Secure Hybrid Compression for Brain MRI Using DCT, DWT, a Convolutional Autoencoder, and Lightweight Encryption

Brain Magnetic Resonance Imaging (MRI) is critical for diagnosing neurological disorders, but the growing volume of MRI data poses challenges in storage, transmission, and security. Conventional compression techniques, including Discrete Cosine Transform (DCT), Discrete Wavelet Transform (DWT), and formats such as JPEG, JPEG2000, and WebP, reduce file size but often degrade fine structural details, affecting clinical interpretation. This study presents a novel hybrid framework that integrates classical transforms, modern image formats, and a Convolutional Autoencoder (CAE) with a lightweight XOR-based transformation for secure handling of compressed MRI representations. The approach includes: baseline compression with DCT, DWT, and standard formats; CAE training with progressively increasing encoder filters (32–256) and a 1024-dimensional latent space; and XOR-based transformation of latent codes to provide a lightweight obfuscation layer during storage or transmission. Feature heatmaps extracted from the encoder validate the preservation of anatomical details. The evaluation uses 253 MRI slices from a single publicly available dataset; therefore, further validation on larger and multi-institutional datasets is required and show that classical and standard methods achieve moderate compression (PSNR 23–27 dB) with artifacts, whereas the standalone CAE provides flexible compression (PSNR ~22.9 dB). The CAE with XOR-based transformation achieves high reconstruction fidelity, with a reported PSNR of 52.11 dB and SSIM of 0.994, shows low reconstruction error in the evaluated MRI slices, including regions containing visible tumor boundaries, and reduces storage by approximately 15:1. The framework simultaneously ensures high-quality compression and secure transmission, indicating potential applicability to medical image storage and transmission scenarios. Future work will explore deeper architectures, multimodal MRI datasets, advanced encryption methods, and real-time edge deployment.