Research Of Skin Cancer Federated Detection Model

Skin cancer has an adverse impact on healthy life of people because of it is difficult to find and treat.In recent years,the development of artificial intelligence and other emerging technologies has promoted the intelligent detection of skin cancer,it provides a more simple and efficient way for skin cancer detection,and to a certain extent alleviated the misdiagnosis problem,which caused by traditional skin cancer detection such as doctors observe pathological pictures and so on.However,the existing studies seldom consider the data privacy of skin cancer patients,which makes it difficult for most studies to be applied in real life.At the same time,when solving the small sample data faced by skin cancer,the quality of the generated image is rarely considered from multiple angles.Therefore,in this paper,the problems existing in skin cancer detection research are analyzed,and based on the innovative research of federated learning and dual generation adversarial network(Dual GAN),a skin cancer federated detection framework combined with dual generation adversarial network is proposed,which not only ensures the detection rate of skin cancer,but also protects the data privacy of patients.At the same time,to further reduce the time required for global model training,starting from the communication cost and local resources,an efficient skin cancer federated detection method and an efficient adaptive aggregation skin cancer federated detection method are designed.The main work of this paper is as follows:To solve the problems of data privacy and small sample data faced by skin cancer data,federated learning is applied to skin cancer detection,and the Dual GAN is introduced,a skin cancer federated detection framework combined with dual generation adversarial network is designed,which can not only protect data privacy of patients but also effectively detect skin cancer symptoms.Firstly,to improve the quality of the generated image,the performance of the Dual GAN is considered from many angles,and the training process of the Dual GAN is transformed into a many-objective optimization problem,which is solved by knee point-driven evolutionary algorithm(Kn EA).Then,to improve the overall performance of the global model in skin cancer federated detection,by optimizing the structure of the global model and transforming it into a many-objective optimization problem,and a many-objective skin cancer federated detection model is designed.It is solved by Kn EA to balance the conflict between global model accuracy,loss,area under ROC curve and communication cost.Finally,by comparing the performance of federated method and non-federated method under different conditions,the experimental results show that the proposed method is effective.To reduce the communication cost of skin cancer federated detection and improve the efficiency of model,an efficient method of skin cancer federated detection is proposed.Inspired by the fitness prediction strategy,the traditional federated average algorithm is improved,it dynamically divides the advantages and disadvantages of the client by setting the accuracy threshold,and a new federated aggregation method is designed to realize the influence of the superior client on the inferior client,which improve the convergence speed of the global model,and then reduce the number of communications.Experiments are carried out to compare different federated aggregation methods on multiple datasets to prove that the proposed method can improve the overall performance of the skin cancer federated detection model.To reduce the waste of local resources in federated learning and further reducing the time needed of training the global model,an efficient adaptive aggregation skin cancer federated detection method is proposed.The client is divided into completed local training and uncompleted local training,and considering the difference of accuracy and degree of completion,a new method of adaptive federated aggregation is designed to further improve the efficiency of skin cancer federated detection model.Finally,through extensive experiments on different datasets,the experimental results are analyzed to verify the effectiveness of the proposed method in the detection of skin cancer.