JoWUA

Abstract

The article focuses on the issue of safe healthcare data management within decentralized systems, specifically, blockchain-based ones. Conventional blockchain-based systems have great concerns regarding exposure of privacy, cost of storage and cryptographic overhead of storing confidential medical records. The proposed work is an integrated system that integrates both Feature-Gated Convolutional Neural Networks (FG-CNN) to predict the disease and Transaction-Structured Selective Encryption (TSSE) to encrypt data efficiently. The FG-CNN model is also trained on the UCI Heart Disease dataset, where it also does binary classification to predict heart disease and gives the sensitivity score of each feature. Selective encryption with these sensitivity scores is then applied to minimize the amount of unwarranted encryption overhead whereby only privacy sensitive data fields are encrypted. The encrypted data will be stored in off-chain MongoDB database, with blockchain-backed integrity proofs stored on Ethereum. The experimental methodology proves the efficiency of the TSSE framework in terms of minimizing the encryption time by 63%, decryption time by 61%, and blockchain storage overhead by 88%. Also, the use of blockchain gas is minimized by 55% which is a massive enhancement as compared to the conventional full-data encryption systems. The system also lowers privacy risks by 75%, although it has high privacy guarantees. The NIST statistical randomness tests confirm the secure and random nature of the encrypted data, indicating the robustness of the proposed encryption approach. Nevertheless, the article does not include clear quantitative forecasts of classification accuracy, precision, recall, and F1-score of the FG-CNN model, which may offer a more detailed evaluation of its performance in diagnosing the disease.