Intelligent Diagnostic Method For T2DM Combined With CHD Based On Multi-modal Medical Data

Objective To establish an intelligent analysis method based on the fusion of multiple sources of medical big data information to achieve automated diagnosis of type 2 diabetes combined with coronary heart disease(CHD)classification and diagnosis.Methods: A total of 699 patients who visited the cardiovascular department of Qingdao University Affiliated Hospital from January 2018 to December 2021 were selected as the research subjects.Among them,there were 395 male patients(56.5%)and304 female patients(43.5%),with ages ranging from 23 to 85 years old.According to whether they had CHD or not,they were divided into CHD group and non-CHD group;then the CHD group was further divided into T2 DM combined with CHD group and pure CHD group according to whether they had type 2 diabetes or not.Forty-eight multi-modal clinical data items were extracted from electronic medical records based on T2 DM combined with CHD risk factors and guidelines for managing cardiovascular disease.Significant feature indicators were selected by significance test method.The original data was split into training set and validation set in a ratio of 7:3,using the training set to train deep fuzzy neural network model,which was compared with support vector machine(SVM),logistic regression(LR)and decision tree(DT)models in terms of model performance evaluation using accuracy rate,precision rate,recall rate,F1-score,and area under ROC curve(AUC).Results: A total of sixteen significant feature indicators have been screened out through significance analysis tests.The deep fuzzy neural network classification model established by these sixteen features has an accuracy rate of89.5%,precision score is89.96%,recall score is89.58%,F1-Score0.903,The AUC value for T2 DM combined with CHD group is 0.893,while the AUC value for the pure CHD group is 0.862 and the AUC value for the non-CHD group is 0.929.which are all better than SVM、LR、DT.Conclusion: The deep fuzzy neural network model constructed based on multi-modal medical examination data can achieve complementary information analysis among various types of data sources,reveal mutual interactions between different modalities,improve integrity utilization efficiency,the relatively strong diagnostic performance demonstrates that this method could be applied effectively for diagnosing type II diabetes mellitus complicated with coronary heart disease.