OPTIMIZED DEEP RESIDUAL NETWORK-BASED KEY GENERATION AND ATTRIBUTE-BASED ENCRYPTION FOR SECURE MEDICAL DATA IN CLOUD ENVIRONMENTS

Data security in healthcare is increasingly critical, particularly for cloud-based Patient Health Records (PHRs) shared among users. This research tackles the persistent challenges of securing healthcare data by introducing a novel approach that integrates Deep Residual Network (DRN) with Gezelle Brown-Bear Optimization Algorithm (DRN_Gez-BOA). The proposed framework ensures robust encryption and efficient storage of sensitive PHRs in cloud environments. The methodology is structured into two crucial phases: encrypting and securely uploading PHRs to the cloud, followed by retrieval and controlled access to the stored records. Multiple entities, like data owners, PHRs, cloud storage, and data users are considered in this work. The various phases include attribute extraction, key generation, attribute-based encryption, access control, and decryption process. The key used for encrypting medical data is generated in the key generation phase using DRN_Gez-BOA. The DRN undergoes training utilizing the Gez-BOA framework, which synergistically integrates the Gezelle Optimization Algorithm (GOA) with the Brown-Bear Optimization Algorithm (BOA). This fusion enhances optimization efficiency, leading to improved cryptographic key generation. The DRN_Gez-BOA model achieves notable computational performance, demonstrating a key generation time of 24.876[Formula: see text]ms, encryption time of 240.643[Formula: see text]ms, decryption time of 273.642[Formula: see text]ms, memory usage of 4.315[Formula: see text]MB, and energy consumption of 0.288[Formula: see text]J when operating with a key size of 16[Formula: see text]MB. Additionally, the proposed DRN_Gez-BOA model has lower memory usage compared to several other methods, with improvements of 33.99% over Customizable Unique Node Access (CUNA), 19.63% over Boneh Goh Nissim Bilinear Attribute-based Optimal Cache Oblivious (BGNBA-OCO), 14.30% over Bee-Foraging Learning-based Particle Swarm Optimization (BFL-PSO), and 4.39% over Verifiable Attribute-Based Keyword Search Scheme (VABKSS). Furthermore, by integrating DRN, which excels in deep feature extraction, with Gez-BOA, which optimizes key generation and encryption parameters, this method ensures robust protection against unauthorized access while maintaining computational efficiency. This approach not only strengthens security but also enhances accessibility and scalability, making it a viable solution for modern medical data management systems.