Compression of medical images should not degrade the quality of the regions that are critical for diagnosis of diseases. Zyout et al. present one such application for mammograms, including its effect on microcalcification detection. The scheme used is a region-of-interest (ROI) encoding approach based on a wavelet transform. The pixels located on the boundary of an ROI have to be handled appropriately so that the coding distortion does not spread to the ROI pixels.

The method optimizes the quality of compressed images by the appropriate choice of integer wavelet transform and the starting bit plane of ROI coding. A binary mask corresponding to the ROI is composed by taking into consideration the importance, size, and center of the ROI. It employs an ROI-based set partitioning in hierarchical trees (SPHIT) coding method that uses a predefined starting bit plane corresponding to the importance of the ROI. At this bit plane, it separates the SPIHT lists into two sets: one corresponding to the ROI, and the other corresponding to the background. SPIHT coding is employed to encode the ROI, only until the bit budget is exhausted or a predetermined stop criterion is reached.

The scheme is extensively evaluated using digital mammograms available from the Mammographic Image Analysis Society (MIAS) database. The results are evaluated quantitatively, using rate-distortion (R-D) measure, as well as qualitatively. However, the ROI are selected manually, a step that could be automated.