Deep Learning Model Management For Coronary Heart Disease Early Warning Research

Coronary Heart Disease(CHD)is one of the common diseases that threaten people’s health and life.Deep learning method based CHD early warning research is becoming the state-of-the-art method.It contains plenty of experiments during the process of establishing and optimizing the CHD early warning models,including the adjustment of hyperparameters,the network and the training data of deep learning models.And then pick the optimal model out by the evaluation of the models.After some investigation,file systems and spreadsheets are usually used to manage the data generated in the modelling lifecycle,which causes a rapid increase of the storage requests and inconvenience for researchers to compare and analyze the difference between experiments.So it is valuable to study the data management of deep learning model in early warning modeling of CHD.This thesis proposes the conception of deep learning model version and design a kind of version code mechanism to build the relationship among the model versions.A deep learning model version tree is established for each kind of model.TrackingAncestors algorithm and Find-Specified-Ancestor algorithm are designed to conduct the lineage management of the deep learning model.According to the characteristics of the deep learning model data generated by each experiment,we propose the SMR(Samples-Models-Results)data links conception to describe the relationship among the training data,model and the model evaluation result.Then we build the conceptual model and logic schema for CHD patients’ data and deep learning model data respectively.Design and implement the query algorithm about SMR data links.Considering the big data characteristics of the patients’ data,we compare the query response time based on the implement of MySQL database and MongoDB database.Finally,we design the architecture for the CHD patients’ data and the deep learning model data management,and implement the prototype system called Cdmdcms(Cohort Data&Deep Learning Model Data,Collaboration Management System),provide services including models selection,model comparison and model data visual exploration service to support the researchers in data model analysis.