Arrhythmia Recognition Based on Multimodal Neural Network

Cardiovascular disease has always been the highest morbidity and mortality of non-communicable diseases, arrhythmia is a relatively common cardiovascular disease. Traditional ECG signal classification methods require manual feature extraction, and feature extraction from a single model is insufficient, so it cannot maintain a high recognition rate for various arrhythmias. Premature beats need to be judged by the connection between beats, and only using CNN model will cause low sensitivity. In this paper, a model combining CNN and LSTM was proposed. CNN extracted local features and LSTM captured the dependence relationship between heart beats before and after, realizing the recognition of 15 arrhythmias with an accuracy of 98.15% and sensitivity of all arrhythmias above 90%. Compared with the single model, the dual-mode model not only improves the accuracy of arrhythmia identification, but also improves the sensitivity of some special arrhythmias.