Observer Studies Involving Search: Modeling and Analysis

涉及搜索的观察者研究：建模和分析

• 批准号: 7103678
• 负责人: DEV P CHAKRABORTY
• 金额: $ 32.63万
• 依托单位: UNIVERSITY OF PITTSBURGH AT PITTSBURGH
• 依托单位国家: 美国
• 财政年份: 2005
• 资助国家: 美国
• 起止时间: 2005-08-01 至 2009-05-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/7103678
• 关键词: bioimaging /biomedical imaging; computer assisted diagnosis; computer program /software; computer system design /evaluation; diagnosis quality /standard; health care personnel performance; image processing; lung neoplasms; mathematical model; model design /development; radiography

DESCRIPTION (provided by applicant): Since the inception of the paradigm in 1961 the analysis of FROC (free-response receiver operating characteristic) data (i.e., mark-rating pairs) has remained an unsolved problem in imaging science. Since the radiologist must search patient images for localized lesions, intrinsically FROC data is generated in virtually all imaging studies in radiology. Currently there is great interest in evaluating different computer aided detection (CAD) algorithms for lung cancer screening with helical CT scans. This task also involves search and depending on the CAD in use there are approximately 4 to 20 suspicious areas reported per patient While in the past the receiver operating characteristic (ROC) paradigm has been the standard method for the evaluation of diagnostic imaging systems, its limitation to one report per patient is now being increasingly felt. This is a proposal to develop the science and perform the requisite testing of the FROC paradigm by accomplishing the following aims. (1) Develop a statistical model for FROC data that includes the effect of search and other relevant perceptual factors and which reduces to the standard ROC model in non-search tasks. (2) Use the model to validate and evaluate currently proposed methods for analyzing location and ratings data, and in particular determine their statistical power. (3) Develop an algorithm to determine the parameters of the model from clinical FROC data and to make statistical inferences regarding the modalities being tested; verify that the estimated search parameters correlate with values inferred from eye-movement recordings. (4) Provide practical and validated software to the user community that will offer more statistical power and therefore require fewer resources of cases and readers. The significance of the proposed work is that on the scale of previous developments, the potential gain in statistical power by the FROC method is very large. Also, the explicit modeling of search and the ability to estimate search related parameters from observer data should open up new areas of vision research that all researchers can exploit.

描述(由申请人提供)： 自1961年该范式问世以来，FROC(自由反应接收器工作特性)数据(即标记-评级对)的分析在成像科学中一直是一个悬而未决的问题。由于放射科医生必须搜索患者图像中的局部病变，因此在放射学几乎所有的影像研究中都会生成FROC数据。目前，利用螺旋CT扫描对不同的计算机辅助检测(CAD)算法进行肺癌筛查有很大的兴趣。这项任务还涉及搜索，根据所使用的CAD，每个患者大约报告4到20个可疑区域，而在过去，接收器操作特征(ROC)范例一直是评估诊断成像系统的标准方法，现在越来越多地感受到其对每个患者一份报告的限制。这是一项发展科学并通过实现以下目标对FROC范式进行必要测试的建议。(1)建立FROC数据的统计模型，该模型包含搜索和其他相关感知因素的影响，并在非搜索任务中简化为标准ROC模型。(2)使用该模型验证和评价了目前提出的位置和评级数据分析方法，特别是确定了它们的统计能力。(3)开发一种算法，从临床FROC数据中确定模型的参数，并对正在测试的模式进行统计推断；验证估计的搜索参数与从眼动记录中推断的值是否相关。(4)向用户社区提供实用和有效的软件，这些软件将提供更多的统计能力，因此需要更少的案例和读者资源。这项工作的意义在于，在以往发展的规模上，FROC方法在统计能力方面的潜在收益是非常大的。此外，搜索的显式建模和从观察者数据估计搜索相关参数的能力应该会打开所有研究人员都可以开发的视觉研究的新领域。