SEISMOCARDIOGRAM AND PHONOCARDIOGRAM FEATURES AS POTENTIAL DIAGNOSTIC BIOMARKERS IN HEART FAILURE WITH LOW EJECTION FRACTION PATIENTS

• Main Author: Naufal, Dziban
• Format: Dissertations
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/78858
• Institution: Institut Teknologi Bandung

Heart failure is a condition where the heart is unable to pump enough blood or fill the heart chambers adequately due to changes in heart function and structure. Left ventricular heart failure examination is done using transthoracic echocardiography (TTE) by analyzing the ejection fraction value and other parameters in the left ventricle which are categorized as mechanical activities. These activities can be recorded with modalities other than ultrasound transducers, such as accelerometers as seismocardiogram (SCG) signals and microphones as phonocardiogram (PCG) signals. Both modalities’ recording might have potential diagnostic biomarkers for heart failure condition thus they able to eliminate the shortcomings of TTE, which depend on image quality and operator ability. Based on the mentioned potential, this dissertation research is designed to capture the biomechanical signals of the heart through SCG and PCG modalities, with the aid of ECG as the marker for systole and diastole period, as well as to perform feature extraction and analyze the contribution of features in distinguishing cardiac complexes in subjects with normal and low ejection fraction hearts. Support vector machine classification with ranked features as input are used to test feature significance. The CTI and dynamic feature groups are identified as having the most significant contribution in recognizing the presence of low ejection fraction heart conditions, which is the first hypothesis of this study. In addition, the utilization of multiple modalities performs better than single modality, which is the second hypothesis of the study. Ranking was done using an approach called Relief on 185 recognized features. Feature categories with high significance in classifying subjects with low and normal left ventricular ejection fraction are cardiac time interval (CTI), dynamics, and cepstral representation. Significant CTI features include PEP/LVET (R2), IVCT, LVET, and MPI (R1 in this study). Signal dynamics feature that show high significance is an SCG’s spectral entropy. iv The potential of the ranked feature set for becoming diagnostic biomarkers was confirmed using support vector machine-assisted classification. Classification was performed to determine the ability of the seismophonocardiogram (SPCG) feature set to detect the presence of systolic heart failure. The use of the features set in the subject oriented SVM classification can provide a maximum performance of 92%, 96%, 92%, 96%, and 99% for accuracy, precision, sensitivity, specificity, and AUC respectively, with the involvement of 121 features. The contribution of CTI in recognizing the presence of systolic heart failure can be explained physiologically because IVCT, LVET, and MPI are closely related to heart mechanical activity. Decreased heart contractility can be indicated by increased IVCT values and decreased LVET values, thereby increasing MPI values. The contribution of signal dynamics features can be understood by looking at this group of features as quasi-deterministic changes in physical values recorded in SCG and PCG, where both signals are generated from heart mechanical activity. Based on the classification results, the involvement of CTI and signal dynamics feature groups with a multiple modality scheme can be the answer to the construction of an affordable and highly accessible non-invasive systolic heart failure detection system.