Human observer efficiency for signal detection and localization in emission tomographic images

For the medically relevant task of joint detection and localization of a signal (lesion) in an emission computed tomographic (ECT) images, it is of interest to measure the efficiency, defined as the relative task performance of a human observer vs that of an ideal observer. Low efficiency implies that improvements in reconstruction algorithms may be possible and also that an ideal observer might be suitably handicapped to derive a model observer that emulates human performance. In our experiments, we use a simplified “filtered noise model” proposed in [1] that simplifies the complex ideal observer calculations. This model is used to emulate the tomographic reconstruction process where the correlation structure of the reconstructed images is a combination of quantum noise and the noise due to background variability both modulated by a form of regularization implemented during the reconstruction process. A two-alternative forced choice (2AFC) test is used to obtain the performance of the human observers. We also introduce two efficiency definitions appropriated for the underlining joint detection-localization tasks. Experimental results show that both the ideal observer and the human observer perform badly in localizing the exact center of the signal but much better in obtaining the rough location of the signal. The human efficiency depends strongly on the amount of smoothing in the image, with efficiency dropping for both over-smoothed case and under-smoothed case. Human efficiency increases approximately monotonically with signal intensity. We compared these results with a signal-known-exactly case and observed similar trends.