Deep Learning-Based Multi-Lead ECG Classification with Short-Time Fourier Transform Attention

ECG classification is critical for diagnosing cardiovascular diseases. This study introduces a novel multi-lead ECG classification model incorporating Short-Time Fourier Transform Attention (STFTA) to enhance performance under complex conditions. STFTA effectively integrates information across both the frequency and time domains, significantly improving classification accuracy. The model was evaluated on the PTB-XL and CPSC2018 datasets. On the PTB-XL dataset, it achieved an F1 score of 0.784 and an AUC of 0.941, outperforming existing methods. On the CPSC2018 dataset, the model obtained an F1 score of 0.862, surpassing other advanced models. These results demonstrate the model's excellent performance in ECG classification, improving accuracy and generalization in cardiac disease diagnosis, and confirming the potential of STFTA in ECG signal processing.