Ensuring the integrity of AI models: a blockchain-based approach for protecting medical imaging training data

Abstract The development of trustworthy AI models is crucial, particularly for critical medical applications such as brain tumor detection using MRI images. However, medical images are being increasingly targeted by adversarial attacks to compromise the diagnostic accuracy of AI-based solutions. Adversarial attacks lead to deliberate perturbations in the medical imaging dataset, which will deceive the functioning of AI models. While these perturbations are visually imperceptible to human beings, they can cause AI models to malfunction if trained on adversarial images. To enhance the trust of healthcare professionals and patients in AI-based diagnosis of brain tumors, this research article presents a novel blockchain-based framework that utilizes Hyperledger Fabric and Private IPFS. This framework will safeguard MRI scans for brain tumor detection from unauthorized access & tampering by adversaries by decentralizing data storage and access control while ensuring data provenance. Hyperledger Fabric and private IPFS enable secure and tamper-proof dataset storage and sharing, leading to reliable and adversarially robust AI-based solutions. Experimental evaluations of the proposed framework demonstrate decentralized cryptographic assurance of image integrity in a permissioned blockchain network of the healthcare and AI fraternity. Performance of this defense strategy is validated using Hyperledger Caliper.