TREE-STRUCTURED VECTOR QUANTIZATION OF CT CHEST SCANS - IMAGE QUALITY AND DIAGNOSTIC-ACCURACY

The quality of lossy compressed images is often characterized by signal-to-noise ratios, informal tests of subjective quality, or receiver operating characteristic (ROC) curves that include subjective appraisals of the value of an image for a particular application. We believe that for medical applications, lossy compressed images should be judged by a more natural and fundamental aspect of relative image quality: their use in making accurate diagnoses. We apply a lossy compression algorithm to medical images, and quantify the quality of the images by the diagnostic performance of radiologists, as well as by traditional signal-to-noise ratios and subjective ratings. Our study is unlike previous studies of the effects of lossy compression in that we consider non-binary detection tasks, simulate actual diagnostic practice instead of using paired tests or confidence rankings, use statistical methods that are more appropriate for non-binary clinical data than are the popular ROC curves, and use low-complexity predictive tree-structured vector quantization for compression rather than DCT-based transform codes combined with entropy coding. Our diagnostic tasks are the identification of nodules (tumors) in the lungs and lymphadenopathy in the mediastinum from computerized tomography (CT) chest scans. Radiologists read both uncompressed and lossy compressed versions of images. For the image modality, compression algorithm, and diagnostic tasks we consider, the original 12 bit per pixel (bpp) CT image can be compressed to between 1 bpp and 2 bpp with no significant changes in diagnostic accuracy. The techniques presented in this paper for evaluating image quality do not depend on the specific compression algorithm and are useful new methods for evaluating the benefits of any lossy image processing technique.