Heterogeneous feature fusion based machine learning strategy for ECG diagnosis

Cardiovascular diseases, as a serious threat to life and health globally, its high misdiagnosis rate has been a challenge in ECG diagnosis. This study is dedicated to improving the accuracy and efficiency of ECG diagnosis through the introduction of artificial intelligence techniques. In this study, we innovatively designed a feature extraction framework named BiAE that combines the advantages of bi-directional long and short-term memory networks (BiLSTM) and autoencoder to effectively extract rich feature information from raw ECG signals. Meanwhile, a large number of high-dimensional features were automatically extracted from the time series data using tsfresh. Some of these key features (e.g., BiAE_Feature22, BiAE_Feature65, and BiAE_Feature45) play a significant role in the time and frequency domain variations of ECG signals, and show unique advantages in global signal identification, QRS wave cluster detection, T-wave analysis, and extreme abnormal signal capture, respectively. Ten machine learning models including support vector machines were subsequently employed for ECG signal classification into five specific categories such as Normal Beat, Unclassifiable Beat, Premature Ventricular Contraction (PVC), Premature or Ectopic Supraventricular Beat (SVPE), and Fusion of Ventricular and Normal Beat (FUSION). Through cross-validation and performance evaluation, the support vector machine (SVM) was finally identified as the optimal model with an accuracy of 96%. Artificial intelligence-assisted ECG diagnosis can significantly improve the efficiency and accuracy of ECG diagnosis, which is expected to provide strong support for early screening and accurate diagnosis of cardiovascular diseases.