A near lossless Region of Interest(ROI)compression algorithm based on the shearlet transform was proposed for medical images to improve the Mean Structural SIMilarity(MSSIM) between the original image and the reconstructed image. Firstly, the ROI was designated in a medical image and the rests were regard as the Background (BG). Then, the ROI and BG were transformed into shearlet domains respectively, and the significant coefficients which could approximate the original region accurately were selected to be denoised and compressed. Furthermore, the main coefficients in ROI were coded by lossless Huffman coding and those in BG were quantized and coded by Huffman coding. Finally, the reconstructed image was obtained by Huffman decoding and inverse shearlet transform. Experiment results show that the MSSIM and Peak Signal Noise Ratio (PSNR) between the original ROI and the reconstructed image ROI obtained by the new algorithm have increased by 4 percent and 135 percent respectively as compared to the modified Set Partitioning in Hierarchical Trees (SPIHT) algorithm with the same compression ratio. Moreover, for the whole image, the MSSIM and PSNR have increased by 3 percent and 28 percent, respectively. With configurable ROI’s and BG’s quality, the proposed algorithm is suitable for the medical image compression in the Picture Archiving and Communication System(PACS).