Context-Based Coding of Ultrasound Medical Images using Shape-Adaptive Transform Representation

Image compression is essential in order to meet storage and transmission requirements for the vastly produced medical image data. While it is desirable to achieve high compression rates, it is crucial that important medical diagnostic details will be preserved. We introduce a new approach to processing medical images, in which Regions-Of-Interest (ROIs) are automatically detected and coded with relatively high bit rates, while the background is coded with low bit rates. Our coding algorithm is based on identifying the ROI pixels and division of the image into blocks, which are classified based on their ROI pixels contents. Each block is coded according to its classification, using Discrete Cosine Transform (DCT) or Shape Adaptive (SA)-DCT. In addition, we propose a tailored interpolation scheme, which significantly reduces artifacts created in ROI segments that are of irregular or noncontinuous shape. Results on echo-cardiac and fetal images demonstrate the performance of our method.