Research And Development Of Medical Data Model Training Platform Based On Searchable Encryption

Nowadays,the world is becoming more and more electronic information-based,in which the emergence of electronic medical record is a cross era performance.It not only improves the efficiency of hospitals,but also greatly facilitates patients.At the same time,machine learning has developed rapidly,and can help workers in all walks of life to better process and analyze data.Therefore,now many hospitals hope to use machine learning algorithm to train the model according to the patient’s electronic medical record data stored in the hospital,so as to get a model that can assist doctors to judge whether patients have a certain disease.However,when hospitals share electronic medical records for machine learning,their patients will involve sensitive personal privacy issues,so it is necessary to share data on the premise of ensuring privacy security.When requesting different disease models,it is necessary to select different features for training,because not all features in a complete electronic medical record are related to a disease.If there are redundant and irrelevant features involved in training,it will seriously affect the accuracy of the model.Moreover,different users sometimes request the same model.If the server cannot balance the processing well,it will consume additional performance.Therefore,how to carry out efficient and flexible machine learning model training on the premise of ensuring the security of electronic medical record information is an urgent problem to be solved.In view of this background,based on searchable encryption technology and homomorphic encryption technology,this topic has developed a medical data model training platform based on searchable encryption.The main work is as follows:(1)A model training framework based on searchable encryption is designed.Homomorphic encryption technology is used to encrypt data,and symmetric searchable encryption technology is used to generate blind indexes and trapdoors.The server can perform ciphertext data search through blind query trap and index,and can use the searched homomorphic ciphertext for model training.This framework significantly improves the flexibility of training and the fit of the model.It also ensures the security of dynamic data management(2)The sigmoid function is approximated by higher-order polynomial,which effectively solves the defect that homomorphic ciphertext can not calculate nonlinear function,so that the server can perform nonlinear model training on ciphertext;At the same time,the homomorphic ciphertext operation is optimized,which improves the accuracy of the ciphertext model and reduces the training time of the model.(3)In the model training framework,the server locally uses the header of the query trap and the storage address of the ciphertext model to build an automatically updated ciphertext model table.On the basis of ensuring the effectiveness of the model,it greatly improves the efficiency of obtaining the model,effectively prevents redundant training when different users send the same request,and reasonably allocates the resources in the server.(4)Build and apply the platform to the real environment.This platform can realize the addition,deletion,modification,query and model training of medical data on the premise of ensuring the privacy and security of user data.