A Secure Blockchain-Federated Deep Learning Model for Privacy-Preserving COVID-19 Diagnosis

Traditional diagnostic techniques have been exposed as having several shortcomings regarding sensitivity, scalability, and data protection due to COVID-19's widespread impact. A centralized training strategy remains hampered by data-sharing limitations, privacy risks, and a lack of trust between medical institutions despite deep learning's potential for accurate disease identification in chest CT imaging. This study presents a federated deep learning framework based on blockchain for privacy-aware diagnosis of COVID-19 via CT scans. A shared model can be trained collaboratively and decentralized, without requiring patients' sensitive information to be exchanged. In addition to homomorphic encryption, model gradients are also encrypted during training to further maintain data confidentiality. To enhance the effectiveness of feature extraction and classification, capsule networks and extreme learning machines are combined in an ensemble learning strategy. In experiments across multiple feature extraction networks, the proposed framework achieves very high recall, reflecting its high ability to detect COVID-19 cases while maintaining reliable precision. Accordingly, the proposed framework offers a practical and reliable solution for large-scale collaborative medical image analysis in pandemic situations that integrates accuracy, privacy preservation, and security.