OPTIMIZATION PERFORMANCE OF CYBERSECURITY WITHIN MULTI-MODAL CLOUD HEALTHCARE INDUSTRY BASED ON DEEP LEARNING TECHNIQUES

Abstract

In next-generation medical systems, the amount of Internet of Things (IoT) devices, edge computing infrastructure, and multi-modal cloud architectures has increased significantly, making healthcare cybersecurity a global concern. Although healthcare organizations invest a lot in storing security measures, they are far more susceptible to the sophisticated cyber threats. More than 90% of them have experienced at least one cyberattack over the past few years, and between 2013 and 2023, ransomware attacks in the healthcare industry have seen a surge to unprecedented heights. There is an important gap in the current literature: current cybersecurity frameworks don't offer a unified, scalable, and efficient solution for securing real-time flows of data across the different layers of the IoT, across the shore (edge) clouds, and across the center (core) clouds in a multi-domain healthcare environment. The current models are either computation intensive or domain specific, with multi-modal healthcare architecture being under-protected. This study introduces a novel hierarchical Deep Neural Network (DNN) framework that can identify and block malicious activities in multi-area healthcare data flows. The framework is inspired by transfer learning, utilizing pre-trained edge cloud models and fusion is to an optimized center cloud model. A Capability List (CP List) access control mechanism is built-in to implement fine-grained data access policies. The proposed model is tested on the structured dataset of healthcare network security incidents from the simulated environments in 2024. The experimental results show that the proposed hierarchical DNN architecture can achieve the accuracy of 95% to 100% for training and testing in the unknown attack. Taking an average of 26.2% less training time than the independently trained center cloud models, the model needs just 6-8 epochs to train, whereas the standalone edge cloud model requires 35-40 epochs. The research has validated that the combination of hierarchical deep learning and cloud-based security systems can greatly improve threat detection and decrease computational load. To protect sensitive patient data, the use of hierarchical AI-driven security is recommended, as well as the standardization of multi-layer access control across all IoT, edge and cloud environments for policymakers and healthcare administrators.