Lossless Medical Image Compression Using ROI Segmentation with Deep Learning Techniques

With The increasing use of medical imaging has put a greater requirement on fast and efficient storage and transmission. Compression algorithms need to strike a balance between reducing file sizes without distortion of valuable diagnostic information. In this paper, a hybrid lossless compression method is proposed that uses deep learning-based ROI segmentation. A U-Net model is used specifically to segment the cardiac anatomy on CT scans accurately. Lossless compression is used to preserve large diagnostic areas, and lossy compression is used to compress small areas. The approach strikes an effective balance between the two to yield high compression ratio without compromising medical integrity. Experimental evidence proves its efficacy in minimizing storage needs without compromising diagnostic accuracy. This work recognizes the potential of deep learning for medical image processing and paves the way for upcoming AI-based compression methods.