Título: Comparative Analysis of ECG Signal Feature Extraction Techniques for Cardiac Dysfunction Classification Using CNN.
Autores: maycon ribeiro, Victor Emanuel Paz Teixeira, Jés de Jesus Fiais Cerqueira, Juan Alberto Leyva Cruz, Eduardo Furtado de Simas Filho, Dhavidy De Almeida Silva & Jaimilton dos Santos Lima
Resumo: An advanced analysis of electrocardiogram signals, through feature extraction techniques such as Fourier transform (FFT), Hilbert-Huang (HHT), Discrete Wavelet Transform (DWT), and Mel Frequency Cepstral Coefficients (MFCCs), enables detailed characterization of the electrophysiological parameters associated with different cardiac dysfunctions. This paper compares these techniques to provide greater interpretability to the ECG-based prediction model. For this purpose, 197 signals extracted from the PhysioNet repository were used. The Convolutional Neural Network (CNN) model used a training set with 167 ECG signals. It had 30 ECG signals in the test set, which included 8 ARR, 7 MI, 10 NSR, and 5 CHF. The results indicated that the selection of the extraction technique should be guided by the particularities of the signal and the objectives of the application. Furthermore, the integration of these processing methodologies with machine learning algorithms has increased the accuracy of cardiac diagnoses and prognoses, consolidating it as a promising tool for clinical analysis.
Palavras-chave: Electrocardiogram; Feature Extraction; Convolutional Neural Network; Cardiac Dysfunction; Machine Learning.
Páginas: 8
Código DOI: 10.21528/CBIC2025-1173612
Artigo em PDF: CBIC_2025_paper1173612.pdf
Arquivo BibTeX:
CBIC_2025_1173612.bib