Visual processing and decision making in search and recognition of targets embedded in pictorial scenes
Abstract The present study examines the distributions of gaze durations associated with true and false, positive and negative decision outcomes resulting from visual search. The gaze-duration measure reflects visual processing time for detecting, interpreting and deciding whether an embedded object encountered during search is truly a target. The targets were simulated lung nodules embedded in the complex pictorial scenery of chest radiographs. The subjects were radiologists whose eye fixations were recorded while they searched the radiographs. Extensive visual scanning and localized focal fixating were required to detect, interpret and decide whether or not a nodule target was present in a particular location of the chest image. The results clearly demonstrate that when a viewer fixated a true target, gaze duration significantly increased regardless of whether or not the target was reported. This suggests that missed targets resulted primarily from decision errors rather than detection errors, assuming that it is possible to detect something object-like and decide not to report it. Target-free locations correctly recognized as such were associated with significantly shorter gaze durations than target-containing locations regardless of whether or not a target was reported. The mean values of gaze durations were found to rank order according to decision outcome: true positive = false positive > false negative > true negative. It was also found that image locations associated with true-negative decisions were re-fixated, after other locations were scanned, significantly fewer times than image locations associated with true-positive, false-positive and false-negative decisions. The results are discussed within the framework of a two-stage model of visual search and its relation to decision theory.