used by par
31. est 2 9 Testing Test 2 9 2 Test procedure 9 2 Test procedure As in test 1 the experimenter would make sure that the participant s mobile phone would not disturb the test and the experimenter would then proceed along the following points similar to test 1 e Explain that the test was about breath pacing and what was meant by that e Explain that there would be 3 different audio pacers e That each pacer would last for 5 minutes preceded by a pause of 1 minute e Assist the participant in placing sensors and finding a comfortable position to rest the hand with the pulse sensor requesting that it be held still during the test e Turn off the monitor facing the participant and check sensor signals via the laptop adjusting sensor placement or gain if necessary e Clear the sensor signal plots from the screen and turn the monitor back on e Enter information about the participant number gender and age e Place headphones on the participant e Start the demo mode of the test application e Explain that the experimenter would leave the room at the start of the test and come back in after 18 minutes the duration of the test e Make sure that the participant was ready to begin the test e Start the test mode of the test application and leave the room e Come back to the test participant after 18 minutes and help remove sensors It was explained that the order in which the pacing sounds would appear during the actual te
32. hat were created as described in the following sections The software application developed for the second test was based on the application developed for the first test with only a few modifications required Please refer to the file BreathPacerTest_2 m on the CD included with this report for the complete code The sensors and data acquisition method used in test 2 were the same as in test 1 and will not be reiterated here 8 1 Continuous audio Waves In order to find a suitable recording of sea waves as the starting point for creating a 10 second clip for breath pacing the freesound org website was searched After lis tening to over a hundred audio recordings of sea waves the choice fell on a file named MareMio wav freesound org id 219120 by the user Kigofix The recording has a duration of approx 3 5 minutes and a sampling rate of 48 kHz in 24 bit stereo It features fairly calm sea waves rolling onto a beach and the trickling sound of the water receding without sounds of sea gulls or people It was difficult to find segments that in themselves had natural attack sustain and decay periods matching to the aforementioned breathing cycle 3 sec inhalation 1 sec hold 4 sec exhalation 2 sec hold Thus segments were copied from one or both channels of the original file and pasted into six new track layers see figure 8 1 were each segment was modified to fit the desired duration and blend with the other l
33. he bar holds at the top or bottom Figure 4 1 The visual pacer as implemented in the BreathPacer test application 4 1 2 Auditory breath pacing There would be virtually countless options for designing auditory stimulus for breath pacing as long as the stimulus provides the user with an indication of the intended timing in the respiratory cycle For this test the choice was made to use the same audio that occurs as the default option in the CardioSense Trainer application see CardioSense Trainer on page 5 The audio in the CardioSense Trainer application is created by using standard Windows MIDI sounds and it is possible to some degree to select your own pacer audio by choosing from a list of MIDI sounds The default pacer audio is composed of four MIDI instruments named TangoAccordian Accordian Harmonica and Reed Organ For the purpose of this test the CardioSense Trainer pacer audio was captured by using Camtasia and exporting the audio as a WAV file The audio was then cropped using Audacity to represent the complete 10 second breath pacing cycle Furthermore the audio was converted to mono since no stereo effects appeared in the original audio Henceforth this audio clip will simply be referred to as the Accordion sound See figure 4 2 for an image of the 10 second waveform The sound file is available on the CD included with this report Camtasia is a screen recording software by TechSmith See http w
34. hen consulting a physician and having ones pulse checked it is common to be informed about ones heart rate by a single number This is an average often found by feeling the pulse beats from an artery near the skin and counting beats within e g 15 seconds and then multiplying that count to arrive at a number of beats per minute bpm 1 However the time interval between beats varies from beat to beat and these consecutive variations are termed heart rate variability HRV 0 859 sec 0 793 sec 0 744 sec 0 721 sec 76 bpm 1 sec 2 sec 4 3 0 seconds of heart beat data Figure 2 1 Illustration of an electrocardiogram displaying the electrical heart signal Signal peaks indicate heart beats The time interval between heart beats vary from beat to beat Image from 1 In recent years there has been a great deal of interest in HRV biofeedback which involves guiding the user to breathe at slower than normal rates in order to maximize the ampli tude of HRV Clinical research in HRV biofeedback has shown to ameliorate a number of disorders such as anxiety depression asthma and more although further research is called for 3 4 The existing research seems to have also led to a number of HRV biofeedback products being made for the consumer market as will be discussed next 2 1 State of the Art This section will take a look at some of the HRV biofeedback products com
35. ial communication at 115200 bps Serial print a sending a character to PC while a a a Serial read wait for an a from the PC timeStart millis the time when we proceed to the main loop void loop loopEntryTime micros keep track of time at beginning of loop if Serial available gt 0 check if PC has sent any data a Serial read if we did NOT receive a b from the PC then send sample data if a b airFlow analogRead Al sample eHealth airflow sensor pulseWave analogRead A0 sample pulse sensor time millis timeStart timestamp for sensor data concatenate string of prefix characters and sensor data stringToSerial preA time preB pulseWave preC airFlow send string to serial port Serial println stringToSerial idle here until nearly 10 milliseconds have passed while micros loopEntryTime lt 9996 14 Appendices 14 2 Appendix B 14 2 Appendix B Questionnaire from test 1 Which breath pacing was the easiest to follow O Audio Audio Visual Visual Neither Which breath pacing was the most difficult to follow Audio Audio Visual Visual 5 Neither If you found any of the breath pacing modes easier or more difficult please explain why Do you have any previous practice with paced breathing Yes O No If yes please elaborate 52 14 Appendices
36. icant effect of breath pacer on peak heart rate and once again the result does not intuitively present its significance see figure 10 2 on 11 Discussion 11 3 General discussion page 41 In this case it concerns the peak heart rate being lower for the PingPong pacer compared to the other two pacers The dominant frequency in HRV during breath pacing was again found to be almost exactly 0 1 Hz as expected and this time with no exceptions The two participants that were the source of exceptions in this measure in test 1 did not participate in test 2 From the questionnaire responses it is interesting to note that the discrete PingPong sound receives the lowest mean score for the question about how easy it was to follow with ones breathing Statistically the Waves sound stands out as scoring significantly higher being easier to follow than the PingPong sound In the evaluation of how pleasant the sounds were to listen to during breath pacing the Waves sound received a significantly higher score than both the Accordion and the PingPong sound The Waves stand out particularly from the Accordion sound originally found in the CardioSense Trainer product CardioSense Trainer on page 5 The Waves used in the test were repeated without any variation it could be considered that adding slight variations in the wave sounds not in the overall timing may lead to an even more pleasant listening experience and thus a more attractive p
37. iration rate Respiration rate breaths min N w D uw oOo bel m Accordion Waves PingPong Breath Pacer Figure 10 3 Respiration rate Error bars represent 1 SD A repeated measures ANOVA was performed for each feature and the results can be seen in table 10 3 F 2 24 p value HRV peak frequency 0 993 0 385 HRV peak power 0 104 0 902 Mean heart rate 1 704 0 203 Peak heart rate 6 977 0 004 Trough heart rate 0 842 0 443 Amplitude of heart rate 2 303 0 122 Respiration rate 0 480 0 929 Table 10 3 Repeated measures ANOVA F and p values rounded to 3 decimal places The repeated measures ANOVA reported a significant effect of breath pacer type on peak heart rate F 2 24 6 977 p 0 004 A post hoc pairwise comparison using Tukey s HSD test showed that the mean for the PingPong breath pacer M 87 690 SD 10 512 was significantly lower than for the Accordion breath pacer M 88 878 SD 10 730 in respect to peak heart rate The PingPong breath pacer mean was also significantly lower than the Waves breath pacer M 89 395 SD 10 028 in respect to peak heart rate 10 0 1 Questionnaire responses Unfortunately the responses from one of the participants were lost due to an error thus only 12 sets of responses is analysed in the following The questions that were responded to with 7 point Likert scales will be referred to as follows 42 1
38. k 2e 4 3 Pauses between breath pacing es 44 Demo mode and Test mode 0 00000 ee eee eee 4 5 Randomization of test sequence Z6 SCNSOLS A RAE ee PR A SRA 4 6 1 LS AAA Pet bai he mom et 4 6 2 Respiration sensor 2 ee ee Rx A Dataacquisition ssthi mos OO ee Ee ee le ee OE A 4 8 Monitoring of sensor signals es 5 Testing Test 1 pt Questionnalte 234g A age al al AU RR URP wie eee cp Dade ALESE SOG Up ore een tte tod npe AAA A ote a N NDOT KF BW WwW oo oo 00 e H O O o 13 13 14 14 15 16 17 17 17 18 18 19 20 21 10 11 12 13 14 5 9 Test procedure 9 du a e aa dk dk ott Processing of Sensor Data Results Test 1 7 0 1 Questionnaire responses ll ers 1 0 2 Remarks uu eb eee BRED he eee ee hh ba as Design and Implementation Test 2 8 1 Continuous audio Waves hs 8 2 Discrete audio PingPong o ee ee 83 Demo mode and Test mode 0 0 00000000 eee eee Testing Test 2 9 1 Questionnaire 2020244244402 Ree hei e A 9 2 Test procedures emitir aa ra eee eee eee See aches Ab Results Test 2 10 0 1 Questionnaire responses ee a Discussion TET Discussion Ob test T c re a ae ae Ee RETE ii E ROS s a 11 2 Discussion of test 2 ees 11 3 General discussion 2 lees Conclusion Bibliography Appendices 14 1 App ndIix
39. lient is at some point encouraged to use the display of his her own heart rate as a pacing guide again with the aim of maximizing HRV amplitude 4 17 However there appears to be very little research investigating any use of different types of auditory breath pacing in relation to HRV 11 3 Analysis 3 3 Breath Pacing A paper from Philips Research explores a system concept for a multimodal breathing guidance system intended to help a user relax 19 The breathing guidance consists of a vibrotactile blanket modulated room lighting and the sound of ocean waves on a shore These three guidance modalities acted in unison either reflecting the user s own breathing or controlled to guide the user to a certain breathing rate The audio is rendered in software by mixing sound recordings of approaching and retreating waves in order to adjust to the user s breathing or create the desired cycle length Personal correspondence from one of the authors is included in Appendix D on page 57 The authors do set out to track HRV but do not report the results due to discovering inaccuracies in their real time detection of HRV amplitude 12 4 Design and Implementation Test 1 The first test was designed to investigate whether a difference in HRV exists between three modalities of breath pacing audio audio visual and visual This chapter will describe the design and implementation of the software application that was created to pe
40. m MyBrainSolutions is also mar keted as a stress reduction tool It is based on an ear clip sensor to track the user s pulse and breath pacing to maximize HRV It appears to differ somewhat from the Heart Math product by first guiding the user through an optimization process where the user s personal best breathing rate 6 is found The application offers a visual breathing guide shown as an illustration of lungs being filled or emptied but the product webpage does not make it clear whether a form of auditory breathing guide is also included http www heartmath com https www mybrainsolutions com mycalmbeat 2 Pre analysis 2 1 State of the Art Figure 2 3 MyCalmBeat product suite Image from https www mybrainsolutions com 2 1 3 CardioSense Trainer The CardioSense Trainer from Complete Coherence is yet another example of a com mercial product based on breath pacing and HRV again with an ear clip pulse sensor The application has options for both visual and or auditory breath pacing Jal Mode View Select Sereen Zommanc Client Session Help Baw ts J e M b ox so BreathPacer Heart Rate bpm Le E ga 66 H E A e m ol al s T E zu E LI NE il z Maximize selected window lo the size ol whole screen ee A Figure 2 4 A screen capture fro
41. m the CardioSense Trainer software The breath pacer is the vertical column on the left side The upper horizontal graph window displays heart rate over time The lower horizontal window displays a so called coherence score Image from video at http www youtube com watch v orKZ nJZTfs The visual breath pacer is a vertical column resembling bellows or a progress bar The auditory pacer is based on MIDI Musical Instrument Digital Interface instruments already installed with the PC s operating system The default sound is a mix of accordion and organ sounds that rise in pitch to indicate inhalation hold pitch to indicate breath http www complete coherence com technology 2 Pre analysis 2 2 Discussion hold and fall in pitch to indicate exhalation It is possible to select other MIDI based instruments to be used for breath pacing 2 1 4 The Dash Although not currently a product marketed along the lines of those mentioned above the Dash is of interest because of its potential for a breath pacing and heart rate variability related application 12 08 Workouts Adjust your workouts Select to workout and control your goals and limits a The Dash in ear headphones next to an iPhone b The Dash in use Figure 2 5 The Dash wireless in ear headphones with built in pulse sensor The product was funded via a Kickstarter campaign by Bragi At the time of writing it is expected to go into production by Jan
42. mercially available today At the end of this section a product will be presented that is not currently available as a HRV biofeedback device but has potential for such an application 2 Pre analysis 2 1 State of the Art 2 1 1 Inner Balance by HeartMath HeartMath offers a number of products based on breath pacing and heart rate variability and markets them as improving wellness and reducing stress One such product the Inner Balance for iOS consists of a photoplethysmographic pulse sensor that can be clipped onto an earlobe and software for iOS devices Apple iPhone iPad iPod e os Carrier P 2 24 PM 82 gt eec Carrier 2 24 PM 82 Session o Session e Breathe with the breath pacer 0 o ese Sg at of o b Screen image from the app c Screen image showing fre a Inner Balance app and The centre figure contracts and quency analysis of heart rate vari pulse sensor for iPhone Image expands to guide breathing Im ability bar chart and pulse sig from http succeeder se age from 5 nal Image from 5 Figure 2 2 Inner Balance for iOS from HeartMath The application offers 3 different visual breath pacers but the manual available online does not mention breath pacing based on audio 5 The user can keep logs of training sessions and receive a so called coherence score based on frequency analysis of his her HRV 2 1 2 MyCalmBeat Similar to the HeartMath product MyCalmBeat fro
43. ng a significant effect it could be considered questionable how significant that effect is in practice after all the two means are at approximately 77 and 75 4 bpm giving a difference of approx 1 6 bpm Furthermore it should be considered that what is sought achieved with breath pacing in relation to HRV is a higher amplitude in the oscillations of heart rate not generally a higher heart rate The dominant frequency in HRV during breath pacing was found to be almost exactly 0 1 Hz as expected with very few exceptions meaning that participants closely followed the breath pacing cycle in each modality A subsequent review of the HRV data from the participant that had a peak HRV fre quency of 0 003 Hz for the Audio pacer showed that the participant s heart rate varied relatively little It also showed a trend towards a slowly but steadily climbing heart rate during that particular breath pacer which also happened to be the first breath pacer in that participant s test The relatively low variability plus the trend in slowly increasing heart rate resulted in the main peak frequency being at the very low end A secondary peak was found around the expected 0 1 Hz in this case The comments about the audio breath pacer being annoying to listen to are interesting from the perspective of testing other sounds and gathering participants responses to those sounds 11 2 Discussion of test 2 In test 2 the ANOVA reported a signif
44. nute 50 second window Peak and trough heart rate was calculated for consecutive 10 second windows and then finally calculating the mean of those consecutive peaks and troughs Amplitude of HRV was calculated as the difference between peak and trough Respiration rate was calculated as a mean rate of the exhalation cycles during the total 3 minute 50 second window Heart Rate 90 f Lo Po on Fo 8 P oU fn So 0 g ap o E E t 5 j 5 a fe a Pr B 80 i 9 9s e P wa f a 65 e OOS i ji P Ed ox fol jJ i 1 N p oa d f amp a 60 Sooo s Saga voo o00 No breath pacing Breath pacing E l l Omn 1min 2 min Figure 6 2 An example of heart rate from one of the test participants The plot shows 2 minutes of beat to beat heart rate The first minute is during a pause with no breath pacing The second minute is with breath pacing with a cycle length of 10 seconds for a full respiration cycle HRV Power Spectral Density HRV Power Spectral Density 30r N e Power s Hz m un Ul ie 0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 0 1 0 2 0 3 0 4 0 5 0 6 0 7 Frequency Hz Frequency Hz Figure 6 3 Example of HRV power spectral density each from one minute of a time series of interbeat intervals The plot on the l
45. pant a file was saved with information about the test sequence experi enced by that participant Also timestamps were saved indicating when the individual pacers started helping to identify relevant sections in the sensor data 4 6 Sensors The basic requirements in terms of tracking physiological data in this experiment is to obtain a time series of beat to beat heart rates and monitor respiration in participants Although respiratory information can be extracted from the heart rate due to respira tory sinus arrhythmia it was decided to supplement the data by adding a sensor for respiration 4 6 1 Pulse sensor For tracking heart rate it was decided to use a photoplethysmography PPG sensor A PPG sensor can be placed e g on a user s fingertip or earlobe and is non intrusive whereas electrocardiographic sensors placed on the participant s chest could feel intrusive Three different PPG sensors were assessed for reliability and consistency of signal quality One sensor was a transmissive PPG sensor from Seeedstudio see figure 4 8a that is designed with an ear clip and circuitry that connects to a digital input on e g Arduino microprocessor boards It was found that this sensor often failed to register a heart beat and since the circuitry was not designed to transmit the actual pulse wave signal it did not readily lend itself to other methods of detecting heart beats Another sensor that was 18 4 Design and Im
46. plementation Test 1 4 6 Sensors considered was the reflective PPG sensor see figure 4 8b designed by Joel Murphy and Yury Gitman and originally brought to market via a Kickstarter campaign It includes options for placement on a finger or earlobe but it was found that it required some effort in order to achieve good signal strength and it was prone to movement artefacts b Pulse sensor by Joel Murphy a Heart rate sensor from Seeedstudio and Yury Gitman Figure 4 8 Two pulse sensors that were assessed for use in the test but discarded Images from a http www seeedstudio com and b http pulsesensor myshopify com The PPG sensor chosen for the test setup was the EasyPulse v1 1 from Embedded Lab This transmissive PPG sensor comes with a flexible rubber sleeve that fits onto a user s fingertip This device appeared slightly more tolerant to movement before showing artefacts in the pulse wave signal and overall signal strength was good Furthermore the included circuitry was designed to allow adjustment of signal gain Figure 4 9 The EasyPulse v1 1 sensor from Embedded Lab was chosen as the pulse sensor for the test setup Image from http embedded lab com 4 6 2 Respiration sensor It was considered to build a respiration sensor by using a stretch sensor an elastic material that changes resistivity according to the amount of stretching which could be https www kickstarter com http embed
47. rate variability HRV In test 1 the breath pacers were auditory audio visual and visual In test 2 the breath pacers were all auditory and consisted of a continuous accordion like sound a continuous sound of ocean waves and a discrete ping pong type of sound The waves and ping pong sounds were constructed for this study whereas the accordion sound was found in a HRV biofeedback application that employs breath pacing Furthermore it was investigated which sounds participants found to be the easiest to follow and most pleasant to listen to A significant result was found in test 1 for the effect of breath pacer type on mean heart rate where the audio visual pacer resulted in a higher mean heart rate than the visual pacer The result should be approached with caution as the cause is unclear and should be investigated further In test 2 a significant result was found for the effect of breath pacer type on peak heart rate where the ping pong sound resulted in a lower peak heart rate than the waves and accordion Again the cause for the result is unclear and should be investigated further The waves sound was found to be significantly more pleasant to listen to than the other two sounds and the waves sound was found to be significantly easier to follow than the ping pong sound 13 Bibliography 1 2 3 4 5 6 7 8 9 10 Alan Watkins The electrical heart Energy in cardiac health
48. rform the test The test application was programmed using Mat lab R2014a The complete code listing is approximately 1000 lines and only snippets will be shown in the following sections For the complete code please refer to the file BreathPacerTest_1 mon the CD included with this report 4 1 Breath pacing cycle For this test it was decided to not include a precursory test to find each participant s resonant breathing frequency A fixed pacing cycle of 6 breaths per minute bpm was implemented for all modalities in the test As mentioned in the Analysis there seems to be slightly differing approaches to the breath pacing applied by various commercial products mainly whether the user is instructed to briefly hold his her breath between inhalation and exhalation For this test the choice was made to adopt the default cycle found in the CardioSense Trainer application by Complete Coherence A 10 second breathing cycle equalling 6 bpm has the following structure e 3 seconds inhalation e 1 second hold e 4 seconds exhalation e 2 seconds hold http www mathworks se products matlab 4 Design and Implementation Test 1 4 1 Breath pacing cycle 4 1 1 Visual breath pacing The visual breath pacer was made as a vertical progress bar centered on the screen see figure 4 1 The test participant would be instructed to inhale as the progress bar fills up exhale as the progress bar goes down and holding his breath while t
49. roduct In regards to the question about how easy or difficult it was to hear the sounds the experimenter asked one participant for reason why he chose to rate the PingPong sound lower than the other sounds His response was that since there was less sound overall because of the long silent gaps he considered the PingPong sound less audible This sort of reasoning may well have led to the slightly lower mean score for the PingPong sound For the question about how easy or difficult it was to remember how the sound should guide the breathing cycle one participant orally expressed being puzzled by the question as he remembered it had been stated in the instructions before the test that all breath pacers would begin with inhalation Having remembered this it seemed trivial to follow the pacer sounds However it was clear from conversations after tests that not all participants remembered this information from the initial instructions 11 3 General discussion A caveat in this study is that the pulse wave signals were not refined by applying parabolic interpolation before performing peak detection and calculating heart rate despite such an approach being recommended for the sampling rate adopted here see Measuring Heart Rate on page 8 The result could be a somewhat higher degree of noise in the data 4T 12 Conclusion Two tests have been performed to investigate if different types of breath pacing differ in effect on heart
50. s of the audiosample for ia l integerRepeats arrayIdx ia 1 xinfo TotalSamples 1 audioLooped arrayIdx info TotalSamplesxria 1 a end if remainderSamples audioLooped integerRepeatsxinfo TotalSamples 1 size audioLooped 1 1 a 1 remainderSamples 1 end player audioplayer audioLooped Fs 16 4 Design and Implementation Test 1 4 5 Pauses between breath pacing 4 3 Pauses between breath pacing To give participants a pause from breath pacing and return to a baseline HRV a one minute pause was given before starting each mode of breath pacing The screen image would display the text Relax and breathe normally Breath pacing will begin in a little while However five seconds before breath pacing would begin the text a little while would be replaced by a countdown showing the number of seconds before pacing began so the participant could prepare to focus on the upcoming stimulus esi facere Relax and breathe normally Breath pacing will begin in a little while Figure 4 5 The screen of the test application while in the pause state The last 5 seconds before the pause would end the text a little while would be replaced by a countdown 4 4 Demo mode and Test mode The test application was programmed so that it could also run a demonstration of the test cycle helping to inform participants of the procedure In the demo mode pauses were short
51. ss than 5 minutes the ULF band is not considered and VLF is defined as 0 04 Hz It is recommended that recordings have a duration of at least 10 times the period of the lowest frequency in the range being investigated 10 When performing spectral analysis of HRV it must be considered that although the initial ECG or PPG signal may have been sampled with a fixed sampling rate e g 100 Hz the time series of interbeat intervals IBI s extracted from the ECG or PPG samples will not be fixed interval data Spectral analysis via a fast Fourier transform FFT assumes evenly spaced samples and so applying FF T directly on the IBI time series will lead to incorrect estimations A resampling of the times series to obtain a new evenly spaced dataset can also lead to an incorrect estimation of the actual frequency spectrum For a more accurate spectral estimation of HRV the Lomb Scargle periodogram method is recommended as it does not assume evenly spaced samples and allows analysis directly on the IBI timeseries 14 15 3 2 2 Respiratory Sinus Arrhythmia Respiratory sinus arrhythmia RSA is a component of HRV seen as a variation in heart rate that coincides with a person s frequency of respiration When breathing air into the lungs the parasympathetic nervous system s influence on heart rate is reduced resulting in an increase of heart rate When breathing air out of the lungs the parasympathetic influence is reinstated and heart rat
52. st may differ from the order of the demo mode and that there would be no visual pacer in the actual test Also it was explained that each pacer audio would always begin with inhalation and that the speed of pacing was identical across pacers Although this test would focus solely on auditory cues for breath pacing it was found useful to include the visual progress bar when demonstrating to participants how the breathing cycle corresponded with the audio Thus the sequence presented in the demo mode was as follows Pause 7 sec Audio visual breath pacing Accordion 20 sec Pause 7 sec Audio breath pacing Accordion 20 sec Pause 7 sec Audio visual breath pacing Waves 20 sec Pause 7 sec Audio breath pacing Waves 20 sec 38 9 Testing Test 2 9 2 Test procedure 9 Pause 7 sec 10 Audio visual breath pacing PingPong 20 sec 11 Pause 7 sec 12 Audio breath pacing PingPong 20 sec In test mode no audio visual pacing was presented and the order of breath pacing audio was randomized The duration of each pause was 1 minute and the duration of each pacing mode was 5 minutes resulting in the following structure Pause 1 min Audio breath pacing 5 min Pause 1 min Audio breath pacing 5 min Pause 1 min Audio breath pacing 5 min From the participant arrived until she he had finished testing and responded to the questionnaire the duration was approximatel
53. ted peak frequency of 0 1 Hz for the other pacers The other participant showed a HRV peak frequency of 0 2 Hz for the Audio and Visual pacer but 0 1 Hz for the AudioVisual pacer For both participants in the above mentioned cases of outlying HRV peak frequencies a second but smaller peak was found at 0 1 Hz where the main peak would have been expected HRV peak frequency Frequency Hz Audio AudioVisual Visual Breath Pacer Figure 7 4 Two of the 15 participants were the source for almost all of the variance seen in HRV peak frequency If those two participants are excluded from the data then the mean HRV peak frequency for each pacer mode is located very closely to 0 1 Hz and standard deviation diminishes drastically Error bars represent 1 SD See figure 7 1 left for comparison If the two participants mentioned above are excluded from the dataset the mean HRV peak frequency still lies around 0 1 Hz but standard deviation is reduced greatly 31 7 Results Test 1 7 0 Audio M 0 100270 SD 0 000410 AudioVisual M 0 100356 SD 0 000314 Visual M 0 100245 SD 0 000324 See figure 7 4 32 8 Design and Implementation Test 2 The second test was designed to investigate whether a difference in HRV exists between using two types of audio for breath pacing continuous and discrete Also the Accor dion sound from test 1 was included for comparison with the two new audio pacers t
54. ted via the autonomic nervous system ANS where the sympathetic branch of the ANS is responsible for increasing heart rate and the parasympathetic branch via the vagus nerve is responsible for de creasing heart rate 11 13 The mechanisms by which the sympathetic and parasympathetic nervous systems influ ence heart rate differ both chemically and in terms of the time it takes for their effect to peak The effect coming from a regulation by the sympathetic branch peaks after approximately 4 seconds and returns to baseline level after approx 20 seconds From the parasympathetic branch the effect peaks after about half a second and goes back to baseline after approximately 1 second 12 Left atrium Right atrium Sinoatrial node pacemaker Atrioventricular node Septum Figure 3 1 Illustration of the heart showing the location of the sinoatrial node Image from 7 3 Analysis 3 2 Heart Rate Variability 3 2 1 Analysis of HRV The aforementioned difference in modulation speed is the basis for applying one of the common methods for statistical analysis of HRV power spectral density PSD which quantifies the amount of variance within a range of frequencies By convention the HRV power spectrum is divided into four frequency bands ultra low frequency ULF 0 003 Hz very low frequency VLF 0 003 0 04 Hz low frequency LF 0 04 0 15 Hz and high frequency HF 0 15 0 4 Hz For short term recordings le
55. the waves sound was found to be significantly easier Copies 3 to follow than the ping pong sound Pages 60 Finished August 7th 2014 Copyright O This report and or appended material may not be partly or completely published or copied without prior written approval from the authors Neither may the contents be used for commercial purposes without this written approval Breath Pacing by Auditory and Visual Cues Christian Toft August 7 2014 Contents 1 Introduction 2 Pre analysis 2 1 State ofthe ArtG aa dur 64 ee dira eee ad 2 1 4 Inner Balance by HeartMath llle 2 1 2 MyCalmbBeat 1e SEES E RUE ERES 2 1 3 CardioSense Trainer 22e 2 L4 PheiDash 333 ss a oor e REGE RR RR RR Ox 2 2 Discussions 2 42 RR al sale wwe RR RACER m m ENS 2 2 1 lt Problemstatements as 5 5 x keke Ree ee te 0e o SN 3 Analysis 3 1 Measuring Heart Rate 2s 3 1 1 Artefacts in ECG and PPG signals o 3 2 Heart Rate Variability 2 2222 32 41 Analysis o HRV agria isa gum Smash 3 2 2 Respiratory Sinus Arrhythmia e 3 2 3 Resonant Breathing Frequency e 3 9 Breath Pacing zoe oom rre eu deu e e bets auus b e esi ao 4 Design and Implementation Test 1 4 1 Breath pacing cycle iii o tt e e kt 4 1 1 Visual breath pacing 24 ow e a a a A ws RS RBS d 4 1 4 Auditory breath pacing een 4 1 8 Audio visual breath pacing en 4 2 Gapless audio playbac
56. ticipants when answering a questionnaire A laptop computer with an Intel Core i7 2 4 GHz processor was used to run Matlab and the test application The laptop display was facing away from the participant An Arduino Uno with an eHealth shield see Design and Implementation Test 1 on page 13 was connected to the laptop for data acquisition The respiration sensor was placed under the participant s nose held in place by a rubber band around the head The pulse sensor was typically placed on the participant s left or right index finger tip Both sensors were connected via the eHealth 5 Testing Test 1 5 3 Test procedure shield The pulse sensor s circuit board was hidden so LED blinks from heart beats would not be seen by the participant Figure 5 1 A test participant wearing the headphones and sensors The test application was run on the laptop and the screen was mirrored on a monitor facing the test participant Checking of sensor signals was done with the monitor turned off so only the experimenter would see the pulse wave and respiratory signals on the laptop display 5 3 Test procedure When a test participant arrived he she would be asked to set his her mobile phone to a silent and no vibration mode and also not have the mobile screen visible during the test This was to prevent any incoming calls or messages to affect the participant possibly raising his her pulse or drawing attention away from the task of paced bre
57. uary 2015 Images from http www bragi com press The Dash by Bragi is a Kickstarter project that was successfully funded in March 2014 It is being marketed as the World s First Wireless Smart In Ear Headphones The headphones integrate a microprocessor 4 GB memory ear bone microphone bluetooth connectivity touch sensitive interface and a number of sensors one of which is a pulse oximeter for tracking the user s pulse and blood oxygen levels An API application programming interface will be made available for developers to create new applications for the Dash Whereas other HRV related products require the user to attach a pulse sensor that otherwise serves no use the Dash seems appealing in that it has a multitude of uses while integrating a pulse sensor 2 2 Discussion A main component of the HRV biofeedback products reviewed above is breath pacing Typically a form of visual or auditory guide that the user is supposed to follow with https www kickstarter com projects hellobragi the dash wireless smart in ear headphones 2 Pre analysis 2 2 Discussion his her respiratory cycle An interesting aspect about the Dash in ear headphones is that they can function in combination with a smartphone s visual display and potentially as stand alone and audio only devices for breath pacing and HRV biofeedback if desired Some questions that arise here is does visual and auditory breath pacing yield the same res
58. ults in HRV What kind of audio would users prefer for breath pacing and will different kinds of auditory breath pacing result in differences in HRV For example the accordion like sound used in the CardioSense Trainer subsection 2 1 3 on page 5 is a continuous type of audio there are no silent gaps But would a discrete type of audio e g a short ping to indicate beginning inhalation and exhalation and no sound in between to indicate the level of progression yield the same results in terms of HRV and user preference Preliminary review on the subject of breath pacing in HRV biofeedback scenarios has indicated that the suggested respiratory cycle is far slower than what is common for healthy adult humans Thus continuous audio designed to indicate the level of progression may yield different results than discrete sounds indicating only beginning and end of respiratory stages 2 2 1 Problem statement In regards to the questions in the discussion above the following problem statements are formed e Is there a difference in HRV between using auditory audio visual and visual breath pacing e Is there a difference in HRV between using continuous and discrete audio for breath pacing 3 Analysis 3 1 Measuring Heart Rate Accurate detection of heart beats can be done with electrocardiography ECG which measures the electrical signal that occurs from contraction of the heart muscles at every heart beat This can be measured b
59. very short term 30 beat analysis Part 1 survey and toolbox In Journal of clinical monitoring and computing 27 5 2013 pp 569 576 cit on p 10 Evgeny G Vaschillo Bronya Vaschillo and Paul M Lehrer Characteristics of reso nance in heart rate variability stimulated by biofeedback In Applied Psychophys iology and Biofeedback 31 2 2006 pp 129 142 cit on pp 10 11 Paul M Lehrer Biofeedback training to increase heart rate variability In Prin ciples and practice of stress management 3 2007 pp 227 248 cit on p 11 Paul M Lehrer Evgeny Vaschillo and Bronya Vaschillo Resonant frequency biofeed back training to increase cardiac variability Rationale and manual for training In Applied psychophysiology and biofeedback 25 3 2000 pp 177 191 cit on p 11 Esko O Dijk and Alina Weffers Breathe with the Ocean a System for Relaxation using Audio Haptic and Visual Stimuli In EuroHaptics 2010 2010 p 47 cit on pp 12 57 50 14 Appendices 14 1 Appendix A The complete code for the Arduino Uno eHealth sensor platform for Arduino include lt eHealth h gt Initialize variables int airFlow 0 pulseWave 0 unsigned long time 0 timeStart 0 loopEntryTime 0 String preA String A String preB String B String preC String C String stringToSerial String char a b void setup Serial begin 115200 initialize ser
60. via a USB connection e Health Airflow sensor sensor shield thermistors circuitry Arduino Uno PC Data Acquisition Hardware Lo eee ee SNE Analog to Digital Matlab 7 Pulse sensor Pulse sensor conversion 27700 IR LED photodetector circuitry Figure 4 11 A functional block diagram of the pathways from the sensors to the PC Shttp www cooking hacks com 20 4 Design and Implementation Test 1 4 8 Monitoring of sensor signals The Arduino was programmed to send a certain character the letter a to the PC when serial communication was initialized and then wait for the same character to be returned from the PC This was done to make sure that serial communication had been properly established before proceeding to run the main function The main function consists of a loop where the Arduino samples the sensors and send the values to the PC every 10 milliseconds resulting in a sampling frequency of 100 Hz In this loop the serial connection is also checked to see if the character b has been sent from the PC in which case the Arduino will stop sending data Figure 4 12 The eHealth shield stacked on top of an Arduino Uno Image from http www cooking hacks com The sensor sample values were sent along with a timestamp given in milliseconds from the Arduino as a concatenated string where each value is prefixed by a single character either A B or
61. which the pacing modes would appear during the actual test may differ from the order of the demo mode and that it was therefore important that the participant payed attention to how the audio corresponded with this cycle Furthermore it was explained that each pacing mode would always begin with inhalation and that the speed of pacing was identical across pacers The sequence presented in the demo mode was as follows Pause 7 sec Visual breath pacing 20 sec Pause 7 sec Audio visual breath pacing Accordion sound 20 sec Pause 7 sec Audio breath pacing Accordion sound 20 sec In test mode the structure was the same except the order of breath pacing modes was randomized The duration of each pause was 1 minute and the duration of each pacing mode was 5 minutes giving the following structure Pause 1 min Breath pacing 5 min Pause 1 min Breath pacing 5 min Pause 1 min Breath pacing 5 min In total from the participant arrived until the participant had finished testing and re sponded to the questionnaire the duration was approximately 30 minutes 25 6 Processing of Sensor Data An additional Matlab application was programmed in order to plot and inspect the recorded sensor data in a manner that was more intuitive than by browsing through very long text files The application also served to perform pulse peak detection so a time series of interbeat interv
62. ww techsmith com camtasia html 3 Audacity is open source audio editing software See http audacity sourceforge net 14 4 Design and Implementation Test 1 4 1 Breath pacing cycle Figure 4 2 Waveform of the Accordion sound file with a length of 10 seconds From left to right the red arrows mark positions of 3 seconds 4 seconds and 8 seconds respectively For the phase of the test that applied the auditory breath pacing a speaker icon was displayed on screen as an indication of this being an audio only mode figure 4 3 n Figure 4 3 Only a speaker icon was displayed on screen while the audio only breath pacer was active 4 1 3 Audio visual breath pacing For the phase of the test that applied audio visual breath pacing the progress bar and the Accordion sound was presented simultaneously Figure 4 4 The screen of the test application while audio visual breath pacing was active To ensure that pacing cues from the progress bar and audio were synchronised the progress bar would update according to the progress of the audio implemented by the following Matlab code 9 currently played sample time in seconds avPacerTime player CurrentSample Fs modulus 10 gets our current location within the 10 sec pacer cycle modAvPacer mod avPacerTime 10 15 4 Design and Implementation Test 1 if within first 3 sec if modAvPacer lt 3 set bar h
63. y 35 minutes 39 10 Results Test 2 Originally 14 people participated in the test The test data from one participant was rejected due to what appeared to be motion artefacts in the pulse wave signal Feature extraction was done with the same method as for test 1 described in Processing of Sensor Data on page 26 The test data considered below is for 13 participants 8 male 5 female age 21 55 years mean 33 077 std 9 106 Five of these participants also appeared in test 1 with approximately three weeks time between participating in the two tests In the following mean and standard deviation will be abbreviated as M and SD Breath pacer sequence participants 1st 2nd 3rd 2 Accordion Waves PingPong 1 Accordion PingPong Waves 1 Waves Accordion PingPong 3 Waves PingPong Accordion 5 PingPong Accordion Waves 1 Visual Waves Accordion Table 10 1 The distribution of breath pacer sequences among the 13 participants Accordion Waves PingPong M SD M SD M SD HRV peak frequency Hz 0 100307 0 000256 0 100211 0 000397 0 100088 0 000473 HRV peak power s Hz 104 728094 18 990621 104 253327 18 141188 102 625702 16 844486 Mean heart rate bpm 18 179672 10 853227 77 942248 9 578062 77 051404 10 581616 Peak heart rate bpm 88 878352 10 729586 89 394995 10 027823 87 690187 10 511956 Trough heart rate bpm 68 19184
64. y placing electrodes on the skin near the heart for the most accurate reading 7 8 ECG is considered the gold standard for measuring heart rate in clinical research but another method photoplethysmography PPG is also widely used today 7 9 PPG involves using a light emitting diode to shine infrared light into the skin and a photodetector to sense the variations in how much of the light is reflected or transmitted These variations are caused by the volume of blood circulating through the blood vessels Typical placement of a PPG sensor is at a fingertip or an earlobe Studies have been made in regards to whether HRV analysis from PPG signals is compa rable with that from ECG signals It has been found that PPG can be considered J sufficiently accurate only for healthy and mostly younger subjects at rest 9 One of the sources for discrepancies between results obtained by PPG and ECG could be pulse transit time the time it takes from the actual heart beat until the blood pressure wave is seen in the blood vessels Pulse transit time depends on age blood pressure stiffness of the arteries and is also affected by respiratory activity 9 In terms of sampling frequency it is recommended to work with 250 500 Hz for accurate peak detection in an ECG signal although a sampling rate as low as 100 Hz may be satisfactory if parabolic interpolation is used to refine signal peaks 10 3 1 1 Artefacts in ECG and PPG signals

Breath Pacing by Auditory and Visual Cues

Contents

Download Pdf Manuals

Related Search

Related Contents