Research On Automatic Identification And Intelligent Typing Of Coronary Artery Disease Based On Deep Forest Using ECG Signals

According to the Chinese Health Statistics Yearbook in 2019,the prevalence and mortality of coronary artery disease(CAD)in urban and rural residents in China has been rising.In 2018,the mortality rate of CAD in urban residents in China was 120.18 / 100,000,and that of rural residents was 128.24 / 100,000.Early diagnosis of coronary artery disease can effectively curb the deterioration of the disease and reduce mortality.The diagnosis of coronary artery disease can be divided into two steps: identification and typing.Based on this research background,in order to solve the problems of poor generalization of artificially designed features and large demand for training data in the current automatic CAD diagnosis and typing algorithms,this paper builds an automatic diagnosis model of CAD based on deep forests to automatically extract effective features And using a small amount of training data to obtain the optimal model.Based on the effective diagnosis of CAD,in view of the different treatment methods of CAD,an intelligent typing algorithm for CAD based on the weighted average deep forest is proposed.By improving the feature fusion method of deep forest,the effectiveness of the proposed features is improved,and the model calculation amount is reduced.The specific research contents are as follows:1.An automatic identification algorithm for CAD based on deep forest is proposed.An appropriate size sliding window is constructed for multi-grained scanning,and the effect of data volume on the accuracy of the algorithm is avoided through data enhancement.Combined with completely random forests and random forests with different selection methods for splitting attributes,automatic extraction of effective features is achieved.Gain comparison is set for each cascade to reduce parameter adjustment.Three sets of experiments with 90%,70%,and 30% of the training set were set,and the identification accuracy was 99.86%,99.81%,and 99.58%,respectively.In addition,in consideration of the clinical situation where the patient lacks previous data,training data and test data are set up from experiments with different subjects,and the identification accuracy is 99.98%.Experiments prove that the proposed algorithm can effectively realize the automatic diagnosis of coronary heart disease,which is of great significance for the early diagnosis and screening of coronary heart disease in clinical practice.2.An intelligent typing algorithm for CAD with weighted average deep forest is proposed.In view of the high efficiency requirements of CAD classification,different weights are applied to each subtree of the random forests in the cascade,which reduces the negative impact of weaker recognition subtrees on the entire model and improves the effectiveness of the feature.Improve the cascaded feature fusion method.After averaging multiple class vectors generated by each cascade,they are aggregated with the original feature vector as the input of the next cascade.Problem of information redundancy and model space complication.Three sets of experiments with 90%,50%,and 30% of the training set were set up,which achieved classification accuracy of 98.94%,98.21%,and 97.78%,respectively.Therefore,the proposed algorithm can effectively realize the intelligent typing of CAD.