Abstract
In the radiological diagnosis of cancer, false negatives are common (e.g., 10-30% in breast cancer screening). A potential cause of these errors is “Satisfaction of Search” (SOS): after having found a first lesion, the miss rate for additional lesions in the same case is substantially elevated. A key limitation to the study of SOS is the need to use trained Radiologists as subjects, as access to this population is limited. Although there has been some success in developing naïve-participant tasks, these have tended to sacrifice similarity to the target task for increased experimental control. In the present study, we developed a laboratory analog of medical image analysis and validated this approach through a series of experiments on SOS. The visual search backgrounds were contour-dense natural scene images, designed to simulate the image properties of dense breast tissue. The targets were semi-transparent face images drawn from two emotion categories: happy and angry. Naïve participants clicked on each detected face in an image, indicated the emotion, and rated confidence. Within participants, we manipulated the number of faces in an image (none, one, or two) and target salience (high, low). SOS was operationalized as a lower detection rate for low-salience targets when the same image contained a high-salience target versus when it did not contain a high-salience target. Between groups, we manipulated the prevalence of two-target images, with the prediction that low prevalence of second targets would lead to increased SOS due to early search termination. The results indicated a robust SOS effect, which was modulated in the predicted direction by two-target prevalence. The general approach provides a means to test specific hypotheses of the causes of SOS, which can then be validated in experiments with Radiologists. The present results demonstrate that the meta-cognitive factor of two-target prevalence contributes substantially to second-target misses.