Privacy Preserved Federated Learning for Skin Cancer Diagnosis
Yifan Li, Yuechen He, Yu‐Fei Fu, Sizhe Shan · 2023
In recent years, the use of artificial intelligence to assist the medical industry in making more accurate diagnoses has become popular. To achieve a more accurate diagnosis, more data is often needed to constantly improve the model. However, when data needs to be shared by patients from different hospitals, patients may be reluctant to share their privacy with unknown institutions that they do not trust. This situation creates a Data Silo problem that hinders progress in AI diagnostics. Therefore, in this paper, we propose to use the Federated Learning framework to train models with private patient data on local clients and aggregate models on a central server, allowing participants to jointly train models without sharing data and technically breaking down Data Silos and enabling AI collaboration. Experiments demonstrate that compared to traditional CNN training, the CNN model base on the Federated Learning framework sacrifices 8% of accuracy to accomplish the above requirements. Due to the inherent nature of the distributed system, the training time and node communication time is also favorable.