Sistem Multitask Classifier Interpretable Deep Learning untuk Penyakit Katup Jantung Menggunakan Sinyal Seismocardiography

Wibowo, Dasha Rhenaldi Arne (2026) Sistem Multitask Classifier Interpretable Deep Learning untuk Penyakit Katup Jantung Menggunakan Sinyal Seismocardiography. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5023221029-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (34MB) | Request a copy

Abstract

Penyakit Katup Jantung (Heart Valve Disease atau HVD) merupakan penyebab utama mortalitas kardiovaskular global, namun metode deteksi dini standar emas seperti echocardiography memiliki biaya tinggi dan aksesibilitas yang terbatas. Metode alternatif seperti auskultasi seringkali memiliki sensitivitas yang rendah. Seismokardiografi (SCG) menawarkan solusi non-invasif untuk menilai mekanika jantung, namun morfologi sinyalnya kompleks dan sulit diinterpretasi menggunakan metode konvensional. Penelitian ini mengusulkan pengembangan sistem berbentuk chest strap berbiaya rendah yang mengintegrasikan sensor SCG dan Photoplethysmography (PPG) menggunakan mikrokontroler STM32 dan Raspberry Pi. Sistem ini menerapkan metode Interpretable Deep Learning (HVDNet) yang dilengkapi mekanisme Self-Attention untuk mengatasi masalah "black box" pada kecerdasan buatan medis. Segmentasi sinyal dilakukan melalui dua pendekatan, yaitu menggunakan referensi PPG dan metode dekomposisi Successive Variational Mode Decomposition (SVMD). Sistem berhasil membuktikan potensial penggunaan HVDNet untuk klasifikasi menggunakan subject split dengan F-1 Score sebesar 82.80%, mengaplikasikan segmentasi tanpa sinyal ECG melalui metode shift dengan rata-rata korelasi tertinggi puncak AO dan puncak sistolik sebesar 0.86% yang menggunakan deteksi AO dan puncak sistolik PPG, serta mendesain dan mengambil data terhadap 45 subjek sehat dan sakit di ITS dan RSUD Dr. Soetomo.
====================================================================================================================================
Heart Valve Disease (HVD) is a leading cause of global cardiovascular mortality, yet gold-standard early detection methods such as echocardiography come with high costs and limited accessibility. Inexpensive alternative methods like auscultation often suffer from low sensitivity. Seismocardiography (SCG) offers a non-invasive solution to assess cardiac mechanics, but its signal morphology is complex and difficult to interpret using conventional methods. This study proposes the development of a low-cost, chest strap-based system that integrates SCG and Photoplethysmography (PPG) sensors using STM32 and Raspberry Pi microcontrollers. The system implements an Interpretable Deep Learning method (HVDNet) equipped with a Self-Attention mechanism to address the "black box" problem in medical artificial intelligence. Signal segmentation is conducted through two approaches: using a PPG reference and the Successive Variational Mode Decomposition (SVMD) method. The system successfully demonstrates the potential of using HVDNet for classification using a subject split with an F-1 Score of 82.80%, applying ECG-signal-free segmentation via a shift method with the highest average correlation between the AO peak and systolic peak at 0.86% utilizing AO detection and the PPG systolic peak, as well as designing and collecting data from 45 healthy and patient subjects at ITS and RSUD Dr. Soetomo.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Seismocardiography, Photoplethysmography, Interpretable Deep Learning, Penyakit Katup Jantung, Chest Strap, Heart Valve Disease
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5102.9 Signal processing.
Divisions:	Faculty of Electrical Technology > Biomedical Engineering > 11410-(S1) Undergraduate Thesis
Depositing User:	Dasha Rhenaldi Arne Wibowo
Date Deposited:	31 Jul 2026 04:15
Last Modified:	31 Jul 2026 04:15
URI:	http://repository.its.ac.id/id/eprint/140678

Actions (login required)

View Item