An automated analysis technique for thallium images.

Abstract

The use of stress thallium-201 scans in the non-invasive assessment of myocardial perfusion is well established, despite several reports of considerable inter-observer variability in the assessment of perfusion defects. By applying a simple statistical algorithm to a set of normal thallium images and using a well defined criterion of abnormality, the threshold of normality in these 'statistical images' was obtained for each of four projections. Subsequently a test set of images from both normal volunteers and patients with arteriographically documented coronary artery disease were reported using statistical images at four levels (70, 75, 80 and 85% of the mean of the hottest pixels) and standard thallium images viewed on the computer monitor in both colour and black and white. Significant reductions in the inter-observer disagreement and enhanced predictive accuracy in the detection of significant coronary artery disease were obtained using the statistical images. The technique described and assessed would permit the reporting of thallium scans at a preselected value of sensitivity and specificity depending on the requirements of the particular study. It could be readily implemented, after local validation, in any department performing thallium scans where the gamma camera is interfaced to a computer.