Enhancing Security and Privacy in Cloud- and IoT-Driven Medical Imaging Using a Cascaded Visual Attention–Deep Spiking Parallel Convolutional Neural Network Optimized with the Improved Orca Predation Algorithm

The rapid adoption of cloud computing and Internet of Things (IoT) technologies in healthcare has significantly improved medical data accessibility and remote diagnosis. However, the integration of distributed storage, real-time transmission, and deep learning–based analysis exposes medical images to serious security, privacy, and computational challenges. The current diagnostic models are mainly concerned with accuracy in classification and ignoring secure storage and transmission, and even traditional encryption algorithms tend to have high computational overhead that does not prefer an IoT setting. In order to overcome these shortcomings, this paper presents a safe and effective model of cloud- and IoT-based medical imaging. A Hyperchaotic System–Fibonacci Q-Matrix (HFQM) is also used to encrypt medical images in the first step to assure high levels of confidentiality at a minimal computational cost. Rolling Guidance Filtering (RGF) works with image quality after secure retrieval, and it does not lose image structural details. The Kolmogorov–Arnold Transformer (KAT) segments the tumor regions, and the classification is done using a hybrid Cascaded Visual Attention–Deep Spiking Parallel Convolutional Neural Network (CVA-DSPCNN). The Improved Orca Predation Algorithm (IOPA) is used to optimize network weights, make convergence faster, and minimize classification error. The results of the experimental assessment of the Figshare and Brain Tumor MRI datasets indicate that the suggested CVA-DSPCNN-IOPA model reaches an accuracy, precision, recall, and F1-score of more than 99% at a low encryption and computation time. The findings validate the effectiveness of the proposed system in terms of balance between security, efficiency, and diagnostic reliability in a cloud- and IoT-based healthcare setting.