Convolutional Neural Network Fused With Recurrent Network for ECG-Based Detection of Hypertrophic Cardiomyopathy.

Background: Hypertrophic Cardiomyopathy (HCM) affects the left ventricle of the heart, leading to thickening of the ventricular wall and potentially life-threatening conditions, such as atrial fibrillation, cardiac failure, and sudden death. Early and accurate detection of HCM from Electrocardiogram (ECG) signals is critical for reducing mortality risk. However, most existing methods fail to simultaneously capture spatial and temporal patterns in ECG data, resulting in reduced diagnostic reliability.

Method: This paper proposes a hybrid Deep Learning (DL) network for detecting HCM using the ECG. Initially, input ECG signals are forwarded for pre-processing by Kalman filter. Then, processed signal is fed to feature extraction phase for extracting Empirical Mode Decomposition (EMD), statistical and medical features, which is followed by feature fusion, wherein the optimal feature is merged by Deep Belief Network (DBN) with Jensen-Shannon distance. Moreover, Convolutional Neural Network Fused with Recurrent Network (CNNFRN) performs HCM detection and final detected output is effectively achieved. The proposed CNNFRN combines Kalman Neural Network (CNN) and Recurrent Neural Network (RNN) based on regression modelling. Finally, the model is trained under a supervised framework using the Adam optimizer.

Results: The proposed model is validated using the PTB Diagnostic ECG Database and Shaoxing and Ningbo Hospital ECG Database. The results show that the proposed CNNFRN model achieved an accuracy of 0.940, sensitivity of 1.000, specificity of 0.913, and an F1-score of 0.956. These findings confirm the model's effectiveness in robust and early detection of HCM, offering significant clinical value.

Conclusion: The proposed model accurately detects HCM by combining advanced feature extraction and a hybrid deep learning approach that captures both spatial and temporal ECG patterns. It also shows strong performance and reliability across multiple databases, making it valuable for early and effective clinical diagnosis.