Pareto Optimization–based Nonsubsampled Contourlet Transform (NSCT) Medical Image Compression

Abstract

The exponential growth of medical imaging data from modalities such as MRI, CT, and X-ray poses significant challenges for storage, transmission, and real-time analysis in clinical settings. Conventional compression techniques, including JPEG, JPEG2000, and wavelet-based methods, often fail to preserve fine anatomical and pathological structures critical for accurate diagnosis. This study proposes a Pareto Optimization–based Nonsubsampled Contourlet Transform (PO-NSCT) framework for medical image compression, integrating multi-scale and multi-directional decomposition with evolutionary multi-objective optimization. The compression process is formulated as a multi-objective problem to simultaneously minimize distortion (MSE), maximize compression ratio (CR), and preserve structural fidelity (SSIM). NSGA-II is employed to identify Pareto-optimal parameter configurations, enabling a systematic trade-off between rate and image quality. Experimental evaluation is conducted on 7,023 brain MRI images from Figshare, SARTAJ, and Br35H datasets. Results demonstrate that PO-NSCT outperforms existing methods, achieving higher compression ratios (up to 65.7%) with superior PSNR (>38 dB) and SSIM (>0.95), while enhancing downstream classification accuracy by 5–10%. Statistical validation via Wilcoxon signed-rank tests confirms significance (p < 0.05). The framework preserves diagnostically relevant features, ensuring clinical reliability, scalability, and suitability for telemedicine and AI-assisted workflows.

Keywords

Compression, NSCT, Pareto, Optimization, Distortion, Fidelity, PSNR, Classification, and medical image.