Brain tumors remain one of the most life-threatening neurological disorders, requiring early and accurate detection to improve patient survival rates.Magnetic Resonance Imaging (MRI) is widely used for non-invasive diagnosis of brain abnormalities due to its high spatial resolution and superior tissue contrast.However, traditional diagnostic techniques, including biopsies and manual interpretation of MRI scans, are often invasive, time-intensive, and prone to inter-observer variability.Moreover, the growing reliance on cloud-based medical imaging systems for storage and analysis introduces additional risks related to data privacy, model integrity, and unauthorized access.Ensuring both diagnostic precision and data confidentiality in such distributed environments remains a critical challenge.This research presents an advanced and secure cloud-based framework for encrypted brain MRI medical image classification, integrating multiple high-performance computational modules.The proposed system employs Memory-Boosted Guidance Filtering (MBGF) for enhanced noise suppression and texture preservation, and Quadratic Phase Quaternion Domain Fourier Transform (QPQDFT) for comprehensive multidimensional feature extraction.These refined features are classified using a Cascaded Visual Gates-Controlled Deep Unfolding Network (CV-GC-DUN), which synergizes visual attention mechanisms and gated optimization processes for improved interpretability and accuracy.Network weights are fine-tuned through the Black-Winged Kite Algorithm (BWKA), a metaheuristic optimizer designed to balance exploration and exploitation during convergence.To secure sensitive data before cloud transmission, the Bit-level Cryptanalyzing Image Encryption Algorithm (BCIEA) ensures high encryption efficiency and resilience against cryptographic attacks, guaranteeing end-to-end data confidentiality.Experimental validation using benchmark MRI datasets demonstrates that the proposed model achieves exceptional diagnostic performance, with an accuracy of 99.92%, recall of 99.93%, and a minimal error rate of 0.08%, alongside strong encryption quality (PSNR: 36.87 dB).These results confirm the framework's potential to deliver both robust medical image classification and uncompromised data security, making it a viable foundation for nextgeneration intelligent healthcare systems and cloud-based diagnostic platforms.