Design And Implementation Of A Cloud-based Follow-up Service Platform For Cancer Patients In A Federated Learning Approach

Ovarian cancer is a type of disease with high morbidity and fatality.Its mortality rate ranks first among all gynecological tumors,which seriously threatens women’s life and health.At present,there is no that good treatment for ovarian cancer,so patients have to live with the disease.Therefore,in order for doctors to keep abreast of the patients’ condition after they leave hospital,it is urgent to construct a set of effective follow-up services.In addition,in the process of diagnosis and treatment,doctors often want to use artificial intelligence algorithms to assist in judging the condition of patients in order to improve work efficiency.However,due to the serious isolated data island problem in medical industry,especially in cancer,high-quality data that can be provided for algorithm training is very scarce.Data on follow-up platform is generated by real patients,so using these data for data analysis is a feasible way to help doctors grasp the patient’s condition comprehensively.According to real demands above,this thesis designs and implements a cloud follow-up service platform for cancer patients based on federated learning to meet the actual needs of doctors.The main work of this thesis can be divided into two aspects:1)After deep communication with clinicians,it is told that the current follow-up work is of little convenience.So this thesis takes ovarian cancer as an example and builds a cloud follow-up service platform suitable for a variety of cancers in an informationized way.Each medical institution can independently deploy the platform on their own edge nodes,connect to their hospital information systems and carry out follow-up work,which carries the cumbersome follow-up process from the We Chat group to the Internet and meets the actual needs of doctors effectively.2)This thesis also uses the federated learning framework FATE to implement a homomorphic logistic regression algorithm that can predict the tumorigenesis and propose a multi-party joint modeling scheme without sharing the original data.Furthermore,this thesis shows how to use the data from the follow-up platform as the input of the algorithm to model on the medical cloud brain,so as to get through the complete process through the real case of tumorigenesis prediction.In order to verify the effectiveness of the process,this thesis conducts a series of tests and analyses on the effectiveness of both the platform and the algorithm.The results show that the platform functions normally after deployment and can withstand the scale of users in the follow-up scenario,so it is a strong support for future ovarian cancer patients’ follow-up services.Meanwhile,the accuracy of the model trained by federated learning is only 0.1% lower than training all data on a single node,however 5%higher than training half of the data on the same single node.The effect is so remarkable that the model obtained through federated learning is capable of predicting work and effectively solves isolated data island problem.In summary,the topic of this article comes from the actual needs of the first-line doctors.Based on the FATE framework for secondary development,a cloud follow-up service platform for cancer patients is designed and implemented,and information technology is used to reduce the workload during the ovarian cancer follow-up services.This paper studies the federated learning algorithm for the sensitivity of medical data,and proposes to apply it to cancer prediction to solve the problem of isolated data islands between hospitals.This paper has also carried out platform function tests and performance tests,and evaluated the effect of the algorithm through experiments.The results show that the platform can meet the concurrent pressure of the user scale in the follow-up scenario,which will help build better artificial intelligence models to assist doctors in treating cancer patients and provide better follow-up services.