Blockchain-Based Access Control Schemes for Medical Image Analysis System

Medical image analysis systems with machine learning have played an important role in the computer-aided diagnosis and treatment for diseases. However, individual privacy of user data is vulnerable since the training data is exposed to unauthorized user. Therefore, this paper designs an access control scheme to prevent illegal users from accessing medical data while achieving high accuracy of lesion classification. Specifically, in the novel lightweight consortium blockchain-based access control scheme, a chosen consortium node is utilized as key generation center instead of a trusted third party in conventional schemes. Two public retinal datasets are utilized for the classification of diabetic retinopathy (DR). Security analysis shows that the proposed scheme can prevent the user data from leakage and malicious tampering. Experimental results demonstrate that the processing of data cleaning is efficient to increase the accuracy of the classification for early lesions of DR by removing low quality images, and the accuracy is up to 90.2%.