eFuseNet: A deep ensemble fusion network for efficient detection of Arrhythmia and Myocardial Infarction using ECG signals

Amitesh Kumar Dwivedi,Gaurav Srivastava,Sakshi Tripathi,Nitesh Pradhan

DOI: https://doi.org/10.1007/s11042-024-19740-5

IF: 2.577

2024-07-14

Multimedia Tools and Applications

Abstract:Myocardial infarction (MI) and cardiac arrhythmias remain leading causes of mortality within cardiovascular diseases. Accurate and timely diagnosis using electrocardiogram (ECG) signals is crucial, yet manual ECG interpretation is prone to error and time-consuming. Addressing these challenges, the authors propose a novel deep neural network architecture for efficient and accurate detection and classification of both MI and arrhythmia from ECG signals. Our model strategically fuses a modified AlexNet for robust feature extraction with Long Short-Term Memory and Gated Recurrent Units networks to exploit the temporal dependencies inherent in ECG data. Fusion techniques combine the insights from these diverse sub-models, aiming to reduce generalization errors and surpass the performance of single-model approaches. Extensive experimentation on the MIT-BIH and PTB datasets demonstrates superior accuracy (98.51% and 99.97%, respectively) and reduced training time compared to other pre-trained networks. Robustness is validated through 10-fold cross-validation, yielding average accuracies of 98.76% and 99.48%. Our findings highlight the potential of this computationally efficient ensemble model for practical clinical implementation, enabling more accurate and timely diagnosis of life-threatening cardiovascular conditions.

computer science, information systems, theory & methods,engineering, electrical & electronic, software engineering