STCT: Spatial-Temporal Conv-Transformer Network for Cardiac Arrhythmias Recognition

AbstractCardiac arrhythmia, one of the primary causes of death, can be diagnosed using the electrocardiogram (ECG), a visual representation of heart activity. Existing methods can annotate ECG signals and achieve satisfactory results. However, they usually require prepossessing the ECG signals by extracting individual heartbeats before being fed into the recognition model. Then, the data is typically processed as a one-dimensional signal that leads to the loss of information. In this context, we propose a Transformer-based Spatial-Temporal Conv-Transformer (STCT) Network that deals with raw ECG data. The STCT considers the raw ECG signal as two-dimensional features by utilising the spatial and temporal information for more accurate identification of irregular heartbeat. Additionally, the model identifies parts of the feature space that are especially relevant and filters out the remainder, ensuring that STCT performance can be enhanced. To demonstrate the effectiveness and efficiency of the proposed method, four separate open ECG datasets, MIT-BIH, EDB, AHA, and NST, were used for the benchmarking. Ultimately, this method achieved 98.96%, 99.29%, 99.88% and 99.13% of accuracy and outperformed the state-of-art methods when compared.KeywordsECGDeep learningRNNCardiac arrhythmia