Research On T-wave Morphological Classification Algorithm Based On Machine Learning

With the rapid development of science and technology,people’s living standards are constantly improving,physical activity is gradually reduced,weight gain and long-term irregular work and rest factors lead to an increasing incidence of cardiovascular diseases.Because of its safety,non-invasive and easy operation,ECG has become the basic means of cardiovascular diseases.Each part of ECG reflects the ventricular electrophysio logical activities of myocardium in different periods.With the promotion of wearable ECG monitoring devices,it’s not realistic to identify all ECG signals manually due to the generation of huge amount of data.Automatic detection and analysis of ECG has become a necessary means for prevention and diagnosis of cardiovascular diseases.As one of the five major ECG waves,T wave represents the repolarization process of ventricular muscle,and its morphological change is a key indicator of some pathology,so the morphological classification and recognition of T wave is very important in the clinical diagnosis of ECG.This paper analyzed and processed the normal T wave and 5 types of abnormal T wave common in clinical practice(Inverted,Peak,Biphasic,Low and Bimodal),and established the T wave detection and classification model by using the method of traditional machine learning and deep learning,hoping to be applied in wearable medical treatment to complete the recognition of T wave in ECG signal.The main research contents of this paper are as follows:(1)T-wave morphological classification based on time domain features.Seven time-domain features of T-wave morphology were extracted.The Gaussian kernel support vector machine(RBF-SVM)was used as classifier,and the grid search method is used as the super parameter optimization algorithm of the network,a model to recognize T waves of different morphology is built.The average classification accuracy of the model on the test set was 90.61%,and the classification sensitivity of the 6 types of T waves were 93.37%,92.57%,88.59%,79.96%,83.89%and 77.53%,respectively.(2)Study on T wave diagnosis model based on improved frequency slice wavelet transform and convolutional neural network(CNN).The one-dimensional ECG signal fragments were modified by frequency slice wavelet transform to obtain the two-dimensional time-frequency graph.As a comparative experiment,the waveform image of the candidate T-wave segments in the time domain and the energy image in the time-frequency domain are taken as the input of CNN to train the model,respectively.The classification accuracy of the model for waveform images was 87.79%,and the classification sensitivity of 6 types of T waves were 90.20%,82.50%,95.29%,81.46%,81.19 and 84.50%.The classification accuracy of the model for energy images was 97.4%,and the classification sensitivity of 6 types of T waves were 96.43%,97.05%,99.37%,95.51%,99.19%and 97.30%,respectively.