Research On Key Technologies Of Assisted Diagnosis Of Chest Pain Based On Machine Learning

Chest pain is the most common clinical disease in cardiology department.The chest pain cases rank first in hospital outpatient visits and accounts for 20% to 30% of emergency cases.The diversity of causes of chest pain leads to the difference of final diagnoses.Although chest pain can be accurately diagnosed by experienced doctors at present,there are often cases of missed diagnosis or misdiagnosis in emergency and out-of-home diagnoses in areas with low medical standards.Therefore,this paper uses the photoelectric plethysmography method to detect patients’ pulse signal,extracts patients’ heart rate after noise reduction treatment of the signal,and then uploads it to the auxiliary examination module of the medical record collection system.At the same time,personal information,past medical history and current medical history of patients are collected,and then a machine learning classification algorithm is used to predict the diagnosis of chest pain,and chest pain risk score is used to classify the chest pain-related risks using,which can provide a reference for the doctors’ final diagnosis and therefore improve the accuracy rate.The contents of this paper are as follows:1.The pulse signal acquisition module was designed to complete the acquisition of the patient’s pulse wave.Then noise reduction processing was performed mainly using the wavelet threshold denoising algorithm based on the feature analysis of the pulse signal.Finally,the heartbeat interval was obtained from the noise-reduced pulse signal using the threshold timing method to calculate the heart rate,aiming to improving the auxiliary examination in the chest pain medical record acquisition module.2.Chest pain history information acquisition module was designed based on the idea of general diagnosis of chest pain,which includes not only acquisition of personal information,current medical information,past medical history information,physical information,etc.,that are described by patients or relatives,but also the acquisition of auxiliary examination information that is completed in combination with pulse wave and clinical testing.The chest pain history information database was established,which aims to provide data for predictive analysis of chest pain causes and quantification of chest pain risk level.3.Three algorithms of machine learning,including decision tree,support vector machine,and K-nearest neighbor classification algorithm were used.In addition,ID3,C4.5,and CART algorithms were used for comparison with decision tree according to different division principles;linear kernel function,polynomial kernel function,sigmoid kernel function,and Gaussian kernel function are used for comparison with support vector machine according to different kernel functions;and Euclidean distance and cosine distance were used to compare with K-nearest neighbor calculation depending on the distance.The algorithms are compared in terms of accuracy,ROC curves,and detection time to select the optimal method,aiming to provide etiological options for the risk scoring module,aiming to provide a choice of etiology for the risk score module.4.The quantitative grading module of chest pain risk level was designed based on TIMI,HEART and GRACE risk scores,and the risk level of chest pain under different etiologies was judged according to cumulative scores,which was given low risk,medium risk and high risk outputs.The feasibility of the quantitative grading module of risk level was verified by real cases,and the univariate,binary and data correlation analysis of the existing data set was performed to screen out the risk factors of chest pain and expand the etiological types for each chest pain risk level.