Variable rate vector quantization for medical image compression with applications to progressive transmission

In this work, a new technique for variable rate VQ design based on tree structures is applied to medical images. It is an extension of an algorithm for optimal pruning in tree-structured classification and regression due to Breiman, Friedman, Olshen, and Stone [1]. The algorithm finds subtrees of a given tree-structured vector quantizer (TSVQ), each one optimal in that it has the lowest average distortion of all subtrees with the same or lesser average rate [2]. Since the resulting subtrees have variable height, natural variable rate coders result. Image reproduction at 1.5 bits per pixel is excellent and pathology in brain magnetic resonance images can be diagnosed in images at less than 0.5 bit per pixel. Finally, TSVQ is stored in a format convenient for progressive transmission of images.