Evaluation of a numerical observer to determine the influence of JPEG data compression on the diagnostic quality of computed tomography images

Abstract

A numerical observer (JPEGNO) to evaluate the influence of JPEG compression on the diagnostic quality of CT image is proposed. JPEGNO is based on the grey scale histogram of the image and is defined as the inverse of the sum of the difference between successive grey levels in the histogram of the image. JPEGNO = 1/? abs[hist(i)-hist(i+1)] with hist(i) the number of pixels in the image with grey level i. The aim of this study was to determine the validity and reliability of the suggested numerical observer and how it can be used to determine the optimum levels of compression. Five sets of images were ranked by human observers and JPEGNO according to image quality. The rank correlation between JPEGNO and human observers was used to determine the validity of the numerical observer. The correlation of the JPEGNO values between different sets of images was used to determine it's reliability. JPEGNO were used to determine the optimum compression ratio for brain, lung and chest (soft tissue) images. The presence of pathological information in uncompressed brain and chest images were compared to that in their decompressed counterparts on which a optimum level of compression was performed. The rank correlation between the humans and JPEGNO is 0,78 and the average rank correlation between the different human observers 0.82 ± 0.10. The average correlation between JPEGNO for the different sets of images is 0,998 ± 0,002. The optimum compression ratios for brain, lung, and chest soft tissue are 14:1, 11:1 and 15:1 respectively. The difference between the presence of pathological information in the uncompressed and optimum compressed images is 0.15 ± 0.86. The rank correlation between JPEGNO and the human observers is of the same order as the human-human correlation. This indicates the validity of JPEGNO to evaluate the influence of JPEG compression on brain CT images. The high correlation in the JPEGNO values for different sets of images indicates the reliability of this numerical observer. The evaluation pathological information in the decompressed images indicates that JPEGNO can be used to determine a optimum compression ratio for a specific type of CT examination.