Leveraging Convolutional Autoencoders for Bundle Branch Block Detection in Electrocardiograms

This paper introduces a novel approach for automated detection and classification of bundle branch blocks(BBB) in electrocardiograms(ECGs) using convolutional autoencoders(CAE). By integrating autoencoders with convolutional neural networks(CNN), our proposed CAE model efficiently learns essential features from high-dimensional input data, reducing dimensionality and enhancing diagnostic accu-racy. Leveraging the latent space representation learned by the autoencoder, our model effectively distinguishes between Normal, Left Bundle Branch Block(LBBB), and Right Bundle Branch Block(RBBB) ECG records, enabling rapid and precise diagnosis and treatment of cardiac conditions. This latent space encapsulates the essential characteristics of the input data, facilitating robust feature extraction and classification. The MIT-BIH Arrhythmia database provided standard ECG recordings, which are then extracted and normalized for each beat using an autoencoder and classified using CNN. It achieved an overall classification accuracy, F1-score, specificity, and sensitivity of 99.91%, 99.5%, 99.89%, and 99.01% respectively showcasing its robustness in distinguishing between these distinct cardiac conditions using the MIT-BIH dataset. The extracted latent space features are pivotal in enhancing the model's ability to discern distinct cardiac conditions with remarkable accuracy, surpassing traditional classification methods.