Tumour-Aware Medical Image Compression via Directional Intra Prediction and Deep Context-Adaptive CABAC
DOI:
https://doi.org/10.70917/ijcisim-2026-2544Keywords:
Medical Image Compression, Brain MRI, Tumour-Aware Compression, Directional Intra Prediction, CABAC, Deep Neural Networks, Rate–Distortion Optimization, ROI-Based Encoding, Prediction Refinement Network, Probability Estimation NetworkAbstract
Medical image compression has proven necessary to minimize storage and transmission overheads whilst maintaining diagnostically important information in clinical imaging systems on a large scale. This paper suggests a hybrid tumour-aware medical image compressor based on Directional Intra Prediction (DIP), Context-Based Adaptive Binary Arithmetic Coding (CABAC), and Deep Neural Networks (DNN) to compress brain MRI efficiently. The model is tested on a mixed dataset of 7,023 MRI pictures on Figshare, SARTAJ, and Br35H that includes four classes. The proposed approach attains the lowest possible bitrate of 0.31 bpp, which is better than baseline approaches like MLic++ (0.40 bpp) and DWT-PCA-Huffman (0.44 bpp). Gain of compression ratio is increased to 69.7 and reconstruction quality remains high with PSNR of 38.7 dB, SSIM of 0.978 and lower MSE of 0.0029. Statistical validation by using the Wilcoxon signed-rank test gives significant results ( p ≤ 0.016). Moreover, the classification accuracy is increased to 98.0-99.5 on compressed images, which indicates that diagnostic features are preserved. The framework is a good balance between rate and distortion optimization and tumour fidelity, and can be used in clinical and telemedicine systems in real-time.
