DETERMINATION OF THRESHOLDS FOR DETECTION OF CEREBELLAR BLOOD-FLOW DEFICITS IN BRAIN SPECT IMAGES

Two observer studies were performed to determine the threshold (i.e., ratio of the counts in a lesion area to the counts in the correspondin g contralateral region) at which two experienced observers diagnosed b lood flow deficits in the cerebellum in Tc-99m-HMPAO SPECT scans to be clinically significant, and investigate the effect of the intensity m apping scale on the detectability of lesions. Method: Lesions represen ting blood flow deficits varying from no decrease to a 12.5% decrease were simulated in 300 patient images. The first study, a receiver-oper ator characteristics (ROC) experiment, used two observers to compare t he detectability of lesions with three intensity mapping scales: two p seudocolor scales, and a linear gray scale, A second ''threshold-crite rion'' study was done to estimate the threshold at which observers det ermine deficits to be clinically significant. Results: In the ROC stud y, the observers were more accurate in detecting lesions displayed in pseudocolor than in gray scale. In the threshold-criterion study, the threshold at which observers assessed clinically significant deficits was found to range between 0.900 and 0.950 (corresponding to a 5%-10% decrease in counts), depending on the observer, and the intensity mapp ing scale. For both observers, the detection threshold was higher (i.e ., closer to 1.0) with the pseudocolor scale than with the gray scale. Conclusion: The definition of a threshold value for use in quantitati ve techniques is dependent on both the observer and the intensity mapp ing scale. Observers were more accurate with the pseudocolor scales.