ABSTRACT
Visual search is defined as grouping and assigning labels to different regions in the image, where the labels correspond to entities that exist in the real world. Two components of expertise in such tasks are identified. Search expertise is the ability to find lesions while avoiding finding non-lesions. Lesion-classification expertise is the ability to correctly classify found suspicious regions. Two experimental methods of studying visual search are described. By far the more common one involves measuring reaction time and accuracy in an image possibly containing a defined target and defined distractors. In the medical imaging context targets and distractors cannot be defined, as these are perceptual constructs existing in the radiologist's mind. The second paradigm involves measuring, using an eye-tracking apparatus, where the radiologist is looking. The Kundel-Nodine schema for modeling visual search in the medical imaging context is described. It consists of two stages, a brief global impression stage which uses peripheral vision to identify localized perturbations, the latent marks of the FROC paradigm. The second stage involves examining each found suspicious region using the fovea and making a decision of whether to mark it. The two-stage process bears a close resemblance to how CAD algorithms are designed. The Kundel-Nodine schema is the starting point of the radiological search model (RSM) described in the following chapter.
