Artificial intelligence-based healthcare cybersecurity system with blockchain: modified parallel convolutional neural network for attack detection
Swarooparani Kolsur, Sridevi Hosmani · Medicine in Novel Technology and Devices · 2025
While smart wearables and remote devices have improved the speed of diagnosis and treatment, they have also created significant cybersecurity risks, especially with regard to the confidentiality and integrity of medical data. Because the primary means of operation for these Internet of Things (IoT) devices is constant data transmission, they are vulnerable to cyberthreats including Distributed Denial-of-Service (DDoS) assaults and data injection. This study suggests an AI-based Healthcare Cybersecurity System (AI-HCSS) that integrates blockchain technology to mitigate these vulnerabilities and provide strong, real-time patient data and healthcare system protection. A new architecture is shown to identify and counteract DDoS attacks on the cloud infrastructure, and blockchain is used for safe and unchangeable data storage. The system extracts statistical, raw, and enhanced entropy-based features after performing improved min-max normalization for data pre-processing. Then, for precise DDoS attack detection, a modified Parallel Convolutional Neural Network (PCNN) is used. The model's output is interpreted using the SHapley Additive exPlanations (SHAP) approach, which identifies important characteristics that affect detection performance in order to improve transparency and aid clinical decision-making. According to experimental results, the modified PCNN outperforms traditional methods with a high detection accuracy of 91.1%. In addition to bolstering the cybersecurity of healthcare IoT ecosystems, this integrated solution guarantees the real-time defense of clinical systems and patient data against changing cyberthreats. • Primarily, the data is acquired and stored in block chain. • To detect attack, pre-processing is done via Improved min-max normalization. • Then, statistical features, Improved entropy and raw features are extracted.