基于联邦深度学习的皮肤病智能诊断研究OACSTPCD

Recently,intelligent diagnosis of skin lesion based on artificial intelligence has become an attractive topic in bio-medical field.However,due to the data scarcity in single institution,locally trained neural networks are difficult to meet the performance requirement of medical services.Traditional learning paradigm,which collects data from distributed institutions to train neural models centrally,has the risk of privacy leakage.To tackle these problems,in this paper,we propose a federated deep learning-based intelligent diagnosis algorithm for skin lesion.Specifically,compared with the centralized learning,federated learning is introduced to prevent privacy leakage when integrating multi-party data.Each institution sends the local model to the central server for aggregation,and then the central server synchronizes the global model obtained.Further,a correction mechanism is proposed to modify the cross-entropy loss for solving the problem of data heterogeneous in federated learning,whereby the model attention to heterogeneous data is increased by constraining the disparity between the local model and the global model,so as to reduce the impact of data heterogeneity on the diagnostic results.Finally,the experimental results demonstrate that the accuracy of proposed algorithm is 3%-4%higher compared with existing representative works,reaching 75.9%.