Pemodelan Entropy Wavelet Sinyal Phonocardiogram untuk Klasifikasi Murmur Jantung pada Anak dengan Kolmogorov-arnold Network

Armelya, Haura (2026) Pemodelan Entropy Wavelet Sinyal Phonocardiogram untuk Klasifikasi Murmur Jantung pada Anak dengan Kolmogorov-arnold Network. Other thesis, Institut Teknologi S.

Text 5049221010-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Penyakit Jantung Bawaan (PJB) merupakan salah satu penyebab utama morbiditas dan mortalitas pada anak, namun masih sering terlambat terdeteksi, khususnya pada layanan kesehatan primer. Pemeriksaan auskultasi jantung konvensional sangat bergantung pada pengalaman dan subjektivitas pemeriksa, sehingga diperlukan sistem bantu skrining murmur jantung yang objektif, efisien, dan dapat dijelaskan secara matematis. Penelitian ini mengembangkan sistem klasifikasi murmur jantung anak berbasis analisis sinyal phonocardiogram (PCG) menggunakan fitur entropy wavelet dan model Kolmogorov–Arnold Network (KAN). Sinyal PCG dari dataset CirCor DigiScope dipra-proses melalui bandpass filtering 20–600 Hz dan segmentasi, kemudian didekomposisi menggunakan wavelet Daubechies orde 4 hingga level lima untuk menghasilkan enam subband, yaitu A5 dan D1–D5. Nilai Shannon entropy dari masing-masing subband digunakan sebagai representasi utama kompleksitas sinyal dan disusun menjadi vektor fitur dasar. Untuk meningkatkan daya representasi tanpa mengubah konsep dasar fitur, diterapkan pengayaan berupa log-energy, rasio antar subband, serta pendekatan multi-lokasi auskultasi. Model KAN dilatih menggunakan fungsi kerugian Binary Cross-Entropy with Logits dengan penyesuaian bobot kelas pada skema stratified train–test split 80:20. Hasil analisis menunjukkan bahwa nilai entropy wavelet pada subband frekuensi menengah hingga tinggi cenderung lebih tinggi dan lebih bervariasi pada sinyal murmur dibandingkan sinyal jantung sehat, meskipun masih terdapat tumpang tindih distribusi antar kelas. Evaluasi performa pada data uji menghasilkan nilai ROC-AUC sebesar 0,73 dan PR-AUC sebesar 0,60, yang menunjukkan kemampuan model dalam membedakan kelas pada kondisi data tidak seimbang. Secara keseluruhan, kombinasi entropy wavelet enam subband dan model KAN mampu merepresentasikan perbedaan kompleksitas sinyal PCG anak secara kuantitatif dan berpotensi menjadi dasar pengembangan sistem bantu skrining murmur jantung yang ringan, mudah diinterpretasikan, dan sesuai untuk layanan kesehatan primer.
==================================================================================================================================
Congenital Heart Disease (CHD) is one of the leading causes of morbidity and mortality in children, yet it is often detected late, particularly in primary healthcare settings. Conventional cardiac auscultation relies heavily on the examiner’s experience and subjective judgment, highlighting the need for an objective, efficient, and mathematically interpretable heart murmur screening system. This study develops a pediatric heart murmur classification system based on phonocardiogram (PCG) signal analysis using wavelet entropy features and a Kolmogorov–Arnold Network (KAN) model. PCG signals from the CirCor DigiScope dataset were preprocessed using bandpass filtering in the 20–600 Hz range and signal segmentation. The signals were then decomposed using a Daubechies wavelet of order 4 up to level five, producing six subbands (A5 and D1–D5). Shannon entropy values from each subband were used as the primary representation of signal complexity and organized into a base feature vector. To enhance representational capacity without altering the core feature concept, additional features based on log-energy, inter-subband ratios, and a multi-auscultation location approach were incorporated. The KAN model was trained using a Binary Cross-Entropy with Logits loss function with class-weight adjustment under a stratified 80:20 train–test split scheme. The analysis results indicate that wavelet entropy values in the mid- to high-frequency subbands tend to be higher and more variable in murmur signals compared to healthy heart sounds, although some overlap in feature distributions between classes remains. Performance evaluation on the test set yielded a ROC-AUC of 0.73 and a PR-AUC of 0.60, demonstrating the model’s ability to discriminate between classes under imbalanced data conditions. Overall, the combination of six-subband wavelet entropy features and the KAN model effectively capture quantitative differences in pediatric PCG signal complexity and shows promise as a foundation for developing lightweight, interpretable heart murmur screening systems suitable for primary healthcare applications.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Phonocardiogram, Entropy Wavelet, Kolmogorov-Arnold Networks, Murmur Jantung, Pediatri. Phonocardiogram, Wavelet Entropy, Kolmogorov-Arnold Networks, heart murmur, Pediatric.
Subjects:	R Medicine > RJ Pediatrics R Medicine > RJ Pediatrics > RJ101 Child Health. Child health services
Divisions:	Faculty of medicine and health (MEDICS) > Medical Technology > 11503-(S1) Undergraduate Thesis
Depositing User:	Haura Armelya
Date Deposited:	03 Feb 2026 02:35
Last Modified:	03 Feb 2026 02:35
URI:	http://repository.its.ac.id/id/eprint/131531

Actions (login required)

View Item