Region of interest (ROI) compression for adrenal tumor images: a hybrid approach

Abstract

Developments in artificial intelligence have made it considerably easier for specialists to identify and diagnose diseases by applying computer-aided diagnosis (CAD) systems to medical images. This study introduces a novel hybrid compression technique that combines lossless compression of the region of interest (ROI) with lossy compression of the surrounding areas. This approach balances compression-ratio performance and diagnostic image quality by leveraging refined ROI extraction using morphological operations. This study proposes a hybrid ROI-aware medical image compression method designed to enhance the efficiency of adrenal tumor computed tomography (CT) image transmission and storage without compromising diagnostic accuracy. While surrounding regions are compressed using lossy techniques like scalar quantization and Huffman coding, tumor regions found by a reliable automated segmentation pipeline are compressed lossless using a Lempel–Ziv–Welch (LZW) algorithm. While maintaining 95% tumor detectability, an experimental evaluation of 95 annotated CT images produced an average compression ratio of 48.00%, outperforming traditional Huffman compression (20.79%). Peak signal-to noise ratio (PSNR), structural similarity index measure (SSIM), and a new ROI fidelity score were used to measure image quality. This approach is compatible with digital imaging and communications in medicine (DICOM) standards, supports bandwidth-efficient telemedicine, picture archiving and communication system (PACS) optimization, and may be incorporated into artificial intelligence-assisted diagnostic workflows. Future strategies should incorporate multimodal fusion, compression, and adaptive ROI tracking to further enhance clinical usability.