Hybrid Optimization-Driven Multi-Attention Deep Learning Framework For Secure Authentication in Healthcare Iot Devices

Abstract

The rapid growth of IoMT-enabled smart healthcare systems has significantly improved remote patient monitoring and medical image analysis; however, existing approaches suffer from critical limitations such as weak encryption robustness, high computational overhead, lack of secure access control, poor scalability in blockchain integration, and overfitting in deep learning-based diagnosis models. To address these challenges, this research proposes a novel IoMT–Blockchain Chaotic Encryption and Optimization-Driven Deep Learning Framework (IBCEODDLF) for secure medical image transmission and intelligent disease diagnosis. Initially, IoMT devices collect real-time patient data, which are securely managed using a blockchain framework incorporating BFT consensus, Merkle hashing, and smart contracts to ensure integrity and traceability. A Modified Logistic Map-based chaotic encryption scheme optimized via the American Zebra Optimization Algorithm (AZOA) enhances key sensitivity and diffusion strength. After secure transmission and decryption, DenseNet-169 with Hiking Optimization-based hyperparameter tuning extracts discriminative features, and final disease classification is performed using a Physics-Informed Neural Network (PINN)  that integrates clinical data with physiological constraints for improved generalization. Experimental results demonstrate superior performance with 98.7% accuracy, 98.3% precision, 98.1% recall, 98.2% F1-score, AUC of 0.992, PSNR of 41.8 dB, entropy of 7.999, NPCR of 99.63%, average BFT consensus time of 2.3 s, and PINN final loss of 0.039, confirming enhanced security, efficiency, and diagnostic reliability.