Observer Studies Involving Search: Modeling and Analysis

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): The research involves measuring imaging system performance in tasks such as detecting breast cancer. Receiver operating characteristic (ROC) methodology, the current gold-standard, uses patient-level information that a woman has suspected breast cancer. The location-specific free-response ROC (FROC) method uses additional location-level information in the radiologist's report, e.g., the cancer is in the left breast and is present at a particular location. Progress during the funded period has resulted in a novel perceptually-based FROC model and data simulator and several validated methods for analyzing data which are applicable to human observers and computer aided detection (CAD) algorithms. Papers using the PI's ideas and software are being presented in increasing numbers at conferences and in journals, and his work has generated healthy debate. The overall goal of the competing renewal project is to continue advancing the state-of-the-art in this field by addressing a number of limitations of current methods. Specific Aim 1: The figure-of-merit (FOM) is a critical determinant of statistical power and clinical relevance but all current FOMs are lesion-based and cases with more lesions contribute more to the FOM than cases with fewer lesions, and clinically less important lesions contribute equally as more important ones; we will develop novel case-based FOMs that overcome these limitations. Specific Aim 2: A realistic simulator yields confidence in methodology validation using that simulator. We will extend the current simulator by incorporating more realistic correlation effects and we will develop methodology to calibrate the simulator to real datasets thereby allowing the methodology developer to tune the simulator to specific applications. The simulator will be used to validate the different methods of analysis developed in Aim 1. Specific Aim 3: We will address several practical issues with current FROC methodology: arbitrariness of the proximity criterion, i.e., how close a mark must be to a lesion in order to credit the observer for a true detection; lack of sample-size estimation methodology for planning prospective studies; and lack of methods for analyzing clinically realistic data acquisition scenarios such as multiple views and breasts and multiple lesion types per case. Specific Aim 4: We will validate the methodology using independently acquired ROC, FROC and outcome-data in mammography. Outcome is defined as GOOD for normal cases returned to screening or abnormal cases sent to biopsy and BAD otherwise. We will test the hypothesis that FROC better correlates with outcome and yields greater statistical power than ROC. The significance is that the field is increasingly moving towards location-specific analyses, because of its intrinsic appeal and clinical realism, therefore methodology capable of analyzing the complex data, well outside the scope of the current gold-standard, is urgently needed. Patients benefit from better designed and optimized equipment leading to early diagnosis and treatment of cancers. Health care benefits because more efficient and cost-effective studies become possible which could serve as surrogates for expensive clinical trials. PUBLIC HEALTH RELEVANCE: The research involves measuring imaging system performance in tasks such as detecting breast cancer where the current gold-standard uses information that a woman is suspected to have breast cancer, but the proposed new method uses additional information, e.g., the suspected cancer is in the left breast at a particular location, thereby yielding a more precise and clinically relevant measurement of performance. Patients benefit from better designed and optimized equipment leading to early diagnosis and treatment of cancers. Health care benefits because more efficient and cost-effective studies become possible which could serve as surrogates for expensive clinical trials.