Stochastic interpretation of thallium myocardial perfusion scintigraphy

A method is presented for the quantitative interpretation of thallium-201 myocardial perfusion studies. The data are planar images collected immediately following the stress injection, and 4 to 6 hours later. Data analysis consists of preprocessing, including thresholding of the original data, and data reduction using a variant of the circumferential profile methods. The profiles are subdivided into segments, and for each segment the difference between the norm and the actual data is computed. This difference is a quantitative symptom, whose size is assumed to be related to the probability of having the disease. The relationship between the size of the symptom in each of nine segments (three segments/view in three views) and the probability of disease is expressed in a table in which for 30 diseases (combinations of vascular lesions) the sensitivity for each of the nine symptoms is described as a Gaussian function whose average and standard deviation are computed from previous validated cases. Using an arbitrary prevalence, the post-testing probability can then be computed using Bayes' formula sequentially. The sensitivities, however, are not expressed as a binary function of the presence or absence of a symptom, but as a distribution function defined by experience. The method is sensitive for the detection of isolated left anterior descending disease (100%) and triple vessel disease (96%). When specific combinations of vascular lesions are recognized the specificity increases from 79% (when unspecified abnormalities are recognized) to 87% when left anterior descending disease or triple vessel disease is recognized