Atrial Fibrillation Classification Using a Deep Spectral Autoencoder

• Autores: Enrique Quezada-Próspero, Dante Mújica Vargas, Luis A. Cruz-Próspero, Christian García-Aquino, Ángel A. Rendón-Castro
• Localización: Computación y Sistemas (CyS), ISSN 1405-5546, ISSN-e 2007-9737, Vol. 29, Nº. 1, 2025, págs. 15-28
• Idioma: inglés
• Enlaces
  • Texto completo
• Resumen

  • Abstract: This study introduces a novel approach for atrial fibrillation classification using a deep autoencoder stacked with a fully connected softmax layer. The model is trained with spectral features extracted from ECG signals through spectral and signal analysis. The primary goal is to enhance existing algorithms in the state-of-the-art by delivering superior results with reduced computational cost. The PhysioNet Challenge 2017 database was used, which contains normal and atrial fibrillation rhythms. The signals were normalized before spectral feature extraction. These features were used to train the autoencoder, which performed additional feature extraction and dimensionality reduction. The resulting features were then used to train the fully connected layer responsible for classification. Performance was evaluated through quality metrics mentioned in the state-of-the-art using cross-validation to ensure robustness. The best median results obtained were: 99.7% Accuracy, 99.8% Precision, 99.5% Recall, 99.8% Specificity, 99.7% F1-score, 99.3% Matthews Correlation, and 99.3% Kappa.