ConvNeXt based multimodal Feature Fusion for ECG classification

What is it about?

This study introduces a ConvNeXt-based multi-module feature fusion approach for accurate and early detection of cardiovascular disease (CVD). ECG signals from the MIT-BIH dataset are denoised using Discrete Wavelet Transform and balanced with SMOTE, then encoded into GASF, GADF, and MTF images. A 2D module employs ConvNeXt with CBAM for feature extraction, while a 1D module integrates CNN, SENet, and BiLSTM to analyze raw signals. The fusion of both modules enhances feature representation and improves classification performance.

Featured Image