Medical Image Compression and CNN Performance: A Systematic Meta-Analysis

Abstract

A Modern medical imaging systems generate a large number of successive images, creating significant challenges for efficient storage and transmission. Imaging modalities such as computed tomography (CT), magnetic resonance imaging (MRI), and fluoroscopy produce image sequences in which a considerable portion of the image content remains unchanged across successive frames. Conventional lossless compression techniques protect image quality but often achieve limited compression ratios, whereas lossy methods may eliminate clinically significant information. To address these limitations, this paper proposes a near-lossless compression framework designed for sequential medical images. The proposed method identifies static image regions and selectively encodes only the changing regions of interest (ROI) between consecutive images. Successive image subtraction is employed to identify non-zero ROI regions while assigning zero values to unchanged areas, thereby reducing spatial redundancy. Furthermore, a double-coding strategy is incorporated to enrich the overall compression efficiency. Unlike earlier approaches that process fluoroscopy images either individually or as conventional video streams, the proposed technique treats the entire image sequence as a coherent dataset to exploit temporal similarity more effectively. Experimental results obtained from fluoroscopy image datasets validate that the proposed approach achieves an improved compression ratio while maintaining diagnostically important image details. Comparative analysis with existing medical image compression methods shows the effectiveness and potential applicability of the proposed near-lossless framework for medical imaging storage systems.