A Secure Hybrid Permissioned Blockchain and Deep Learning Platform for CT Image Classification

Abstract

Pneumonia is a life-threatening and prevalent disease and needs to be diagnosed within a short time because of the lungs' fluid flow. Late detection of the disease may result in the patient’s death. Thus, advanced diagnosis is a critical factor besides the disease progress. In addition to advanced diagnosis, the privacy of datasets is important for organizations. Due to the great value of datasets, hospitals do not want to share their datasets, but they want to share their trained network weights. Therefore, in this paper, we combine deep learning and blockchain to implement the blockchain as distributed storage. Using permission blockchain, weights are broadcasted among other hospitals securely. Because of the security, the dataset is shared with five hospitals equally. Each hospital trains its network model and sends its weights to the blockchain. The goal is to broadcast the aggregated weights among hospitals securely and have good enough results because the whole dataset is not implemented to train a network. The dataset contains 5856 images, and hospitals implement a residual neural network with 28 layers. The results show that hospitals can increase the accuracy of their model using shared weights compared to a model without using shared weights.

Ähnliche Arbeiten