Skip to Main Content
Different implementations of the receiver operator characteristic (ROC) method in cardiac perfusion SPECT have been described. However, none has attempted to independently assess cardiac function. The aim of this study was, therefore, to design and execute a human observer ROC study that includes the evaluation of cardiac function and perfusion. Due to the lack of a gold standard, our initial design used an improved Mathematical Cardiac Torso (MCAT) phantom to generate normal and abnormal regional function. Abnormal heart function included hypokinesis, akinesis, and dyskinesis. Two heart sizes (124 ml and 100 ml left ventricular cavities) were used for male and female patients, respectively. Eight different locations around the LV myocardium were selected and perfusion defects of various sizes generated. Sixteen gated hearts across the cardiac cycle with abnormal cardiac function according to the different motion models were generated. The SIMIND Monte Carlo package was used to simulate a clinical Tl-201 perfusion SPECT acquisition protocol on the 3-headed IRIX gamma camera (Philips Medical Systems, Cleveland, Ohio). Data were reconstructed using the rescaled block iterative (RBI) technique with 17 subsets (4 projections/subset) and 5 iterations. Three sets of sixty-five cases were shown to an observer, the first 10 cases were used as training while the remainder (55 cases) were read and scored. The readings of the total 165 observed cases served as input data for ROC curve generation. The observer read the data in two ways. In the first reading the observer gave a confidence rating for regional myocardial function followed by a separate confidence rating for a perfusion defect. In the second reading, the order was reversed. ROC curves and areas under curve (AUCs) were determined separately for each reading approach. ROCs and AUCs were calculated for the three coronary artery territories as well as an overall calculation for the heart. While several ideas to further- - improve the methodology were generated during this study, we believe that we have clearly demonstrated that one can perform an independent quantitative task based assessment of cardiac function.