Purnamasari, Khalissa Shafadilla (2025) Deteksi Risiko Kematian Jantung Mendadak pada Fibrilasi Ventrikel Menggunakan Rompi ECG Berbasis Delineasi dan Convolutional Neural Network. Other thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 5023211014-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
5023211014-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (13MB) | Request a copy |
Abstract
Kematian jantung mendadak (Sudden Cardiac Death/SCD) merupakan kondisi fatal yang disebabkan oleh gangguan mendadak pada fungsi jantung, baik pada individu dengan maupun tanpa riwayat penyakit jantung. Sekitar 80% kasus SCD dipicu oleh aritmia ventrikel, khususnya fibrilasi ventrikel (VF), yang ditandai oleh perubahan morfologi kompleks QRS, panjang siklus, dan amplitudo sinyal ECG. Kondisi ini memerlukan penanganan yang cepat untuk meningkatkan peluang keselamatan pasien sehingga pengembangan teknologi pemantauan ECG berbasis wearable diperlukan untuk deteksi dini. Penelitian ini mengembangkan sistem deteksi dini risiko SCD berbasis wearable ECG berbentuk rompi dengan tiga elektroda sesuai kaidah Einthoven lead II, yaitu RA di dada kanan atas, LA di dada kiri atas, dan LL di bawah LA yang menempel langsung pada kulit. Sistem dilengkapi dengan Raspberry Pi 4B sebagai unit pemrosesan dan modul ADS1115 untuk mengubah sinyal ECG analog menjadi digital. Sinyal ditampilkan secara real-time dengan klasifikasi setiap satu menit berdasarkan data terkumpul. Data penelitian terdiri atas 1.335 data (primer dan sekunder). Data primer diperoleh dari 10 subjek (5 laki-laki dan 5 perempuan) dalam kondisi duduk santai. Sebanyak 1.024 data digunakan untuk pelatihan dan 311 data untuk pengujian. Sinyal ECG diproses melalui filtering, normalisasi, dan delineasi menggunakan Discrete Wavelet Transform (DWT) hingga level lima. Proses delineasi mendeteksi lima titik penting: Q onset, R peak, S offset, T peak, dan T offset. Enam parameter berupa amplitudo dan waktu dari masing-masing titik tersebut dijadikan input ke dalam model Convolutional Neural Network (CNN) tanpa ekstraksi fitur tambahan. Model menunjukkan performa tinggi dengan akurasi 98,7%, sensitivitas 98,1%, spesifisitas 99%, dan presisi 98,7%, sehingga sistem ini berpotensi menjadi solusi praktis, efisien, dan non-invasif untuk pemantauan risiko SCD secara dini dalam kehidupan sehari-hari.
======================================================================================================================================
Sudden cardiac death (SCD) is a fatal condition caused by a sudden disruption in heart function, both in individuals with and without a history of heart disease. Approximately 80% of SCD cases are triggered by ventricular arrhythmias, particularly ventricular fibrillation (VF), which is characterized by changes in QRS complex morphology, cycle length, and ECG signal amplitude. This condition requires rapid treatment to increase the chance of patient safety, so the development of wearable-based ECG monitoring technology is needed for early detection. This study develops a wearable ECG-based SCD risk early detection system in the form of a vest with three electrodes placed according to Einthoven lead II configuration: RA (right upper chest), LA (left upper chest), and LL (below the LA), directly attached to the skin. The system is equipped with a Raspberry Pi 4B as the processing unit and an ADS1115 module to convert the analog ECG signal into digital. The signals are displayed in real-time with classification every one minute based on the collected data. A total of 1,335 ECG samples (from both primary and secondary sources) were used, with 1,024 for training and 311 for testing. Primary data were collected from 10 subjects (5 males and 5 females) in a relaxed sitting position. ECG signals underwent filtering, normalization, and delineation using Discrete Wavelet Transform (DWT) up to the fifth level. The delineation process identified five key points: Q onset, R peak, S offset, T peak, and T offset. Six amplitude and time-based parameters from these points were used as input to a Convolutional Neural Network (CNN) model without additional feature extraction. The model achieved high performance, with 98.7% accuracy, 98.1% sensitivity, 99% specificity, and 98.7% precision. This system shows strong potential as a practical, efficient, and non-invasive solution for early SCD risk monitoring in daily life.
Actions (login required)
 |
View Item |