The impact of skull bone intensity on the quality of compressed CT neuro images

The increasing use of technologies such as CT and MRI, along with a continuing improvement in their resolution, has contributed to the explosive growth of digital image data being generated. Medical communities around the world have recognized the need for efficient storage, transmission and display of medical images. For example, the Canadian Association of Radiologists (CAR) has recommended compression ratios for various modalities and anatomical regions to be employed by lossy JPEG and JPEG2000 compression in order to preserve diagnostic quality. Here we investigate the effects of the sharp skull edges present in CT neuro images on JPEG and JPEG2000 lossy compression. We conjecture that this atypical effect is caused by the sharp edges between the skull bone and the background regions as well as between the skull bone and the interior regions. These strong edges create large wavelet coefficients that consume an unnecessarily large number of bits in JPEG2000 compression because of its bitplane coding scheme, and thus result in reduced quality at the interior region, which contains most diagnostic information in the image. To validate the conjecture, we investigate a segmentation based compression algorithm based on simple thresholding and morphological operators. As expected, quality is improved in terms of PSNR as well as the structural similarity (SSIM) image quality measure, and its multiscale (MS-SSIM) and informationweighted (IW-SSIM) versions. This study not only supports our conjecture, but also provides a solution to improve the performance of JPEG and JPEG2000 compression for specific types of CT images.