AUTOMATED DETECTION OF ATRIAL FIBRILLATION USING DEEP LEARNING

Bio Medical and Computer Science Engineering

The condition of atrial fibrillation (Afib) involves irregular beating of the heart's upper chambers (atria), which can increase the risk of stroke caused by a blood clot. The identification of paroxysmal AF can be improved through prolonged cardiac monitoring. To identify AF beats in Heart Beat (HR) signals, a machine learning model was employed, where the dataset is divided into sliding windows of 100-beat sequences. These sequences are then fed into a model that comprises a Bi-Directional LSTM layer, a Global max pooling layer, a Dense layer, a Dropout layer, and an output layer. The model was trained and tested using the MIT-BIH Atrial Fibrillation Database. The approach achieved high accuracy rates during training and validation, with a 98.15% accuracy rate. Additionally, the 7-fold cross-validation on 20 subjects yielded an accuracy rate of 93.43%, while testing with unknown data from 3 subjects resulted in an accuracy rate of 99.2%. The model performed well on untrained data, as demonstrated by the complete setup. The neural network architecture used in the proposed model was straightforward and consisted of simple deep-learning layers. Moreover, the proposed model demonstrated better efficiency.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.