Prevalence effects in visual search: Theoretical and practical implications

描述(申请人提供)：项目摘要：在视觉搜索任务中，观察者(O)在包含干扰刺激的视觉场景中寻找目标。一些医学筛查任务(例如乳腺癌和宫颈癌)是困难的视觉搜索任务。这些任务的一个特点是目标流行率较低。也就是说，与许多其他现实世界的搜索任务(例如，在水果过道中寻找香蕉)和典型的实验室搜索任务不同，这些搜索目标很少出现。例如，在常规的乳房X光检查中，只有不到1%的乳房图像出现了病理。我们发现，这种较低的目标流行率本身可能是视觉搜索中遗漏错误的一个潜在来源(Wolfe等人，2005年)。以Exp为单位。在一系列不同的搜索刺激下，我们发现，当O在仅1-2%的搜索试验中出现时，O错过了0.30%到0.40%的目标。当同样的目标出现在50%的搜索试验中时，操作系统只会错过这些目标中的0.05%到0.10%。视觉搜索任务可以作为信号检测问题来研究。操作系统正试图将包含目标信号的显示器与仅包含噪声的显示器区分开来。未命中错误可能是由于对目标缺乏敏感度，或者由于将决策标准设置到导致可检测目标被归类为非目标的位置而引起的。在其他背景下(如警觉和分类文献)，对目标频率变化的反应被理解为标准的变化，而不是灵敏度的变化。我们已经证明，这在视觉搜索中也是如此。低流行率导致标准的巨大变化，这是令人惊讶的难以抵消的(例如，通过操纵不同类型反应的成本和收益)。相关性：医学筛查任务(如乳房X光检查和宫颈涂片)是搜索罕见目标的例子。我们的基础研究表明，这种稀缺性本身就使目标更难找到。由于实验室搜索任务和临床筛查任务之间的差异，以及由于临床任务是由训练有素的专业人员执行的，从现有数据得出关于医疗筛查错误的结论是不明智的。因此，我们建议确定这是否真的适用于医疗筛查任务，如果是，则测试基于理论的解决方案。这项研究有三个具体目的：1)检验患病率效应是乳腺癌和宫颈癌筛查中错误的潜在来源的假设；2)开发和测试视觉搜索中患病率效应的模型；3)测试理论和临床激励的减少错误率的策略。

DESCRIPTION (provided by applicant): PROJECT SUMMARY: In visual search tasks, observers (Os) look for a target in a visual scene containing distracting stimuli. Some medical screening tasks (e.g. breast cancer & cervical cancer) are difficult visual search tasks. A characteristic of these tasks is low target prevalence. That is, unlike many other real world search tasks (e.g. finding bananas in the fruit aisle) and unlike typical laboratory search tasks, these are searches for targets that appear only rarely. In routine mammography, for example, pathology is present on less than 1% of breast images. We have found that this low target prevalence, by itself, can be a potent source of miss errors in visual search (Wolfe, et al, 2005). In exps. with a range of different search stimuli, we have found that Os miss 0.30 to 0.40 of targets when those targets are present on only 1-2% of search trials. Os miss only 0.05 to 0.10 of the same targets when those targets are present on 50% of search trials. Visual search tasks can be studied as signal detection problems. Os are trying to distinguish displays containing a target signal from those containing only noise. Miss errors can arise from a lack of sensitivity to the target or from setting a decision criterion to a position that causes detectable targets to be classified as non-targets. In other contexts (e.g. vigilance & categorization literatures), the response to a change in target frequency is understood as a shift in criterion, not in sensitivity. We have shown that this is also true in visual search. Low prevalence produces a large shift in criterion that is surprisingly hard to counteract (e.g. by manipulations of costs and benefits for different types of response). Relevance: Medical screening tasks (such as mammograms and pap smears) are examples of searches for rare targets. Our basic research suggests that this rarity, by itself, makes targets more difficult to find. Because of differences between laboratory search tasks and clinical screening tasks and because clinical tasks are performed by highly trained professionals, it would be unwise to generalize from the existing data to the conclusions about errors in medical screening. Therefore, we propose to determine whether this is really true for medical screening tasks, and, if so, test theory-based solutions. The proposed research has three specific aims: 1) to test the hypothesis that prevalence effects are a potential source of errors in breast cancer and cervical cancer screening, 2) to develop and test a model of the effects of prevalence in visual search and 3) to test theoretically and clinically motivated strategies to reduce miss errors.