Federated learning-based colorectal cancer classification by convolutional neural networks and general visual representation learning

摘要

Colorectal cancer is the fourth fatal disease in the world, and the massive burdenon the pathologists related to the classification of precancerous and cancerouscolorectal lesions can be decreased by deep learning (DL) methods.However, the data privacy of the patients is a big challenge for being able totrain deep learning models using big medical data. Federated Learning is a risingstar in this era by providing the ability to train deep learning models on differentsites without sacrificing data privacy. In this study, the Big Transfermodel, which is a new General Visual Representation Learning method andsix other classical DL methods are converted to the federated version. Theeffect of the federated learning is measured on all these models on four differentdata settings extracted from the MHIST and Chaoyang datasets. The proposedmodels are tested for single learning, centralized learning, and federatedlearning. The best AUC values of federated learning on Chaoyang are obtainedby the Big Transfer and VGG models at 90.77 and 90.76, respectively,whereas the best AUC value on MHIST is obtained by the Big Transfer modelat 89.72. The overall obtained results of models on all data settings show thatthe contribution of Federated Learning with respect to single learning is 4.71and 11.68 for the “uniform” and “label-biased” data settings of Chaoyang,respectively, and 6.89 for the “difficulty level-biased” data setting of MHIST.Thus, it is experimentally shown that federated learning can be applied to thefield of computational pathology for new institutional collaborations.