Research on Arrhythmia Classification Algorithm Based on ECG-BERT Model with Temporal Semantic Enhancement

The research of arrhythmia classification algorithms is faced with the problems of huge amounts of data, less effective information of feature extraction, and the accuracy of classification and recognition not being high enough. Electrocardiogram (ECG) is a non-invasive diagnostic tool for arrhythmia, so it is very important for accurate and intelligent classification and interpretation of ECG. This study proposes an ECG-BERT model algorithm based on time sequence enhancement, using MIT-BIH arrhythmia clinical standard database. The arrhythmia data is removed by wavelet transform, and the ECG signal is reconstructed into a time sequence-related pseudo text sequence by adaptive segmentation strategy. The multi-layer Transformer structure of the Bert model is used to automatically fuse the low-level local features with the high-level global features, and capture the previous and subsequent information. In addition, the parallel computing ability of the transformer is more suitable for processing long-sequence ECG data. Finally, the accuracy of classification is 0.9874, the F1 score is 0.933, and the precision of S, V, F, and Q classification is above 0.99 Good results, to promote the development and application of ECG-assisted diagnosis.