Analiza kardiosignala pomoću druge generacije talasić

Abstract

Današnja istraţivanja vezana za kardiovaskularne bolesti usmerena su ka raĉunarskoj analizi kardiosignala, digitalnih zapisa kardiovaskularne aktivnosti i funkcionalnosti, imajući u vidu dva aspekta: ekonomski i telemedicinski. U cilju izbegavanja komplikovanih i skupih sistema za analizu kardiološkog stanja, efikasne tehnike automatske (mašinske) obrade signala imaju veliki znaĉaj za potrebe primarne nege. ObezbeĊivanje visokokvalitetnog signala koji opisuje stanje organizma, radi dobijanja informacije o patofiziologiji, predstavlja prioritet. U ovoj disertaciji se pod kardiosignalima podrazumevaju elektrokardiogram (ECG) i fonokardiogram (PCG), kao tzv. primarni kardiosignali, koji predstavljaju reprezentativne primere elektriĉnog i vibroakustiĉnog zapisa rada kardiološkog sistema. Talasna (vejvlet) transformacija se ĉesto koristi za potrebe reprezentacije sloţenih signala, pored ostalih i kardiosignala, i izdvajanja njihovih vaţnih karakteristika. Ova transformacija je otvorila vrata razvoju razliĉitih tehnika koje mogu biti od znaĉaja za konkretne primene, kao što je kliniĉka praksa u medicini. Nakon uvoĊenja talasne transformacije, javila su se i poboljšanja u vidu tzv. druge generacije talasića. Druga generacija predstavlja samo alternativan pristup za implementaciju diskretne talasne transformacije, dok je lifting šema efikasan algoritam za konstrukciju FIR filtar banki zasnovanih na talasićima (vejvletima). U ovoj disertaciji je pretpostavljeno da se upotrebom talasne transformacije i lifting šeme moţe efikasno analizirati kardiosignal. Naime, pretpostavljena je veza izmeĊu kardiosignala i primene diskretne talasne transformacije u cilju bolje vizuelizacije njegovog sadrţaja. Za takvu namenu znaĉajni su spektralni i spektrogramski prikazi signala. Zdruţena vremensko-frekvencijska transformacija, kao što je kontinualna talasna transformacija, multifraktalan spektar, ali i empirijski pristup, moţe biti primenjen u analizi kardiosignala. Shodno potrebi analize, neophodna su rešenja koja omogućavaju izdvajanje i/ili detektovanje srĉanih dogaĊaja od kliniĉke vaţnosti. Neadekvatna primena diskretne talasne transformacije u redukciji šuma utiĉe na visokofrekvencijske komponente kardiosignala. Ovakve komponente ne moraju biti amplitudski dominantne i jasno uoĉljive, a mogu biti kliniĉki relevantne u dijagnostici. Manifestacija klik-sindroma u fonokardiografiji je dobar primer. Za potrebe istraţivanja obavljena je akvizicija fonokardiograma pedijatrijskih pacijenata, zdravih i pacijenata sa prolapsom mitralne valvule (PMV). Njihovom upotrebom i analizom, predloţen je model za vizuelnu verifikaciju nalaza primenom diskretne talasne transformacije. Upotrebljena je interpretacija detalja pomoću greške predikcije i njihova efikasna lokalizacija u vremenskom domenu. Izbor reda prediktora kontrolisan je detekcijom dominantnih srĉanih zvukova i indikacijom potencijalnih abnormalnosti. Ovakva kontrolisana analiza smanjuje rizik zanemarivanja kliniĉki vaţnih informacija, kao što je klik-sindrom. Predloţeni model i eksperimentalni rezultati ukazuju na nekoliko mogućnosti za dalja istraţivanja i razvoj samopodešavajućih pristupa za analizu kardiosignala.

Current research of cardiovascular diseases is oriented towards the computer analysis of cardiosignals, digital records of cardiovascular activity and functionality, bearing in mind two aspects: economic and telemedical. In order to avoid complex and expensive systems for cardiovascular state analysis, efficient automatic (machine) signal processing techniques are of great importance in primary care. Providing high-quality signal which decribes the state of the organism, in order to obtain information about the pathophysiology, is a priority. In this thesis, electrocardiogram (ECG) and phonocardiogram (PCG) are refered to as cardiosignals, so called primary cardiosignals, which are representative examples of electrical and vibro-acoustical records of cardiac system functionality. Wavelet transform is often used in representation of complex signals, among others cardiosignals, and their valuable feature extraction. This transform opened the door for development of different techniques that can be of importance in particular applications, such as in clinical practice in medicine. Following the wavelet transform, novel enhancement approches are introduced, like so called second generation wavelets. The second generation represents an alternative approach for discrete wavelet transform implementation, where a lifting scheme is an efficient algorithm for FIR wavelet filterbank construction. In the thesis it is assumed that wavelet transform and lifting scheme can be used for efficient cardiosignal analysis. Namely, a relationship between cardiosignal and application of discrete wavelet transform is hypothesized in order to improve the visualization of its content. For such purpose, spectral and spectrogram based representation of a signal are significant. Joint time-frequency representation, such as continious wavelet transform, multifractal spectrum, as well as empirical approach, can be applied in cardiosignal analysis. In accordance with the analysis, it is inevitable to develop approaches that enable extraction and/or detection of cardiac events of clinical importance. Inadequate application of discrete wavelet transform in noise reduction may affect high frequency cardiosignal components. Such components are not necessarily either characterized by dominant amplitude or even clearly noticeable, but they are usually clinically relevant in diagnostics. Click-syndrome manifestation in phonocardiorgaphy is such an example. For the purpose of research, phonocardiograms are recorded from pediatric patients: healthy patients and patients with prolapsed mitral valve (PMV). Their use and analysis have led to the proposition of a model for visual click finding verification via discrete wavelet transform. Details are interpreted through prediction error and efficiently localized in time domain. The choice of prediction order is carried by detection of dominant heart sounds and potential abnormality indication. Such controlled analysis decreases the risk of neglecting clinically important information, such as the click-syndrome. Proposed model and experimental results show several possibilities for further investigations and development of self-adjusting approaches for cardiosignal analysis.