Pathologies affect the performance of ECG signals compression

1 Vol.:(0123456789) Scien i ic Repo s | (2021) 11:10514 | h ps://doi.o g/10.1038/s41598-021-89817-w www.na u e.com/scien i ic epo s Pa hologies a ec he pe o mance o ECG signals comp ession And ea Nemco a1*, Rado an Smisek1,2, Ma in Vi ek1 & Ma ie No ako a3,4 The pe o mance o ECG signals comp ession is in luenced by many hings. Howe e , he e is no a single s udy p ima ily ocused on he possible e ec s o ECG pa hologies on he pe o mance o comp ession algo i hms. This s udy e alua es whe he he pa hologies p esen in ECG signals a ec he e iciency and quali y o comp ession. Single-cycle ac al-based comp ession algo i hm and comp ession algo i hm based on combina ion o wa ele ans o m and se pa i ioning in hie a chical ees a e used o comp ess 125 15-leads ECG signals om CSE da abase. Rhy hm and mo phology o hese signals a e newly anno a ed as physiological o pa hological. The comp ession pe o mance esul s a e s a is ically e alua ed. Using bo h comp ession algo i hms, physiological signals a e comp essed wi h be e quali y han pa hological signals acco ding o 8 and 9 ou o 12 quali y me ics, espec i ely. Mo eo e , i was s a is ically p o en ha pa hological signals we e comp essed wi h lowe e iciency han physiological signals. Signals wi h physiological hy hm and physiological mo phology we e comp essed wi h he bes quali y. The wo s esul s epo ed he g oup o signals wi h pa hological hy hm and pa hological mo phology. This s udy is he i s one which deals wi h e ec s o ECG pa hologies on he pe o mance o comp ession algo i hms. Signal-by-signal hy hm and mo phology anno a ions (physiological/pa hological) o he CSE da abase a e newly published. Elec oca diog am (ECG) is he mos equen ly used echnique o e eal and diagnose hea diseases1. In clinical p ac ice, s anda d 12-lead ECG is used p edominan ly2. Some diso de s (especially a hy hmias) a e pa oxys- mal and appea only ime o ime and/o du ing speci ic ac i i y3. In such cases, he Hol e ECG is indica ed4. Hol e signals a e eco ded o a leas 24 h using mobile ECG de ice4. The ob ained signal is s o ed o line o online ansmi ed o he medical cen e and e alua ed by ca diologis and/o Hol e echnician. To speed up he ansmission5, sa e he memo y5 and ene gy o he de ice6, he ECG signal is comp essed. The aim o comp ession is o each maximum e iciency o da a educ ion wi hou loss o diagnos ic in o ma ion5. Comp ession algo i hm may be accep able only i he diagnos ic in o ma ion (ECG mo phology) is nei he los no dis o ed7. To gain signi ican da a educ ion and minimize powe consump ion in eleheal h moni o ing, lossy comp ession is p e e ed8. Howe e , lossy comp ession is always connec ed wi h in o ma ion loss. As a ma e o p inciple, lossy comp ession is always comp omise be ween size o he da a and hei quali y9. The e o e, he assessmen o ECG signal quali y a e comp ession and he de e mina ion o comp ession e i- ciency should be an essen ial pa o comp ession i sel 10. In e ms o comp ession algo i hms’ app oach o pa hologies in ECG signal, wo g oups can be dis inguished. Fi s , algo i hms which do no ake in o conside a ion pa hologies in ECG signal, such as Se Pa i ioning in Hie a chical T ees (SPIHT)11,12, comp essed sensing based me hod13 o skele on (local ex eme ex ac ion) based me hod14. Second, comp ession algo i hms which ake in o conside a ion pa hologies in ECG signal (such as15–18). The p inciples o hese algo i hms a e usually based on de ec ion o QRS complexes o bea s classi ica- ion. A ew examples o comp ession algo i hms which ake in o conside a ion pa hologies a e desc ibed in he nex pa ag aphs in mo e de ail. Comp ession me hods based on QRS complex de ec ion exhibi lowe pe o mance in case o abno mal ECG signals19. This p oblem mainly conce ns 2D comp ession me hods in which he ECG signal is always segmen ed in o bea s. The changes in bea s’ pe iodici y lowe he pe o mance o 2D comp ession algo i hms7. The pe o mance o 2D comp ession me hods is dependen on he accu acy o QRS complex de ec ion which may be a ec ed by noise, a i ac s, sudden changes in ampli udes, RR in e als o QRS complex mo phology7. Chen16 de eloped an algo i hm based on empla e ma ching and s a ed ha i was no e y sui able o i egula OPEN 1Depa men o Biomedical Enginee ing, Facul y o Elec ical Enginee ing and Communica ion, B no Uni e si y o Technology, Technická 12, 616 00 B no, Czech Republic. 2Ins i u e o Scien i ic Ins umen s, The Czech Academy o Sciences, K álo opolská 147, 612 64 B no, Czech Republic. 3Depa men o Physiology, Facul y o Medicine, Masa yk Uni e si y, Kamenice 753/5, 625 00 B no, Czech Republic. 4In e na ional Clinical Resea ch Cen e , S . Anne’s Uni e si y Hospi al B no, Pekařská 53, 656 91 B no, Czech Republic. *email: nemco aa@ u b .cz 2 Vol:.(1234567890) Scien i ic Repo s | (2021) 11:10514 | h ps://doi.o g/10.1038/s41598-021-89817-w www.na u e.com/scien i ic epo s/ wa e o ms, including a ying pa e ns. Chou17 de eloped a 2D ECG p ep ocessing algo i hm o be e com- p ession o i egula ECG signals. Be a e al.18 used hyb id comp ession o ECG signal. I s a ed wi h suppo ec o machine (SVM)-based bina y classi ie o ECG bea s o no mal and abno mal and con inued wi h wa ele -based comp ession o abno - mal bea s and combined wa ele - and PCA-based comp ession o no mal bea s. Rakshi e al.15 used di e en app oach. They p ecede econs uc ion e o s by using h ee dic iona ies con- side ing no mal, p ema u e en icula con ac ion and paced bea s o eco e he signal in comp essed sensing (CS) app oach (i is one speci ic algo i hm based on CS; no all CS-based comp ession algo i hms ake in o con- side a ion pa hologies). The ECG mo phology di e s in hese h ee ypes o bea s and due o he ailo ed econ- s uc ion can be p ese ed. This algo i hm was es ed on a pa o MIT-BIH A hy hmia Da abase, No mal Sinus Rhy hm Da abase and Comp ession Tes Da abase. The au ho s s a ed ha hei me hod ou pe o med exis ing me hods (wa ele dic iona y, adap i e dic iona y, s anda d dic iona y based CS app oaches and wa ele -based lossy comp ession scheme). Howe e , his algo i hm may ha e p oblems wi h signals including o he pa hologies. Nasimi e al.20 in oduced comp ession scheme which disc imina es be ween no mal and abno mal hea - bea s. In case o no mal hea bea s, he edundancy is emo ed, which leads o inc eased spa si y o he signal and such pa is be e comp essible using CS. Dissimila hea bea s can be caused by pa hology and such pa s a e no comp essed. In bo h cases, he quan iza ion and Hu man encoding a e applied. Comp ession pe o mance o ECG signal may be dec eased by noise, unless i is il e ed ei he be o e o wi hin he comp ession p ocess. Noisy ECG signals a e usually s o ed using mo e bi s7,21. Bu noise is no diagnos ically aluable; hus i is ad an ageous no o s o e i a all (in ideal si ua ion). PhysioNe con ains a da abase o ECG signals dedica ed o es ing o comp ession algo i hms— he MIT-BIH ECG Comp ession Tes Da abase22,23. I consis s o 168 wo-channel ECG signals which include wide a ie y o pa hologies—a hy hmias (a ial, AV junc ional, en icula ), dis u bances in conduc ion and noise23. I can be used o es ing o comp ession abili ies o a ious algo i hms, pa icula ly how hey can comp ess di e - en ypes o ECG signals and p ese e he diagnos ic in o ma ion. Ne e heless, his da abase is used a he spo adically5,15,21,24. The main eason is he ac ha i is no anno a ed25. To he bes o ou knowledge, he e is no a single s udy p ima ily ocused on he possible e ec s o ECG pa hologies on he pe o mance o comp ession algo i hms. Manikandan e al.7 epo ed ha he p esence o bo h egula and i egula hy hm wi h di e en mo phologies in he signals may lead o di e en comp ession a ios (CRs) o a gi en pe cen age oo mean squa e di e ence (PRD) in comp ession me hods. Se e al quali y-gua - an eed me hods a e a ailable, e.g.26–28, in which he quali y me ic’s h eshold can be se o comp ess he signal wi hou loss o diagnos ic quali y. Howe e , his h eshold can be se empi ically o acco ding o o he au ho ’s ecommenda ions which a e usually de e mined on he whole da abase. In p e ious s udy10, we ecommended a ious quali y me ics’ h esholds. Howe e , i eal gua an ee o he diagnos ic quali y o he signal is needed, he quali y me ics’ alues a e e y s ic in case he whole da abase is conside ed. As a esul , he e iciency o comp ession will be low. Po en ially, he signals wi h any pa hology need o be comp essed mo e ca e ully wi h lowe e iciency o each desi ed quali y. On he o he hand, physiological signals can be comp essed mo e e i- cien ly and he quali y may be p ese ed. This s udy deals wi h his assump ion. The aims o his s udy a e o e alua e whe he he pa hologies p esen in ECG signals a ec (a) he quali y o he comp essed and econs uc ed signal and (b) he e iciency o he comp ession. Algo i hms such as SPIHT11,12 enable o se he comp ession e iciency di ec ly and hus hey a e sui able o sol ing he i s aim. In his case, all signals will be comp essed wi h cons an e iciency (a e age alue leng h = a L) o assess he in luence o pa holo- gies on quali y. Algo i hms such as single-cycle ac al-based (SCyF)29 do no enable di ec se ing o e iciency (a L) o quali y a e comp ession. Thus, hey a e sui able o sol ing he second aim, whe e he e iciency (a L) as well as he quali y change. The pa ame e s (desc ibed in “Me hods” sec ion) o he SCyF algo i hm (excluding he e iciency and quali y ones) will be se equally o all signals. Fo his s udy, he ECG signals om he CSE da abase we e used. Fi s ly, he analysis was pe o med on signals classi ied in o wo g oups—physiological and pa hological. Secondly, in pa hological signals, i was dis inguished whe he he hy hm and/o mo phology is pa hological. Thus, mo e de ailed analysis was pe o med as well. Noise is no conside ed wi hin his s udy and i is discussed in limi a ions. Me hods CSE da abase and anno a ions. Fo he pu pose o his s udy, he second mos ci ed30 da abase—Com- mon S anda ds o Quan i a i e Elec oca diog aphy (CSE) da abase31 was used. In his s udy, only he 125 o igi- nal (non-a i icial) signals om da ase 3 o CSE da abase we e used. These signals a e bo h wi hou and wi h a ious pa hologies. Each signal was ob ained om 15 leads—12 s anda d leads and 3 F ank leads. The leng h o each signal is 10 s, sampling equency is 500 Hz, bi esolu ion is 16 bps. In ou p e ious s udy, we eely published he anno a ions o pa hologies o he CSE da abase30. Fi e ca diologis s diagnosed he signals inde- penden ly and hen he 4R consensus was p o ided by wo ECG expe s and inal diagnoses we e de e mined. Final diagnoses include 38 s anda d diagnos ic s a emen s acco ding o Ame ican Hea Associa ion (AHA) ecommenda ions32. De ails can be ound in30. In he p esen s udy, 125 signals om he CSE da abase we e classi ied in o 4 g oups acco ding o he anno- a ions o pa hologies om30. Two ea u es we e assessed— hy hm and mo phology o he signals, each in o wo classes—physiological (F) and pa hological (P); hus, 4 g oups we e c ea ed. Physiological hy hms include ollowing ca ego ies: sinus hy hm, sho and p olonged PR in e al, sinus achyca dia, and sinus b adyca - dia. Physiological mo phology includes no mal QRS complexes and sup a en icula p ema u e complexes. Pa hological hy hms include second-deg ee and hi d-deg ee AV blocks, a ial ib illa ion, a ial lu e , and paced hy hm. Pa hological mo phologies include en icula hype ophy, myoca dial in a c ion, p ema u e 3 Vol.:(0123456789) Scien i ic Repo s | (2021) 11:10514 | h ps://doi.o g/10.1038/s41598-021-89817-w www.na u e.com/scien i ic epo s/ en icula complexes, paced complexes, and a ious diso de s o en icula conduc ion. Classi ica ion was p o ided by one o ECG expe s engaged in he p e ious s udy30. In Table1, o e iew o he g oups and he numbe o signals classi ied in each g oup is p esen ed. Table wi h anno a ions om ou p e ious a icle30 is wi hin he p esen s udy enhanced by hy hm and mo phology assessmen signal-by-signal. Enhanced able is p esen ed in he Supplemen a y TableS1. In case o wo-g oup analysis, h ee g oups wi h any pa hology (FP, PF, PP) we e clus e ed in o one g oup o pa hological signals. This g oup includes 86 signals. The second g oup includes only FF g oup consis ing o 39 signals. Fo es ing, all signals om he CSE da abase we e used, only F ank leads o signals no. 60, 68, 76, 84, 92, 100, 108, and 124 we e excluded, since hey a e o cons an alue. Comp ession and assessmen o i s e iciency and quali y. Comp ession was p o ided using wo algo i hms (a) combina ion o wa ele ans o m (WT) and SPIHT11,12 and (b) SCyF29. In his s udy, bo h com- p ession algo i hms a e lossy ones. The i s algo i hm decomposes he signals using WT in o wa ele coe icien s which c ea e empo al o i- en a ion ees. The SPIHT algo i hm i e a i ely codes he coe icien s based on hei impo ance using hei compa ison wi h h eshold. Ou pu o he algo i hm is a bi low. This algo i hm can be di ec ly con olled in e ms o bo h e iciency (a L) and quali y (no malized PRD = PRDN)12. SPIHT algo i hm was p ima ily dedi- ca ed o image comp ession33. La e , 1D e sion o he o iginal 2D SPIHT algo i hm was published and applied o ECG11. In his s udy, he SPIHT algo i hm implemen ed by H ubes e al.12, u he called SPIHT-H was used. SCyF algo i hm which is based on ac als was in oduced in ou ecen s udy29. SCyF algo i hm uses down- sampled single-cycle o ECG as a domain. The ECG signals a e hen di ided on ange blocks (RB) o block size (BS, in his s udy BS = 256). Fo each RB, he mos simila domain block (DB, o e lapping pa s o domain, o e lapping is se by jump s ep = JS which is 1 in his s udy) is sea ched. As a simila i y me ic, he ac al oo mean squa e (FRMS) is used (in his case, FRMS = 12). Simila i y be ween RB and DB can be inc eased using ac al coe icien s and a ine ans o m applied on domain block and/o di ision o RB (BS is hal ed, in his case, we used maximally 2 di isions). The ou pu s o he SCyF comp ession algo i hm a e domain, index o DB, ac al coe icien s, ype o used ans o m, and numbe o di isions (all in bina y o m). This algo i hm does no enable di ec se ing o he quali y o he e iciency o comp ession. O he pa ame e s o he algo i hm we e se equally as desc ibed abo e. The i s aim was o e eal whe he he quali y a e comp ession and econs uc ion is he same o physi- ological and pa hological signals o whe he he e a e signi ican di e ences. Fo his kind o es ing, he SPIHT- H algo i hm was used. The e iciency o comp ession was se equally o a L = 1 bps o all signals. The second aim was o e eal whe he physiological signals a e comp essed mo e o less e icien ly (a L) and wi h di e en quali y a e comp ession and econs uc ion han pa hological signal. In his case, he SCyF algo i hm was used. To assess he quali y o he signals a e comp ession and econs uc ion, 12 me ics ecommended in ou p e ious s udies10,34 we e used. These a e namely: pe cen age simila i y using s anda d de ia ion o NN (PSim SDNN), quali y sco e (QS), signal o noise a io (SNR1), mean squa e e o (MSE), no malized pe cen age oo mean squa e di e ence (PRDN1), maximum ampli ude e o (MAX), s anda d e o (STDERR), wa ele -ene gy based diagnos ic dis o ion using s a iona y wa ele ans o m (WEDD SWT), spec a di e ence (Spec um), simila i y—posi ions wi h ole ance o 10 (SiP10), simila i y—posi ions and ampli ude wi h ole ance o 10 (SiPA10), and dynamic ime wa ping—pe cen age ma ch o iducial poin s (DTW pm p2). S a is ics. A i s , he possible di e ence be ween wo g oups o signals (physiological and pa hological) was es ed. No mali y o he da a was es ed using Shapi o–Wilk and Lillie o s es . The null hypo hesis is ha he da a come om a no mally dis ibu ed popula ion. Acco ding o hese wo es s, da a we e no no mally dis ibu ed, hus nonpa ame ic es was applied. Two g oups (physiological and pa hological) a e conside ed in his case, hus he Mann–Whi ney es —a nonpa ame ic es dedica ed o wo independen g oups is chosen. I examines whe he hese wo g oups we e selec ed om popula ions wi h he same dis ibu ion. The null hypo hesis is ha he e is no di e ence be ween he g oups o physiological and pa hological signals. Secondly, we wen deepe and es ed whe he he e is he di e ence be ween ou g oups o signals (FF, FP, PF and PP). Ex ension o Mann–Whi ney es is K uskal–Wallis es dedica ed o wo o mo e independen g oups. I examines he null hypo hesis whe he hese ou g oups o igina e om he same dis ibu ion. Table 1. O e iew o 4 g oups o signals acco ding o he p esence o absence o pa hologies in hy hm and mo phology. F s ands o physiological and P s ands o pa hological signals. Rhy hm Mo phology G oup abb e ia ion Numbe o signals F F FF 39 F P FP 72 P F PF 7 P P PP 7 4 Vol:.(1234567890) Scien i ic Repo s | (2021) 11:10514 | h ps://doi.o g/10.1038/s41598-021-89817-w www.na u e.com/scien i ic epo s/ Mul iple compa ison o mean anks is used o de ail es ing a e ejec ing he null hypo hesis in K uskal–Wallis es . Each pai o g oups es ed using K uskal–Wallis es can be compa ed. Mean and compa i- son in e al o each o 12 me ics in each o 4 g oups (FF, FP, PF, and PP) is calcula ed. The null hypo hesis is ha he co esponding mean di e ence be ween wo g oups is equal o ze o. Resul s and discussion SPIHT-H (a L = 1 bps). Two g oups. In Table2, mean and median quali y is exp essed by 12 quali y me - ics o g oups o physiological and pa hological signals om CSE da abase. I was compa ed which o g oups shows highe quali y a e comp ession and econs uc ion while he e iciency is cons an . Such g oup is high- ligh ed in g een. In mos cases, be e quali y is eached in g oup con aining physiological signals. Acco ding o QS, he mean and median quali ies a e be e in pa hological g oup. Median PRDN, SNR and WEDD SWT show also be e quali y in pa hological g oup. Al hough in hese 5 cases, he pa hological signals a e o be e quali y han physiological signals, he di e ence be ween he quali y me ics is e y small. On he o he hand, i he physiological signals a e o be e quali y, he di e ence is in mos cases high. S a is ical analysis was pe o med on hese da a as well. As desc ibed in “Me hods”, Shapi o Wilk and Lil- lie o s es we e used o es no mali y o da a. As bo h es s showed ha he da a we e no no mally dis ibu ed, nonpa ame ic Mann–Whi ney (M–W) es was used o es ing whe he he e was a di e ence be ween physi- ological and pa hological g oups. The p alue o M–W es is shown in Table2 in which he cases when he null hypo hesis was ejec ed (p < 0.05) a e highligh ed in blue. The M–W p alues co espond wi h median. In case he physiological signals show be e esul s in e ms o median, he null hypo hesis o M–W es was ejec ed. In case he physiological signals show wo se esul s acco ding o median, he null-hypo hesis o M–W is no ejec ed. Fou g oups. In Table3, mean and median esul s o ou g oups o signals (FF, FP, PF, and PP) a e exp essed by 12 quali y me ics and one e iciency me ic (a L). The bes esul s ( he bes quali y o signals a e comp es- sion and econs uc ion) a e highligh ed in g een and he wo s ones in ed. In 8 cases ou o 12, he bes mean esul s a e eached in g oup FF. In 4 cases, g oup FP shows he bes esul s. On he o he hand, he wo s esul s Table 2. E alua ion o he di e ences be ween g oups o physiological (F) and pa hological (P) signals om CSE da abase comp essed by SPIHT-H me hod. Mean and median quali ies a e exp essed by 12 quali y me ics. G een colo highligh s be e esul s. M–W p means p alue o Mann–Whi ney es . Blue colo s ands o ejec ion o null hypo hesis. a L PRDN MSE SNR STDERR MAX WEDD SWT PSim SDNN QS Spec umSiP10SiPA10 DTW pm p2 F1.0275 4.6647 82.6010 28.3661 8.5846 43.6559 2.0402 99.5521 4.8928 2151019.7758 82.1918 32.5669 32.5104 mean P1.0271 4.7221 135.2911 28.3490 10.3788 54.3007 2.081698.5393 4.98552681905.0379 78.289028.3585 30.0635 F1.0298 3.3928 63.2859 29.3888 7.9422 43.5275 1.5592 100.0000 4.8356 2016289.0847 84.5833 34.0685 31.8681 median P1.0264 3.2728 82.1192 29.7019 9.0453 45.8556 1.515399.9905 4.92002311843.5171 81.934129.3619 29.7633 M-W p 0.855 0.762 7.69E-15 0.7624 7.96E-15 1.35E-08 0.411 2.78E-10 0.796 1.15E-15 1.71E-13 1.02E-12 1.18E-11 Table 3. E alua ion o he di e ences be ween ou g oups o signals om CSE da abase, namely: physiological hy hm and physiological mo phology (FF), physiological hy hm and pa hological mo phology (FP), pa hological hy hm and physiological mo phology (PF), pa hological hy hm and pa hological mo phology (PP). Mean and median quali y a e exp essed by 12 quali y me ics. G een colo highligh s he bes esul s and ed colo highligh s he wo s esul s. K–W p means p alue o K uskal–Wallis es . Blue colo s ands o ejec ion o null hypo hesis. a L PRDN MSE SNRSTDERRMAX WEDD SWT PSim SDNN QS Spec um SiP10 SiPA10 DTW pm p2 FF 1.0275 4.6647 82.6010 28.3661 8.5846 43.6559 2.0402 99.55214.8928 2151019.7758 82.1918 32.5669 32.5104 FP 1.0272 4.4818 109.0559 28.7266 9.6935 48.4807 1.9311 98.4828 5.1788 2492371.2574 79.8762 29.8273 30.4351 PF 1.0261 5.9971 141.2735 26.1849 10.9543 56.2650 2.6305 99.0924 3.7980 2880079.9420 73.0449 23.3463 29.4397 mean PP 1.0267 5.9478 404.2641 26.5539 16.9791 113.4969 3.1034 98.58724.1352 4465495.0433 66.8779 18.0042 26.7590 FF 1.0298 3.3928 63.2859 29.3888 7.9422 43.5275 1.5592 100.00004.8356 2016289.0847 84.5833 34.0685 31.8681 FP 1.0250 3.2138 75.1361 29.8597 8.6631 45.3415 1.4932 99.9854 5.0440 2182139.8476 83.2265 31.3773 29.6947 PF 1.0256 4.7433 98.4481 26.4785 9.9231 49.4513 2.1855 99.8916 3.4259 2525744.3042 76.2835 22.3835 29.8086 median PP 1.0246 4.3093 161.1616 27.3119 12.6796 58.4145 2.1440 99.96243.6321 3402761.4576 71.5924 18.7474 28.7234 K-W p 0.998 2.99E-06 7.83E-30 2.99E-06 8.23E-30 4.94E-23 4.76E-09 3.91E-09 5.15E-06 1.79E-33 1.27E-40 3.12E-33 2.06E-13 5 Vol.:(0123456789) Scien i ic Repo s | (2021) 11:10514 | h ps://doi.o g/10.1038/s41598-021-89817-w www.na u e.com/scien i ic epo s/ a e ound in g oup PP (8 ou o 12 me ics). Th ee wo s esul s a e shown also in g oup PF and one in g oup FP. Fo median, he esul s a e simila . In 8 ou o 12 cases, he bes esul s a e in g oup FF. In ou cases, he FP g oup shows he bes esul s. The wo s esul s a e in g oups PP and PF in 8 and 4 cases, espec i ely. I is ob ious ha he g oup o signals wi h physiological hy hm and physiological mo phology can be gene ally comp essed wi h he lowes e o . Acco ding o mos o quali y me ics, signals wi h physiological hy hm and pa hological mo phology a e comp essed wi h lowe e o han signals wi h pa hological hy hm and physiological mo phol- ogy. Finally, he signals wi h pa hological hy hm and pa hological mo phology a e comp essed wi h he highes e o ( he wo s quali y). The null hypo hesis whe he all ou g oups o igina e om he same dis ibu ion was es ed. Fo his pu - pose, non-pa ame ic K uskal–Wallis es was used. The K uskal–Wallis p alue is shown in Table3. Blue colo highligh s he ejec ion o null hypo hesis (p < 0.05) in all cases. I means ha no all g oups o igina e om he same dis ibu ion. Thus, u he es ing was pe o med using mul iple compa ison o mean anks (g oup mean anks) o compa e each pai o g oups (Fig.1). The esul s (p alues) a e summa ized in Table4. In Table4, he ejec ion o null hypo hesis is highligh ed in blue. The ejec ion o null hypo hesis co esponds wi h disjoin o in e als in Fig.1. I means ha mos pai s o g oups di e signi ican ly. The di e ence be ween he g oups is illus a ed by Fig.1. The di e ence be ween FF and PP g oups is s a is ically signi ican acco ding o all 12 quali y me ics. FF and PF g oups also di e acco ding o all quali y me ics, FF and FP g oups di e acco ding o 8 me ics. PP and PF g oups di e acco ding o 6 quali y me ics. F om Table3 and Fig.1, i is e iden ha signals wi h physiological hy hm and pa hological mo phology a e comp essed wi h lowe e o han signals wi h pa hological hy hm and physiological mo phology in a s a is ically signi ican manne (in 10 cases). Acco ding o me ic PSim SDNN we canno ejec he null hypo hesis in hal o pai s o g oups. I can be s a ed o ha his me ic is he leas dependen on p esence/absence o pa hologies in he ECG signal. SCyF. Two g oups. Table5 shows he esul s o es ing he SCyF comp ession algo i hm on wo g oups o signals (physiological and pa hological). The SCyF algo i hm was se equally o all signals. Thus, he e ec o pa hology p esence on comp ession e iciency (a L) can be e alua ed. I is e iden ha he g oup o physiologi- cal signals is comp essed mo e e ec i ely (wi h lowe a L, highligh ed in yellow) han he g oup o pa hological signals. The di e ence is s a is ically signi ican (highligh ed in o ange). The mean and he median quali ies o he econs uc ed signals a e also be e in he g oup o physiological signals acco ding o 11 ou o 12 me ics (excep o WEDD SWT). In 9 ou o he p e iously men ioned 11 me ics, he di e ence is s a is ically signi i- can (highligh ed in blue). WEDD SWT STDERR Spec um SNR SiPA10 SiP10 QS PSimSDNN PRDN MSE MAX DTWpm p2 G oupmeanand compa ison in e al FP FF PP PF Figu e1. G aphs o mul iple compa ison o mean anks. Table 4. The esul s o mul iple compa ison o mean anks o ou g oups o signals om he CSE da abase. Rejec ion o null hypo hesis is highligh ed in blue. a L PRDN MSE SNR STDERR MAX WEDD SWT PSim SDNN QS Spec a Di e ence SiP10 SiPA10 DTW pm p2 FP FF 9.98E-01 7.70E-01 9.35E-08 7.70E-01 9.62E-08 1.62E-03 9.61E-02 4.19E-087.60E-01 4.96E-08 4.24E-05 2.64E-05 5.49E-08 FP PP 1.00E+00 2.12E-03 3.77E-09 2.12E-03 3.77E-09 3.77E-09 1.22E-064.92E-01 2.88E-03 3.77E-09 3.77E-09 3.77E-09 5.28E-04 FP PF 1.00E+00 1.25E-04 6.03E-03 1.25E-04 5.78E-03 2.03E-03 1.73E-049.31E-01 1.72E-04 7.19E-03 5.45E-09 4.02E-076.02E-01 FF PP 9.98E-01 1.58E-02 3.77E-09 1.58E-02 3.77E-09 3.77E-09 5.45E-04 1.93E-04 2.07E-02 3.77E-09 3.77E-09 3.77E-09 4.13E-09 FF PF 1.00E+00 1.55E-03 2.94E-08 1.55E-03 2.80E-08 1.27E-06 1.83E-02 4.14E-03 2.09E-03 2.87E-08 3.77E-09 3.77E-09 4.22E-04 PP PF 1.00E+00 9.56E-01 6.83E-04 9.56E-01 7.21E-04 3.86E-03 8.75E-019.33E-019.54E-01 4.79E-05 1.76E-02 3.18E-021.92E-01 6 Vol:.(1234567890) Scien i ic Repo s | (2021) 11:10514 | h ps://doi.o g/10.1038/s41598-021-89817-w www.na u e.com/scien i ic epo s/ Lossy comp ession is gene ally ade-o be ween i s e iciency and quali y o he signals a e comp ession and econs uc ion. I he a L would be se equally o each signal (as in case o SPIHT-H algo i hm), he quali y di e ence be ween he g oup o physiological signals and he g oup o pa hological signals would be highe and e en mo e quali y me ics would be be e o physiological signals. Fou g oups. The es ing o comp ession pe o mance on signals om 4 g oups was pe o med by SCyF algo- i hm. The esul s a e shown in Table6. A i s , K uskal–Wallis es was used o examine whe he all ou g oups o igina e om he same dis ibu ion. The null hypo hesis was ejec ed in all cases (p < 0.05, blue colo in he second line o Table6) and hus u he es ing using mul iple compa ison o mean anks was pe o med. The assump ion in case o compa ing o FF wi h any o he g oup is ha FF signals show be e quali y a e comp ession and econs uc ion. In case o compa ison o PP wi h any o he g oup, he assump ion is ha PP signals show wo se quali y a e comp ession and econs uc ion. F om p e ious esul s (using SPIHT-H algo- i hm) we also know ha signals wi h pa hological hy hm (PF) a e comp essed wi h wo se quali y han signals wi h pa hological mo phology (FP). The same we assume in case o using SCyF algo i hm. Mul iple compa ison o mean anks shows ha pai s o g oups FP and FF, FP and PP, FP and PF, FF and PP, FF and PF di e om each o he signi ican ly acco ding o he majo i y o quali y me ics (blue and ed colo in Table6). The blue colo means ha he quali y is be e o he g oup acco ding o assump ion. I he assump ion is no ul illed, he esul is highligh ed in ed. I con i ms he heo y ha pa hological signals a e comp essed wi h lowe pe o mance han physiological signals. Mo eo e , he e iciency o comp ession (a L alue) di e s be ween he g oups (excep o compa ison o PP and PF) signi ican ly as well which can be seen in he a L column in Table6. I he a L would be he same o FF, FP, PF, and PP g oups, he di e ences in quali y would be e en highe because lossy comp ession is a comp omise be ween comp ession e iciency and signal quali y. Analysis o pa hological signals comp ession. The analysis o pa hological signals comp ession was pe o med using he SCyF algo i hm o e eal pa icula di e ences in comp ession o physiological signals and signals wi h pa hologies. O iginal signal is colo ed in blue, comp essed and econs uc ed signal in ed. The Table 5. E alua ion o he di e ences be ween g oups o physiological (F) and pa hological (P) signals om he CSE da abase comp essed by he SCyF algo i hm. Mean and median quali ies a e exp essed by 12 quali y me ics. Yellow colo highligh s be e esul in e ms o comp ession e iciency. G een colo highligh s be e esul s in e ms o he quali y o signals a e comp ession and econs uc ion. M-W p means p- alue o Mann- Whi ney es . O ange colo s ands o ejec ion o null hypo hesis in case o a L. Blue colo s ands o ejec ion o null hypo hesis in case o he quali y me ics. a L PRDN MSESNR STDERR MAX WEDD SWT PSim SDNN QS Spec um SiP10 SiPA10 DTW pm p2 F0.9723 7.5780 243.7472 23.3544 14.6600 197.1236 3.1059 95.5514 2.7252 3738879.5223 76.4075 26.9060 27.6156 mean P1.0730 8.2560 2755.9656 23.3243 23.0566 326.4450 2.7157 92.0486 2.6293 5781766.8892 73.7963 24.2497 26.4213 F0.9434 6.3097 178.4755 23.9998 13.3303 140.3195 2.6897 99.1715 2.6886 3373053.7958 79.5114 27.9387 25.9259 median P1.0234 6.5808 193.5184 23.6344 13.9116 175.2892 2.2107 98.5819 2.3252 3511952.1003 76.8311 24.7621 25.6944 M-W p 6.75E-280.7303 4.13E-11 0.7303 4.13E-11 2.84E-05 3.40E-12 9.33E-07 0.0014 4.01E-08 7.36E-05 1.93E-10 0.0274 Table 6. The second line shows he esul s (p- alues) o K uskal–Wallis es o examine whe he all ou g oups o igina e om he same dis ibu ion. The es o he able shows he esul s o mul iple compa ison o mean anks o e eal whe he each pai o g oups di e om each o he in a s a is ically signi ican way. Rejec ion o null hypo hesis is highligh ed in blue and ed. The ed colo means ha he quali y o he signal a e comp ession and econs uc ion is be e o he o he g oup han i was assumed. a L PRDN MSESNR STDERRMAX WEDD SWT PSim SDNN QS Spec a Di e ence SiP10SiPA10 DTW pm p2 K-W p 4.76E-52 1.70E-16 2.38E-26 1.70E-16 2.12E-26 5.15E-10 1.45E-14 2.64E-18 1.09E-26 1.08E-18 5.70E-28 4.53E-33 1.12E-02 FP FF 3.77E-09 7.92E-02 6.25E-05 7.92E-02 6.35E-05 1.19E-02 3.77E-09 2.17E-07 9.40E-01 6.18E-04 5.53E-01 1.33E-03 4.61E-02 FP PP 3.77E-09 3.77E-09 3.77E-09 3.77E-09 3.77E-09 5.70E-07 2.18E-03 5.82E-02 3.77E-09 3.77E-09 3.77E-09 3.77E-09 6.06E-02 FP PF 3.85E-09 1.99E-07 6.88E-05 1.99E-07 6.51E-05 7.99E-01 2.31E-02 3.77E-09 3.77E-09 3.81E-01 3.77E-09 3.79E-09 9.85E-01 FF PP 3.77E-09 3.70E-08 3.77E-09 3.70E-08 3.77E-09 3.97E-09 9.89E-01 2.35E-06 3.77E-09 3.77E-09 3.77E-09 3.77E-09 6.46E-01 FF PF 3.77E-09 1.67E-04 5.05E-09 1.67E-04 4.96E-09 8.71E-02 7.49E-01 1.29E-04 3.77E-09 3.74E-03 3.77E-09 3.77E-09 8.00E-01 PP PF 2.42E-01 5.89E-01 3.21E-02 5.89E-01 3.23E-02 5.81E-03 9.54E-01 3.77E-09 9.34E-01 8.23E-05 9.98E-01 5.80E-01 3.82E-01 7 Vol.:(0123456789) Scien i ic Repo s | (2021) 11:10514 | h ps://doi.o g/10.1038/s41598-021-89817-w www.na u e.com/scien i ic epo s/ e iciency o comp ession in e ms o a L and he quali y a e comp ession and econs uc ion in e ms o he mos common PRDN and ad anced WEDD SWT me ics a e shown in Fig.2. In Fig.2a,b, examples o comp ession o wo signals wi h physiological hy hm and physiological mo phol- ogy a e p esen ed. These signals a e comp essed wi h qui e low a L (high e iciency) and low dis o ion (PRDN, SWT WEDD). Figu e2c shows he de ail o signal no. 70 (lead aVF) wi h a pacemake peak. This signal belongs in o PP g oup (pa hological hy hm and pa hological mo phology). Pacemake peak is sha p and i also includes high- equency componen s. The esul s o comp ession using SCyF algo i hm shows signi ican dis o ion and highe a L (lowe e iciency). In his case, i is caused by he se ing o he SCyF algo i hm in which he downsampling o he domain is used and hus he sha p pacemake peak canno be p ese ed. This p oblem can be sol ed by di e en se ing o he algo i hm (no downsampling o he domain) a he cos o he e iciency educ ion (a L = 1.8756 bps, PRDN = 4.1994%, WEDD SWT = 3.0441%). Figu e2d shows he de ail o signal no. 111 ( he 2nd F ank lead) wi h pa hological hy hm and physiological mo phology. The signal was diagnosed as an a ial lu e . The dis o ion may be caused by i egula occu ence o lu e wa es and hei ansi ion o QRS complex. In SCyF comp ession algo i hm, as he domain only single ECG cycle is used and hus he algo i hm may ha e p oblem wi h cycles ha look di e en ly. Limi a ions o he s udy. The in luence o pa hologies on comp ession pe o mance was es ed only on one s anda d da abase (CSE da abase). Al hough his da abase includes a ious pa hologies, i does no include all usual ypes (e.g. en icula achyca dia, en icula ib illa ion o idio en icula hy hm). On he o he hand, each signal o his da abase is anno a ed. Fi e ca diologis s diagnosed each signal and he ea e wo ECG expe s made 4R consensus. This p ocess was e y ime- and sou ce-consuming and i would be e y di icul o epea i o any o he da abase. Al hough signal comp ession is in luenced by noise, his e ec has no been s udied. I is a e y complex p oblem, which is ou -o -scope o his s udy. I would dese e a sepa a e s udy. The o iginal signals om s and- a d CSE da abase we e used in his s udy. These signals we e no il e ed. The eason is ha he il a ion pu s e o s in he signals and he analysis would no be objec i e. Al hough bo h used comp ession algo i hms ha e il a ion p ope y, o each he objec i e esul s o comp ession pe o mance, he clea signal (wi hou noise) should be known which is no possible unde eal condi ions. The solu ion would be o c ea e a i icial signals wi h di e en pa hologies and known le el o noise. Since in his s udy only wo comp ession algo i hms we e employed, i s esul s canno be gene alized on all exis ing comp ession algo i hms. On he o he hand, al hough bo h used algo i hms a e based on di e en p inciples, hei ou come is e y simila . In ano he wo ds, bo h algo i hms used in his s udy show he same end—physiological signals a e comp essed be e han pa hological signals. U[µV] (a )( b) (d)(c) 1250 1300 1350 1400 1450 1500 1550 Samples[-] -400 -200 0 200 400 600 800 1000 a L=1.0952bps PRDN =4.5412% WEDD SWT=1.2295% Pwa e QRScomplex Twa e Samples[-] 1950 2000 2050 2100 2150 2200 2250 2300 2350 -400 -200 0 200 400 600 800 1000 1200 1400 1600 a L=0.8554 bps PRDN =3.6284% WEDD SWT=1.5619% Pwa e QRScomplex Twa e -10000 -5000 0 5000 -7500 -2500 2500 a L=1.3074 bps PRDN =67.9243% WEDD SWT=29.8300% a ial pacing QRScomplex en icula pacing 2900295030003050 3100 3150 3200 3250 3300 Samples[-] lu e wa es 4200 4300 4400 4500 4600 4700 -200 0 200 400 600 800 1000 1200 a L=1.0954 bps PRDN =10.9933% WEDD SWT=4.9056% Samples[-] Figu e2. An example o comp ession o ECG signals wi hou and wi h pa hologies in hy hm and/o in mo phology. The SCyF algo i hm was used o comp ession. The blue colo s ands o he o iginal signal and he ed one o comp essed and econs uc ed signal. Diagnoses a e ma ked in g een. (a) No mal ECG and sinus achyca dia; physiological hy hm and physiological mo phology, lead V2 o signal 16; (b) no mal ECG and sinus hy hm; physiological hy hm and physiological mo phology, lead V3 o signal 73; (c) AV dual-paced complexes o hy hm; pa hological hy hm and pa hological mo phology, lead aVF o signal 70; (d) a ial lu e ; pa hological hy hm and physiological mo phology, he 2nd F ank lead o signal 111. 8 Vol:.(1234567890) Scien i ic Repo s | (2021) 11:10514 | h ps://doi.o g/10.1038/s41598-021-89817-w www.na u e.com/scien i ic epo s/ Conclusion As s a ed in “In oduc ion”— o he bes o ou knowledge, he e is no a single s udy p ima ily ocused on he possible e ec o ECG pa hologies on he pe o mance o comp ession algo i hms. The aims o his s udy we e o e alua e whe he he pa hologies p esen in ECG signals a ec (a) he quali y o he comp essed and econ- s uc ed signal and (b) he e iciency o he comp ession. Acco ding o his s udy esul s, signi ican di e ences exis be ween comp ession o physiological signals and signals wi h a ious hy hm and/o mo phology pa hologies. The di e ences we e p o en based on wo comp ession algo i hms o di e en p inciples (SPIHT-H and SCyF) and majo i y o 12 quali y me ics. The di e ences a y among quali y me ics which suppo s ou p e ious ecommenda ion o use a combina ion o se e al me hods o he obus assessmen o ECG signal quali y a e comp ession10. While he e iciency o comp ession is cons an (SPIHT-H algo i hm was used), i was p o en ha pa hologi- cal signals we e comp essed wi h lowe quali y han physiological signals acco ding o he majo i y o used quali y me ics. The mo e de ailed analysis o ou g oups o signals showed ha signals wi h physiological hy hm and physiological mo phology we e comp essed wi h he bes quali y acco ding o majo i y o quali y me ics. The wo s esul s epo ed he g oup o signals wi h pa hological hy hm and pa hological mo phology. The di e ence be ween hese wo g oups was signi ican acco ding o all 12 quali y me ics. Signals wi h physiological hy hm and pa hological mo phology we e in mos cases comp essed wi h lowe e o han signals wi h pa hological hy hm and physiological mo phology. I was obse ed ha PSim SDNN quali y me ic was he leas dependen on he p esence o pa hologies. I was s a is ically p o en ha pa hological signals we e comp essed wi h lowe e iciency han physiologi- cal signals using he same se ing o SCyF algo i hm o all signals. Al hough, comp ession is always a ade-o be ween e iciency and quali y o comp ession and in ou s udy he e iciency was lowe o pa hological signals, he quali y was also lowe o pa hological signals acco ding o majo i y o quali y me ics. Mo e de ailed analy- sis o ou g oups e ealed e y simila conclusions as he same analysis using SPIHT-H algo i hm (desc ibed abo e). Thus, i is e iden ha pa hologies a ec he pe o mance o comp ession algo i hms. This s udy is he i s one o i s kind and b ings new knowledge in o he a ea o comp ession. Mo eo e , i enhances he anno a ions o he CSE da abase in e ms o classi ica ion he signals in o physiological and pa hological ones om wo poin s o iew— hy hm and mo phology. The enhanced anno a ions can be used in b oade a ea o ECG signal p ocessing and a e no limi ed o comp ession. Da a a ailabili y CSE da abase analyzed in he cu en s udy a e a ailable om p o . Paul Rubel, INSERM ERM107-MTIC Hôpi al Ca diologique, 28 a enue du Doyen Lépine, 69677 BRON Cedex, F ance bu es ic ions apply o he a ailabili y o hese da a, which we e used unde license o he cu en s udy, and so a e no publicly a ailable. Da a a e howe e a ailable om he au ho s upon easonable eques and wi h pe mission o p o . Paul Rubel, INSERM ERM107-MTIC Hôpi al Ca diologique, 28 a enue du Doyen Lépine, 69677 BRON Cedex, F ance. Recei ed: 28 June 2020; Accep ed: 29 Ap il 2021 Re e ences 1. Bayés de Luna, A. Clinical Elec oca diog aphy 4 h edn. (Wiley-Blackwell, 2010). 2. Zhang, O. & F ick, K. All-ECG: A leas -numbe o leads ECG moni o o s anda d 12-lead ECG acking du ing mo ion*. P oc. IEEE Heal hc. Inno . Poin Ca e Technol. 20, 103–106 (2019). 3. Giada, F. & Ra iele, A. Clinical app oach o pa ien s wi h palpi a ions. Ca d. Elec ophysiol. Clin. 10, 387–396 (2018). 4. Thale , M. S. The Only EKG Book You’ll E e Need 5 h edn. (Lippinco Williams & Wilkins, 2007). 5. Alesanco, A. & Ga cia, J. Au oma ic eal- ime ECG coding me hodology gua an eeing signal in e p e a ion quali y. IEEE T ans. Biomed. Eng. 55, 2519–2527 (2008). 6. Elgendi, M., Al-Ali, A., Mohamed, A. & Wa d, R. Imp o ing emo e heal h moni o ing: A low-complexi y ECG comp ession app oach. Diagnos ics (Basel) 8, 1–17 (2018). 7. Manikandan, M. S. & Dandapa , S. Wa ele -based elec oca diog am signal comp ession me hods and hei pe o mances: A p ospec i e e iew. Biomed. Signal P ocess. Con ol 14, 73–107 (2014). 8. C a en, D., McGinley, B., Kilma in, L., Gla in, M. & Jones, E. Impac o comp essed sensing on clinically ele an me ics o ambula o y ECG moni o ing. Elec on. Le . 51, 323–324 (2015). 9. Lee, S., Kim, J. & Lee, M. A eal- ime ECG da a comp ession and ansmission algo i hm o an e-Heal h de ice. IEEE T ans. Biomed. Eng. 58, 2448–2455 (2011). 10. Nemco a, A., Smisek, R., Ma sano a, L., Smi al, L. & Vi ek, M. A compa a i e analysis o me hods o e alua ion o ECG signal quali y a e comp ession. Biomed. Res. In . 2018, A . no. 1868519 (2018). 11. Lu, Z. T., Kim, D. Y. & Pea lman, W. A. Wa ele comp ession o ECG signals by he se pa i ioning in hie a chical ees algo i hm. IEEE T ans. Biomed. Eng. 47, 849–856 (2000). 12. H ubes, J., Vi ek, M. & Kozumplik, J. Possibili ies o wa ele decomposi ion o SPIHT comp ession o ECG signals. Anal. Biomed. Signals Images 20, 451–454 (2008). 13. Bales ie i, E., De Vi o, L., Pica iello, F. & Tudosa, I. A no el me hod o comp essed sensing based sampling o ECG signals in medical-IoT e a. P oc. IEEE In . Symp. Med. Meas. Appl. 20, 20 (2019). 14. Fi a, C. M. & Go as, L. An ECG signals comp ession me hod and i s alida ion using NNs. IEEE T ans. Biomed. Eng. 55, 1319–1326 (2008). 15. Rakshi , M. & Das, S. Elec oca diog am bea ype dic iona y based comp essed sensing o eleca diology applica ion. Biomed. Signal P ocess. Con ol 47, 207–218 (2019). 16. Chen, W. S., Hsieh, L. & Yuan, S. Y. High pe o mance da a comp ession me hod wi h pa e n ma ching o biomedical ECG and a e ial pulse wa e o ms. Compu . Me hods P og ams Biomed. 74, 11–27 (2004). 17. Chou, H. H., Chen, Y. J., Shiau, Y. C. & Kuo, T. S. An e ec i e and e icien comp ession algo i hm o ECG signals wi h i egula pe iods. IEEE T ans. Biomed. Eng. 53, 1198–1205 (2006). 9 Vol.:(0123456789) Scien i ic Repo s | (2021) 11:10514 | h ps://doi.o g/10.1038/s41598-021-89817-w www.na u e.com/scien i ic epo s/ 18. Be a, P., Gup a, R. & Saha, J. P ese ing abno mal bea mo phology in long- e m ECG eco ding: An e icien hyb id comp ession app oach. IEEE T ans. Ins um. Meas. 69, 2084–2092 (2020). 19. Ibaida, A., Al-Shamma y, D. & Khalil, I. Cloud enabled ac al based ECG comp ession in wi eless body senso ne wo ks. Fu u . Gene . Compu . Sys . 35, 91–101 (2014). 20. Nasimi, F., Khayyambashi, M. R. & Mo ahhedinia, N. Exploi ing simila p io knowledge o comp essing ECG signals. Biomed. Signal. P ocess. Con ol 60, 13 (2020). 21. Alesanco, A. & Ga cia, J. A simple me hod o gua an eeing ECG quali y in eal- ime wa ele lossy coding. EURASIP J. Ad . Signal P ocess. 2007, A . No. 93195 (2007). 22. Goldbe ge , A. L. e al. PhysioBank, PhysioToolki , and PhysioNe —componen s o a new esea ch esou ce o complex physi- ologic signals. Ci cula ion 101, E215–E220 (2000). 23. Moody, G. B., Ma k, R. G. & Goldbe ge , A. L. E alua ion o he “TRIM” ECG da a comp esso . P oc. Compu . Ca diol. 20, 167–170 (1988). 24. Sil ei a, R. M. R., Agulha i, C. M., Bona i, I. S. & Pe es, P. L. D. A gene ic algo i hm o comp ess elec oca diog ams using pa am- e e ized wa ele s. P oc. IEEE In . Symp. Signal P ocess. In o m. Technol. 20, 402–440 (2007). 25. Raja shi, G. & Biswas, D. Heal h Moni o ing Sys ems: An Enabling Technology o Pa ien Ca e (CRC P ess, 2019). 26. Wang, F. e al. A no el ECG signal comp ession me hod using spindle con olu ional au o-encode . Compu . Me hods P og. Biomed. 175, 139–150 (2019). 27. Hsieh, J.-H., Hung, K.-C., Liu, J.-H. & Wu, T.-C. Wa ele -based quali y-cons ained ECG da a comp ession sys em wi hou decod- ing p ocess. IEEE Mul imed. 27, 33–45 (2020). 28. Jha, C. K. & Koleka , M. H. Diagnos ic quali y assu ed ECG signal comp ession wi h selec ion o app op ia e mo he wa ele o minimal dis o ion. IET Sci. Meas. Technol. 13, 589–598 (2019). 29. Nemco a, A., Vi ek, M. & No ako a, M. Complex s udy on comp ession o ECG signals using no el single-cycle ac al-based algo i hm and SPIHT. Sci. Rep. 10, 15 (2020). 30. Smisek, R. e al. CSE da abase: Ex ended anno a ions and new ecommenda ions o ECG so wa e es ing. Med. Biol. Eng. Compu . 55, 1473–1482 (2017). 31. The CSE Wo king Pa y, Da abase: Common s anda ds o quan i a i e elec oca diog aphy [CD-ROM], 1990. 32. Mason, J. W., Hancock, E. W. & Ge es, L. S. Recommenda ions o he s anda diza ion and in e p e a ion o he elec oca diog am– Pa II: Elec oca diog aphy diagnos ic s a emen lis —a scien i ic s a emen om he Ame ican Hea Associa ion Elec oca di- og aphy and A hy hmias Commi ee, Council on Clinical Ca diology; he Ame ican College o Ca diology Founda ion; and he Hea Rhy hm Socie y—endo sed by he In e na ional Socie y o Compu e ized Elec oca diology. Ci cula ion 115, 1325–1332 (2007). 33. Said, A. & Pea lman, W. A. A new, as , and e icien image codec based on se pa i ioning in hie a chical ees. IEEE T ans. Ci c. Sys . Video Technol. 6, 243–250 (1996). 34. Nemco a, A., Vi ek, M., Ma sano a, L., Smisek, R. & Smi al, L. Assessmen o ECG signal quali y a e comp ession. P oc. IUPESM Wo ld Cong . Med. Phys. Biomed. Eng. 20, 169–173 (2019). Acknowledgemen s This esea ch was unded by Masa yk Uni e si y as pa o he p ojec no. MUNI/A/1246/2020 and by he p ojec no. LQ1605 om he Na ional P og am o Sus ainabili y II (MEYS CR). Au ho con ibu ions All au ho s pa icipa ed in he in e p e a ion o he esul s, and e iew o he manusc ip . A.N. did he es ing and analysis, w o e he manusc ip and p epa ed he igu es and ables. R.S. did he anno a ions and con ibu ed o p epa a ion o igu es. R.S., M.V. and M.N. con ibu ed o c i ical eading o he manusc ip . Compe ing in e es s The au ho s decla e no compe ing in e es s. Addi ional in o ma ion Supplemen a y In o ma ion The online e sion con ains supplemen a y ma e ial a ailable a h ps:// doi. o g/ 10. 1038/ s41598- 021- 89817-w. Co espondence and eques s o ma e ials should be add essed o A.N. Rep in s and pe missions in o ma ion is a ailable a www.na u e.com/ ep in s. Publishe ’s no e Sp inge Na u e emains neu al wi h ega d o ju isdic ional claims in published maps and ins i u ional a ilia ions. Open Access This a icle is licensed unde a C ea i e Commons A ibu ion 4.0 In e na ional License, which pe mi s use, sha ing, adap a ion, dis ibu ion and ep oduc ion in any medium o o ma , as long as you gi e app op ia e c edi o he o iginal au ho (s) and he sou ce, p o ide a link o he C ea i e Commons licence, and indica e i changes we e made. The images o o he hi d pa y ma e ial in his a icle a e included in he a icle’s C ea i e Commons licence, unless indica ed o he wise in a c edi line o he ma e ial. I ma e ial is no included in he a icle’s C ea i e Commons licence and you in ended use is no pe mi ed by s a u o y egula ion o exceeds he pe mi ed use, you will need o ob ain pe mission di ec ly om he copy igh holde . To iew a copy o his licence, isi h p:// c ea i eco mmons. o g/ licen ses/ by/4. 0/. © The Au ho (s) 2021