Grouping Similar Seismocardiographic Signals Using Respiratory Information

Grouping Similar Seismocardiographic Signals Using Respiratory Information Amirtaha Taebi, Student Member , IEEE , and Hansen A. Man sy Biomedical Acoustics Research Laboratory , University of Central Florida , Orlando , FL 32816 , U SA {taebi@knights. , hansen .man sy@ } ucf. edu Abstract — Seismocardiography (SCG) offers a potential non - invasive method for cardiac monitoring. Quantification of the effects of different physiological conditions on SCG can lead t o enhanced unde rstanding of SCG gene sis , and may explain how some cardiac pathologies may affect SCG morphology. In this study, the effect of the respiration on the SCG signal morphology is investigated. SCG, ECG, and respiratory flow rate signals were measured sim ul taneously i n 7 healthy subjects . Results showed that SCG events tended to have two slightly different morphologies. The respiratory flow rate and lung volume information were used to grou p the SCG event s in to inspiratory/expiratory groups or l ow /hig h lung - volume groups, respectivel y. Although respiratory flow information could separate similar SCG events into two differe nt groups, t he lung volume information provided bet ter grouping of similar SCG s . This suggests that variations in SCG morphology may be du e , at least in part, to changes in the intrat horacic press ure or heart location since those parameters correlates more with lung volume than respiratory flow. Categorizing SCG events into different groups containing similar events allows more accurate estimation of SCG features, and better signal characterization, and classification. Keywords— Seismocardiographic signal; respiration effect; cardio- respiratory ; lung volu me , intrathoracic pre ssure . I. I NTRODUCTION ardiovascular disease is a major cause of mortality in the United States as it accounts for 24.2% of total deaths [1]. Developing new technologies for cardiac monitoring and diagnosis can help improve patient management and reduce mortality. Hence, analysis of bloo d flow dynamics [2], [3] a nd the heart related signals [4] has b eco me an active area of research. Seismocardiographic signals (SCG) are the mechanical vibratio ns measure d noninva sively at the chest surface [5], [6] . SCG are believed to be caused by th e mechanical processes associated with the hear t activity (such as cardiac muscle contract ion, blo od mom entum change s, valve closure, etc.) [7] – [9] . SCG signals contain i nforma tion relati ng to both ca rdiova scular and respiratory systems [10] that m ight be com plem entary to other hea rt moni toring methods s uch as e lectrocar diogra phy and pho nocardi ogr aphy. T he SCG signals mainly contain low - frequency waves where the human auditory sensitivity is low and cannot sufficient ly extract the signal characteristics accur ately [11], [1 2] . Hen ce measurement and analysis of t hese signals may b e done using compute r ized data acquisition and analysis , which woul d provide enhanc ed qualitative and quantitative d escription of the signal characteristics in both time and frequency domains [13] – [15] . SCG wa s previously used to estimate the respiration r ate, which was found compa rable to th at derived from a reference respiration b elt [10] . SCG sign al morphology was r eported to vary with differen t factors, including res piration cycle (inspira tion vs expi ration ), sensor l ocation on the chest, etc. [16], [17]. T he effect of respiratory cycle has been studied [10] on some of the SCG features such as timing interval changes . Howev er, the ef fect of respirat ion on t he SCG si gnal morphol ogy needs m ore atte ntion [18] . During inspiration, the diaphra gm moves downwar d, the c hest wa ll expand s, the intrathoracic pressure decreases, the lung s inflate [19] , and the heart positon is displaced almost linearly with th e diaphragm [20] . Th e decrease d intrathoracic pressure increases the pulmonary blood volum e , leading to an increase and decrease in the right and left a trial filling, and reduction in the left ventric ular stroke volum e [21] . These hemodynam ic change s can affect the SCG signal morphology. As describe d, the SCG variation du ring a re spiration cy cle h as been mentioned before. T his study , howe ver, aims at investiga ting the possibl e physiological co rrelates of this morphologi cal variat ion. For this purpose , the SCG events in a recording were first g rouped based on the criteria that are physiologi cally measureable (e .g. inhalati on and exhalation), and then the be st crite rion that could gr oup simil ar SCG events together was id entified. Tha t criterion was then s tudied to explain why SCG morphology varies duri ng the respirat ion cycle. Achievements of thi s study include quan tification of th e differences in SCG signals due to respirati on, and determi nation of optim al respirat ion criter ion for groupi ng the different SCG wavef orms. Materia ls and methods are given in section II . Results are pres ented and discussed in section s III and IV , respect ively , followed by conc lusions i n section V . II. M ETHODO LOGY A. Pa rticipants The study pr otocol was appr oved by the i nstitutiona l review boar d of the Uni versity of Central Fl orida, Orla ndo, FL. A total of 7 young ind ividuals wi th no history of cardiovascular diseas e participated in the study after informed consent . Mean age, height, weight , body mass index ( BMI ), and heartbeat o f the subjects were obtained and r eported in Table I . C TABL E I. O VERVIEW OF THE SUBJECTS ’ CHARACTERISTICS ( MEAN ± SD). Age (years) 24. 3 ± 5.0 Height (cm) 17 0.8 ± 8. 2 Weight (kg) 78 .7 ± 1 3.0 Heart rate (bpm) 66.6 ± 9.0 BMI (kg/m 2 ) 26. 9 ± 3. 4 Number of subjects 7 B. Data Collect ion All participants were instru cted to lay supine on a tab le and breathe normally. The SCG signal was m easured using a triaxial accelerometer ( Model: 35 6A32, PCB P iezo tronics, Depew, NY). The accelerometer output was ampl ified using a s ignal c onditioner (Mo de l: 482C , PCB P iezotronics, Depew, NY) with a gain factor of 10 0. The sensor was placed at th e left ste rnal border and the 4 th intercosta l space us ing a doubl e - sided ta pe s ince this location tend ed to have high si gnal - to - noise ratio . The accelerometer’s x - and y - axes were aligned parallel to the anteropos terior and mediolateral direction s, respectively, wh ile the z - axis was aligned in dor so - ventral direction . I n this study , the z- component of acceleration tended to be stro nge st , simi lar to previous studi es [18] . Therefor e, atte ntion in t he current study was focus ed on t he analysis of this acceleration component. The respirat ory flow rate of the subjects was measured using a pre - calibrated spirometer (Mo del: A - FH - 300, iWorx Syst ems, Inc., Do ver, NH ) . The flow rate signal h ad positive and negative amplitu de during the inspira tion and e xpirati on, respe ctively . The voltage signal f or both respiratory flow rate and SCG signals were acquired using a Contro l Module ( Model: IX - TA - 220 , iWorx Syst ems, Inc., Dover, NH) . The lung volume was calculated as the integral of the respiratory flow rate. The sensors location s ar e shown in Fi g. 1.a. The SCG, ECG, a nd respirat ory signals were all measured simult aneously at a sam pling freque ncy of 1 0 kHz an d down - sampl ed to 320 Hz. A 5 s of sim ultaneousl y recor ded signal s are shown in Fig. 1.b. The SCG signals were then filter ed using a low - pass filter with a cut - off of 100 H z to remove the respiratory noise , which mostly has energy above t his cut - off frequency [22] . Matlab (R2015b, The MathWor ks, Inc, Natick, M A) was us ed to pr ocess a ll signals. C. S CG S egment ation and G rouping based on R espiration The SCG events in ea ch signal were found using a matched filtering with a template c onsisting of a previously identified SCG . The filtering algorithm wa s obtaine d from [23] . T he matched filter coefficients,  (  ) , were calculated as  (  ) =  (    + 1) (1) where  (  ) ,  , an d  = 1,2, … ,  were the lib rary template (an SCG event manually c hosen by t he user), num ber of sam ple points i n the tem plate, and t he coef ficient i ndex, res pectively. The filter ou tput ,  (  ) , was then calculated as  (  ) =  (  )   (  ) (2) where  (  ) wa s the raw SCG signal. T he filter output had maximums at locations that the raw SCG signal best matched the template. The envel ope of the filter out put was f ou nd using Hi lbert tra nsform . The pea ks of this envelope signal with an a mplit ude above a certain t hreshol d were then identified. The indices of t he peaks were then used to determ ine the loc ation of the SCG event s. Identi fied SCG events were checked manually to confirm the absence of distorte d SCG (for e xampl e due to m otion arti facts) . SCG events we re then divide d into two gro ups using tw o differe nt (a) (b) Fig. 1. (a) The location of the accelerometer, ECG electrodes and spir ometer on the subject body. The accelerometer and spiromet er se nsors wer e used to measure the SCG and respiratory f low rate signals , respectively . The dashed and dash - dot lines show the 4 th intercostal sp ace and sterna l border, resp ectively. (b) A 5 s portion of simultaneously acquire d SCG, ECG, and respiratory flow rate signals. (c) Summ ary of the signal processing algorithm used in this study. (c) Fig. 2 . (a) Ensem ble averaged ECG in the top panel and ensem ble averaged SCG in the bottom panel during the high lung volum e, (b) Ensemble averaged ECG in the top panel a nd ensemble averaged SCG in the bottom panel during the low lung volume. respiratory criteria. F irst, the respiratory flow rate was used t o group the SC G events into inspira tory and e xpirato ry groups corresponding to positive and negative respi ratory flo w s, respectively. Similar ly, the lung volum e was used to g roup SCG events. Here the average lung volume was first calculated , and l ow and hi gh lung volume s (i.e., L L V and HL V) were define d as those below a nd above t he me an lung volum e. T he SCG ev ents were then labeled as LL V and HL V events, depending on if they occu r re d during L L V or HL V, respectively . The two gro uping m ethods wer e then com pared to determin e which criterion is m ore effective in group ing similar SCG events. The details of quantify ing the SC G event similarity and effectiveness of the grouping cr iteri a are described in the next section. D. Group ing Criteria Effectiveness After separat ing the SCG events int o two groups (e.g., inspirat ory and e xpirato ry) , they were aligned in time (by minimizing the cross - corr elation function ), and an ensembl e average SCG was calculated for each group separately. Fig . 2 shows the ensem ble avera ge of SCG e vents d uring LLV a nd HLV. To qu antify the dis similarity o f each SCG with respect to the two group s , the d ifference between each SC G waveform and the average waveform of both groups was calculated . T hen the RM S (root - mean - s quare) of th e se difference s w ere determ ined (Eq. 3 ). T his quant ity was t hen divided by the RMS am plitude of the averag e waveform for each gr oup (Eq. 4 ).    ,  =  (    ,     ,  ) (3)    ,  = 󰈅    ,     ,  󰈅 × 100 (4) where   [1, … ,        ] , j is the group (i.e., inspiration or expiration) , an d    ,  =  (   ,  ) (5) wher e   ,  is the ensemble av erage d SCG event of group j . The avera ge dissim ilarit y of grouped SC G event s was calculated as,          =     ,   (6) This is calculated w ithin the sa me g roup as well as with respect to the alternate group. For exa mple, f or events that were grou ped as inspi ratory , their average dissimilarity was calculated with r espect to inspiratory (i.e., sam e group) and expirat ory (i.e., alternativ e group) , separately. The d ifference between t he se two average dissimilarities is indicative of how well was th e groupin g and can be calc ulated from ,    = 󰇡          ,            ,  󰇢 /          ,  (7.a) w here RD FR is the normalized difference of mean dissimilarity of inspira tory event s with res pect to inspirat ory and ex piratory groups, respectively. The same dissimilarity difference was cal culated f or expi ratory e vents. Another groupi ng choice for SCG events that was tested in the current s tudy wa s based on low and high lun g volum e, LLV and HLV, respe ctively. Here , dissimilarit ies were also calculated to determine the dissimilarity of each S CG event group with respect to its ow n group and the al ternative group . For exam ple , for LLV SCG events, the difference between average dissimilarities relative to LLV and HL V groups was calculated from ,    = 󰇡          ,            ,  󰇢 /          ,  (7 .b) To determ ine whic h group ing criteri on ( i.e., inspiratio n vs expirat ion or LLV vs HL V) provide bet ter groupi ng of SCG events , the difference in the average d issimilarity was compared. For example, the RD FR and RD LV wer e compared for each subject. This will help determine whether th e respiratory flow rate or lung vol ume more effec tively separate SCG ev ents . III. R ESULTS The me an dissimilar ity of inspir atory events with respect to same or alternat ive group (i.e . , inspiratory and expirat ory groups, res pectivel y) are listed in Table II (Colum n 2 and 3 , respectively) . Column 4 sho ws the num ber of events i n the group. The same information is listed fo r the expiratory gro up in column s 5, 6, and 7 , resp ectively. The difference in dissimilarity b etween altern ate and same group is listed in column s 8 and 9, respecti vely, where posi tive values indicate more dissimila rity with the alt ernate group compared to th e same group. The differenc e was positi ve in 6 out of 7 s ubjects. Hence it can be concluded that in most subjects, the mean dissimilarity w ithin the s ame group was sm aller tha n that for the alternative group , indicating pro per g rouping . The fact that two differe nt morphologi es of SCG can be separated ba sed on TABLE II . RMS BETWEEN SCG EVENTS IN TH E INSPIRATORY AND EXPIRAT ORY GROUPS AND THE E NSEMBLE AVERAGED INSPIRAT ORY / EXPIRATORY SCG EVENT (E QUATION 6 ). T HE VALUES ARE SHOWN AS MEAN ± SD. T HE NUMBER OF SCG EVENTS IN EACH GROUP IS SHOWN IN P ARENTHE SIS . T HE LAST COLUMN SHOWS THE RELATIVE DI FFERENCES IN PERCE NTAGE . Subject # RMS between inspiratory events and … RMS between expiratory ev ents and … Relati ve Difference (%) Averaged inspiratory SCG           ,   ±SD Averaged expiratory SCG           ,  ±SD Averaged inspiratory SCG           ,  ±SD Averaged expiratory SCG           ,  ±SD       1 25.0252 ± 6.0923 33.2976 ± 7.3152 (53) 28.4763 ± 7.5722 24.0721 ± 7.2852 (55) 33.06 18.29 2 42.1056 ± 12.382 8 46.6644 ± 12.047 5 (38) 50.7218 ± 11.969 8 47.8863 ± 13.232 6 (54) 10.83 5.92 3 47.7511 ± 13.471 2 61.4332 ± 15.162 0 (37) 56.7565 ± 20.021 3 49.7977 ± 21.262 3 (35) 28.65 13.97 4 45.8798 ± 13.6270 64.3130 ± 18.742 9 (28) 65.5481 ± 14.501 3 43.6113 ± 15.797 0 (55) 40.18 50.30 5 33.5099 ± 9.16038 36.2629 ± 7.9248 (46) 36.3790 ± 10.163 3 31.8645 ± 8.5258 (28) 8.21 14.17 6 32.6761 ± 2.1323 45.2334 ± 8.3686 (26) 48.2500 ± 17.037 0 48.9503 ± 11.366 5 (53) 38.43 - 1.43 7 33.2406 ± 8.8591 44.2626 ± 7.9633 (36) 40.0089 ± 8.6755 33.1932 ± 8.1408 (43) 33.16 20.53 TABLE III . RMS BETWEEN SCG EVENTS IN THE HLV AND LL V GROUPS AND TH E ENSEMBLE AVERAGE D HLV/LL V SCG EVENT (E QU ATION 6 ). T HE VALUES ARE SHOWN AS MEAN ± SD. T HE NUMBER OF S CG EVENTS IN EACH GROUP IS SH OWN IN PARENTHESIS . T HE LA ST COLUMN SHOWS THE RELAT IVE DIFFERENCES IN PERCENTAGE . Subject # RMS between L L V events and … RMS between H L V events and … Relative Difference (%) Averaged L LV SCG           ,  ±SD Averaged HL V SCG           ,  ±SD n Averaged LL V SCG           ,  ±SD Averaged HL V SCG           ,  ±SD n       1 22.4070 ± 5.6409 34.1765 ± 9.4193 (53) 31.4550 ± 5.4819 24.9368 ± 6.8360 (58) 52.52 26.14 + 2 46.1731 ± 11.318 9 49.5236 ± 14.710 5 (46) 63.2857 ± 11.203 3 43.2033 ± 13.234 8 (42) 7.26 46.48 + 3 47.5796 ± 12.750 7 87.6964 ± 13.840 9 (32) 86.4990 ± 18.515 4 51.9888 ± 20.064 2 (39) 84.31 66.38 + 4 44.0150 ± 11.353 1 77.7045 ± 21.245 4 (44) 69.7121 ± 9.4823 34.0905 ± 12.619 7 (37) 76.54 104.49 + 5 30.4948 ± 8.1303 44.8703 ± 7.6175 (44) 59.2533 ± 10.821 1 25.3256 ± 7.1247 (31) 47.14 133.97 + 6 33.1170 ± 10.145 1 63.0525 ± 13.669 0 (51) 69.8478 ± 18.929 8 37.1869 ± 10.401 2 (31) 90.39 87.83 + 7 26.0518 ± 6.7255 60.4353 ± 13.994 4 (40) 43.9254 ± 8.3870 34.5265 ± 7.9952 (36) 131.98 27.22 + The “+” sign in t he right column indicates that the dissimilarity results improved when lung volume was used instead of respiratory flow rate. Fig. 3. Rel ative diff erences calcula ted from Eq . 7.a and 7.b for inspi ratory, e xpiratory, LLV , and HLV S CG events. respirati on is consist ent with repo rts that t he SCG morphol ogy changes wi th different phases of res piration [17] . For one subject (subj ect #6), t he expirato ry events were mor e dis simila r to their g roup than the ins piratory group. For this subjec t, the two mean dissimilarities wer e close (~ 48.5) . Results of grouping SC G based o n lung volum e (i.e. LL V vs HLV) are prese nted in Table III, where the format is parallel to Table II. The SCG ev ents in ea ch group (LLV and H L V ) were more similar t o their own gro up. The RD FR and RD LV values ( listed in the last column of TABLE II and TABLE III , respectiv ely ) are a measure of h ow well SCG grou ping wa s . These r esults show that the lung volume s ignal was m ore successful t han the flow rat e signal i n grouping the SCG events i nto two differe nt groups (F ig. 3) where the eve nts in each grou p are morphol ogically sim ilar to each other. IV. D ISCUSSIONS A. Intrathoracic P ressure and H eart Displaceme nt The chest wa ll expa nsion and downward m oveme nt of the diaphra gm during inspi ration ca uses a mo re negative intrathoracic pressure and a d ownward m ovem ent of the heart . The negative pressure in creas es th e expansi on of the right atrium, right ventr icle an d th oracic supe rior a nd inferior vena cava , wh ich causes the i ntravascul ar and int ra - card iac pressures to fall . As a result, the t ransmural pressure (the difference between pressure inside the h eart chamber and th e intrathoracic pressure) increases . This causes a rise in cardiac chambe r expansion , preload and st roke vol ume th rough the Fran k - Starling mechanism . T he opposite phenomena h appens during e xpirat ion [21] . It can then be concluded that intrathorac ic pr essure variations due to respiratio n changes the heart chamber pressures , preload, stroke volum e and st roke work [24] . The se mechanical changes are expected to affect the heart muscle contractile movements and blood f low momentum which can manifest themselves as variations in SCG signal morpholog ies. B. Lim itations The primary limitation of th is study wa s the small number of subjects that particip ated. Future studies need to enro ll larger n umber of su bjects fr om a dive rse popul ation inc luding different age, gender, weight, race, and clinical status. V. C ONCLUSIONS The result s of this st udy showed t hat the SC G demonstr ated morphologi cal differences during respiration. SCG even ts were grouped acc ording to thei r waveform morphology . Two grouping cri teria we re implem ented. One gro uping relied on inspiratory vs. expiratory flow whi le the othe r relied on L LV vs HLV (w hich corresp onds to h igh an d low in trathoracic pressure ). The second criteri on resulted in more similarity within the SCG groups, suggest ing that int rathoracic pressure variat ions can lead t o detectabl e SCG morphology changes. Studying the effect of respi ration al lows separating S CG into groups wit h simila r events. Thi s reduces SC G waveform variability and enables more precise estimatio n of SCG characteristics. In addition, be cause respira tion triggers known changes in p hysiologica l paramet ers (such as int rathoraci c pressure, s troke volume, etc.) , it allows studying the effects of these param eters on SCG. S uch investiga tions can hel p enhance our understan ding of SCG ge nesis, an d explain SCG changes wit h cardiac pat hology. Future studies may pe rform compari son s between the spectral characteristics of two groups of SCG ( e.g., LLV v s HL V) as this might reveal fu rther useful SCG characteristics and may cont ribute to fu rther eluci date SCG genesis . In addition, a rtificial intellige nce methods suc h as neural networks or suppo rt vector m achines might be used to classify the SCG even ts into two grou ps . An ongoing study [25] aims at deve loping cla ssificati on algorithm s for this purpose. A CKNOWLEDGMEN TS This study was supporte d by NIH R44HL099053. R EFERENCES [1] S. L. Murphy, J. Xu, an d K. D. Kochanek , “Deaths: final data fo r 2010.,” Natl. Vital Stat. Rep. , vol. 61, no. 4, pp. 1 – 117, 2013. [2] F. Khalili and H. A. Mansy , “Blood Flow through a Dysfunctional Mechanical Heart Valve,” in 38th Annu Int Conf IEEE Eng Med Biol Soc , 2016. [3] F. Khalili, P. P. T. Gamage, and H. A. Mansy, “Hemodyn amics of a Bileaflet Mechanical Heart Valve with Dif ferent Levels of Dysfun ction,” J Appl Biotechn ol Bioeng , vol. 2, no. 5, p. 00044, 2017. [4] A. Taebi and H. A. Mansy, “Time - frequency Analysis of Vibrocardiographic Signals,” in 2015 BMES Annual Meeting , 2015. [5] A. Taebi and H. A. Mansy, “ Time - frequency Description of Vibrocar diographic Signals,” in 38th Annual Internat ional Conference of the IEEE En gineering in Medicine an d Biology Society , 2016. [6] A. Taebi and H. A. Mansy, “Time - Frequency Distrib ution of Seismocardiographic Sign als: A Comparative Study,” Bioengineering , vol. 4, no. 2, p. 32, 2017. [7] J. Zanetti and D. Salerno, “Seismocardiograph y: A new technique for recording cardiac vibration s. Concept, method, and initial observations,” J. Cardio vasc. Technol. , vol. 9, no. 2, pp. 111 – 120, 1990. [8] R. S. Crow, P. Hann an , D. Jacobs, L. Hedquist, and D. M. Salerno, “Relationship between Seismocardio gram and Echocardiogram for Events in the Cardiac Cycle,” Am. J. Noninvasive Cardiol. , v ol. 8, no. 1, pp. 39 – 46, 1994. [9] A. Taebi, R. H. Sandler, B. Kakavan d, and H. A. Mans y, “Seismocardiographic Signal Ti ming with Myocardial Strain,” in Signal Processi ng in Medicine and B iology Symposium ( SPMB), 2017 IEEE , 2017, pp. 1 – 2. [10] K. Pandia, O. T. Inan, G. T. a Kovacs, and L. Giovangrandi, “Extracting respiratory infor mation from seismocardiogram signals acquired on the chest us ing a miniature accelerometer.,” Phys iol. Meas. , vol. 33, no. 10, pp. 1643 – 60, 2012. [11] L. P. Feigen, “Physical characteristic s of sound and hearing,” Am . J. Cardiol. , vol. 28, no. 2, pp. 130 – 133, 1971. [12] A. A. Luisad a, The sounds of the norm al heart . Warren H. Green , 1972. [13] A. Taebi and H. A. Mansy, “Eff ect of Noise on Time - frequency Analysis of Vibrocardiog raphic Signals,” J. Bioeng. Biomed. Sci. , vol. 6(202), p. 2, 2016. [14] A. Taebi and H. A. Mansy, “Noise Cancellation fro m Vibrocardiographic Signals B ased on the Ensemble E mpirical Mode Decomposition,” J. Biotechnol. Bioeng. , vol. 2, no. 2, p. 00024, 2017. [15] A. Taebi and H. A. Mansy, “ Analysis of Seismocardiograph ic Signals Using Polyno mial Chirplet T ransform and Smoothed Pseudo Wigner - Ville Distribution,” in Signal Proces sing in Medicine and Bi ology Symposium (SP MB), 2017 IEEE , 2017, pp. 1– 6. [16] W. Lin, W. Chou , P. Chang, C . Chou, M. W en, M. Ho, a nd M. Lee, “Identification of Location Specific Feature Poin ts in a Cardiac Cycle Using a Novel Seismocardiogr am Spectrum System,” IEEE J. Biomed. Heal. Informa tics , 2016. [17] K. Tavakolian, “Charachterization and analysis of seismocardiogram for estimation of hemodynamic parameters,” Simon Fraser University, 2010. [18] O. T. Inan, P. F. Migeott e, K. S. Park, M. Ete madi, K. Tavako lian, R. Casanella, J. Zanetti, J. Tank, I. F untova, G. K. Prisk, and M. Di Rienzo, “Ballistocardiography and Seismocardiography: A Review of Recent Advances,” IEEE J. Bio m ed. Heal. Informatics , vol. 19, no. 4, pp. 1414 – 1427, 2015. [19] R. E. Kl abunde, “Cardi ovascular Phys iology Concepts (2nd Ed.),” Lippincott Williams & Wilkins , p. 256, 2012. [20] Y. Wang, S. J. Ried erer, and R. L. Eh man, “Respira tory Motion o f the Heart: Kinematics and the Implications f or the Spatial Resolution in C oronary Im aging,” Magn. Re son. Med. , vol. 33, no. 5, pp. 713 – 719, 1995. [21] R. E. Kl abunde, “Cardi ovascular Phys iology Concepts (2nd Ed.),” Lippincott Williams & Wilkins , p. 256, 2012. [22] H. A. Mansy, T. J. Royston, R. A. Balk, and R. H. Sandler, “Pneumothorax detection using computerised analysis of breath sounds.,” Med. Biol. Eng. Comput. , vol . 40, no. 5, p p. 526 – 532, 2002. [23] H. A. Mansy and R. H. Sand ler, “Bowel - sound signal enhancem ent using adaptive filtering,” I EEE Eng. Med . Biol. Mag. , vol. 16, no. 6, pp. 105 – 117, 1997 . [24] J. V. Tyberg, D. A. Gra nt, I. King ma, T. D. Moore, Y. H. Sun, E. R. Smith, and I. Belenkie, “Effects of positive intrathoracic press ure on pulmonary a nd systemic hemodynamics,” in Resp iration Physiology , 2000, vol. 119, no. 2 – 3, pp. 163 – 171. [25] B. E. Solar, A. Taebi, and H. A. M ansy, “Classification of Seismocardiographic Cycles into Lu ng Volume Phases,” in Signal Processing in Medicine and Biology Symposium (SPMB), 2017 IEEE , 2017, pp. 1 – 2.