ANALYTICAL MODELS OF DETECTION AND DISCRIMINATION

点击翻译按钮获取中文摘要

DESCRIPTION (Adapted from Applicant's Abstract): The proposed research would develop analytical models relating human performance in detection tasks to physical properties of the imaging system. The tasks to be investigated here resemble diagnostic tasks more closely than do the widely used signal-known- exactly (SKE) tasks in that they involve uncertainty about certain aspects of the signals, such as size or location. Therefore, the applicants expect that they would be more valid bases on which to design imaging systems. The proposed analytical approach generalizes detection tasks to estimation of certain parameters of the signal, such as amplitude or size. The models to be developed would be based on the Cramer-Rao (CR) and Barankin lower bounds on the precision with which the parameter of interest can be determined in the presence of uncertainty as to the values of other parameters. These bounds depend on the imaging task (i.e., which parameters are considered unknown), the signal, and the physical properties of the imaging system; they do not require specification of an estimation procedure. The bounds quantify the propagation of pixel noise into variance of the parameter estimates, incorporating the effects of nonstochastic ambiguities.Parameter estimation tasks, e.g., signal amplitude estimation, can be associated with binary tasks, e.g., signal detection. Increased variance in a parameter estimate implies decreased ideal observer signal to noise ratio (SNR) in the relevant binary task. It was proposed that increased variance in amplitude estimates due to uncertainty about the signal would be reflected in reduced human performance in detection. This hypothesis would be tested by psychophysical experiments which estimate the receiver operating characteristic (ROC) curve for human performance in the detection tasks, allowing direct comparison between human and ideal performance.