BioU-Net: Diagnosis Network Based on Spectral Feature Enhancement for Myocardial Infarction

Myocardial infarction (MI) is a dangerous cardiovascular disease. Electrocardiogram (ECG), as a non-invasive testing tool, plays an important role in the diagnosis of cardiovascular diseases. In recent years, deep learning technology has provided new opportunities for automated diagnosis of heart diseases. This study aims to make up for the shortcomings of existing deep learning methods and fully mine the spatial and spectral information of ECG signals to improve diagnostic accuracy. First, we designed the AGF (Attention-Guided Fourier convolution) block, which combines the attention mechanism of SENet and fast Fourier convolution. Secondly, we proposed BioU-Net, which can process ECG signals in both directions and achieve multi-scale information fusion through skip connections. Experiments on the latest PTB-XL dataset prove that BioU-Net significantly improves the performance of the tasks of MI diagnosis. These innovative contributions enable our model to have a potential clinical impact in the medical field, providing new possibilities for early diagnosis and treatment of heart diseases.