Abstract Tongue images are increasingly employed as diagnostic data in intelligent clinical systems. However, the storage and bandwidth costs of high-resolution image data can reduce diagnostic efficiency, making image compression a critical tool for balancing data fidelity, transmission cost, and retrieval latency. General-purpose methods fail to fully exploit the cross-channel correlations and smooth textures of tongue images, leading this study to propose a lossless compression algorithm based on a residual prediction framework. The compression pipeline comprises four stages: First, integer lifting performs reversible color transformation, eliminating interchannel redundancy. Second, an enhanced LOCO-I model incorporating gradient detection and “Left-fused” mode captures smooth tongue textures. Third, an S-shaped scan reorders residuals to preserve spatial locality. Finally, cross-channel differential type mapping, dynamic bit-width packing, and hybrid entropy coding mitigate statistical redundancy. Experimental results across three tongue datasets demonstrate that the proposed algorithm exceeds the compression ratios of standard methods by 6%, 9% and 1%, respectively. It significantly outperforms standard PNG and JPEG 2000 formats while maintaining a decoding latency comparable to QOI. Furthermore, generalization tests on endoscopic images of colonic polyps, retinal fundus images, and dermatological images show compression improvements of 7%, 5% and 6%, respectively. The proposed method achieves a robust balance between high compression efficiency and low-latency decoding while ensuring faithful reconstruction within practical engineering constraints. Funding This work was supported by the National Science and Technology Innovation 2030 Major Program (Grant No. 2025ZD0544900), entitled “Research on Early Warning Technology for Colorectal Cancer Based on Multi-dimensional Analysis System of Traditional Chinese Medicine Tongue”; and the Postgraduate Research & Practice Innovation Program of Jiangsu Province (Grant No. KYCX25_2274), entitled “Research on Model Compression Algorithms for Medical Image Recognition”. Ethics declarations Competing interests The authors declare no competing interests. Additional information Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in