Computer versus visual analysis of exercise thallium-201 images: A critical appraisal in 325 patients with chest pain

Abstract

Although computer-aided analysis of planar thallium images has been previously reported to be superior to visual analysis, a segment-by-segment comparison of all thallium variables (presence of initial defect, presence of redistribution, and presence of abnormal lung:heart thallium ratio) has not been performed with the use of both methods. In the present study, such a comparison was performed to determine which thallium variable(s) are superior, when the computer method is used. In addition, both methods were compared to determine their correlations with the angiographic extent of coronary artery disease (no disease, or one-vessel disease vs multivessel, or left main disease) and the occurrence of future events (cardiac-related death or nonfatal myocardial infarction). Computer-aided analysis was superior to visual analysis (an average of three independent blinded observers) for detection of coronary disease because: (1) it was more specific for detecting initial defects despite having the same sensitivity (89.6% vs 91.0%, p = NS) and (2) it was more sensitive for detecting redistribution in the territory of a stenosed vessol (68% vs 46%, p < 0.001). However, absolute myocardial clearance of thallium had very poor specificity compared to visually assessed redistribution. Only when segmental clearance was considered abnormal on a relative basis (if it was slower by 98% or more than the fastest clearing segment in the heart) did abnormal clearance achieve the same specificity (99%) and a better sensitivity (60% vs 48%, p < 0.01) than visual analysis. Computer and visual analysis were similar in their correlations with the angiographic extent of coronary artery disease (number of diseased vessels), and correctly predicted the angiographic extent of coronary artery disease in two thirds of the cases. Similarly, both methods were equally good in determining long-term (mean follow-up period of 5.3 ± 1.0 years) prognosis and correctly identified occurrence or absence of events in 90% of the cases. Lung:heart ratio and number of segments demonstrating redistribution were the two variables that determined prognosis when both methods were used. In conclusion, computer-aided analysis of thallium images is superior to visual analysis of thallium images with multiple observers and should be the preferred method of analyzing thallium images in the clinical setting.