Heart Disease Prediction From Varied Length ECG Data Using Bi-Directional Recurrent Attention Network

Introduction: Heart ailments are one of the foremost causes of global mortality; thus, an early and accurate diagnosis is paramount. ECG is a common method for evaluating heart conditions, often considered lacking prognostic accuracy. To support diagnostic accuracy, machine learning and AI are beginning to be used for automated ECG analysis. Objectives: The major objective is to develop and evaluate a new deep learning-based hybrid model that can leverage better classification accuracy and computational efficiency. Methods: In this research article, a deep learning approach is presented that integrates convolutional neural networks (CNNs) for feature extraction from the signal into a sequential fashion by using bidirectional long short-term memory (Bi-LSTM) and gated recurrent units (GRU) supplemented with attention mechanisms. A hybrid model is proposed and tested on the ECG Heartbeat Classification Dataset (MIT-BIH), which is known as the Bidirectional Recurrent Attention Network (BRAN). Results: With an astonishing accuracy of 99.65%, the proposed BRAN model surpasses existing methods in both classification accuracy and computational efficiency. This showcases the efficacy of merging CNN with recurrent architectures and attention mechanisms in ECG-based heart disease detection. Conclusions: The study confirms that deep learning models, especially the BRAN hybrid type, greatly enhance the recognition of heart disease from ECG data. The findings emphasize AI-assisted diagnosis inheriting higher accuracy and automation capability, which can help the medical personnel in turning towards early diagnosis and treatment. Real-time implementation and further validation over multiple datasets can be a consideration for future research.