Cardiogram Audio-Based Arrhythmia Detection Using a Hybrid CNN-GRU Model - Dalam bentuk buku karya ilmiah
DAWWI RAISSA DAMARJATI MULJANA

Informasi Umum

Kode

25.04.6616

Klasifikasi

000 - General Works

Jenis

Karya Ilmiah - Skripsi (S1) - Reference

Subjek

Deep Learning

Dilihat

125 kali

Informasi Lainnya

Abstraksi

Cardiovascular diseases (CVDs) are the main cause<br /> of global mortality, and heart rhythm disorders pose major<br /> diagnostic challenges, especially in environments limited by<br /> resources. To address this problem, we introduced a hybrid deep<br /> learning architecture that combines the Convolutional Neural<br /> Network (CNN) and the Gated Recurrent Unit (GRU), and<br /> classified a PCG heartbeat sound into normal and abnormal<br /> categories. CNN components extract spatial features from the<br /> sound of the heartbeat, while GRU models temporal patterns<br /> within the heart cycle. This fusion allows the system to effectively<br /> capture the structural and sequential characteristics of<br /> heartbeat signals. The data sets include 1,000 PCG recordings<br /> evenly divided into five classes: aortic stenosis, ventricular<br /> regurgitation, ventricular stenosis, ventricular valve collapse,<br /> and normal. In the binary classification (normal or abnormal),<br /> the weight loss function was used to solve class imbalances.<br /> Further techniques such as stratified partitioning, dropout, early<br /> stopping, and real-time validation have been used to improve<br /> model training stability and generalization. Methods of data<br /> augmentation, including noise addition, volume change, time<br /> expansion and spectral filtering, were implemented to simulate<br /> the variability of heart sounds in the real world. CNN-GRU<br /> achieved 97% percent test accuracy and 98% percent F1-<br /> scores, with 198 of 203 test samples correctly identified. The<br /> high sensitivity and specificity of the model highlight its ability<br /> to learn key cardiac features from raw PCG data. These<br /> results confirm the clinical feasibility of this model, which offers<br /> robust and scalable tools for detecting and improving diagnostic<br /> accuracy both in well-equipped and low-resource environments.

Koleksi & Sirkulasi

Tersedia 1 dari total 1 Koleksi

Pengarang

Penerbit

Sirkulasi

Download / Flippingbook