The applicants proposed to develop a computer observer model of the human visual system to be used to evaluate medical image quality more accurately than presently available measures and more quickly and efficiently than human observer performance studies. Such a computer observer model could predict the effect of image compression on performance in clinically relevant tasks and enable automated optimization of image compression algorithms for maximum compression with no degradation in clinical diagnosis of coronary angiograms. To this goal, the applicants propose four specific aims: 1) To perform psychophysical measurements of the effect of image compression on performance in clinically relevant visual tasks (lesion detection and classification) for two popular compression algorithms (JPEG and wavelet) using test images that combine real clinical image noise with simulated lesions; 2) To compare conventional measures of image with current and newly developed computer observer models with respect to their ability to predict the observer task performance measured in Specific Aim 1; 3) To use the computer observer model to optimize the two compression algorithms to achieve maximum compression with minimal performance degradation; and 4) To perform a clinical validation of the optimized compression algorithms using clinical images and physician observers to confirm the predicted minimal degradation of clinical task performance with compression. If successful, the applicants proposed to establish a computer observer model as an accurate and efficient measure of image quality based on clinical task performance, which will be a valuable tool for developing optimal image processing and compression algorithms. The impact of this research will be more rapid and cost effective communication and storage of digital coronary angiograms without loss of diagnostic information.
Showing the most recent 10 out of 22 publications
