DIAGNOSTIC TASK SIMULATION WITH MULTI-ELEMENT PHANTOMS

点击翻译按钮获取中文摘要

This project will explore a new method of measuring the relative diagnostic accuracy of imaging systems. The current widespread use of clinical images has several drawbacks, among which is the extensive duplication of effort, since results obtained with clinical cases at one site are difficult to extrapolate to other sites. We will simulate clinical cases with a multi- element phantom. A set of weights, one for each phantom element, will be found with multivariate linear regression techniques. The weights will be such that when relative observer performance measured on the phantom elements is appropriately combined, the result is a prediction of the relative clinical performance. In this study we will study the applicability of the simulation methodology for two clinical tasks, pneumothorax and nodule detection. We will test the hypothesis that relative performances predicted with the multi-element phantom, and directly measured with clinical cases, will agree with each other. A consequence of this is that the same rankings will be obtained for different imaging systems whether one measures them with clinical cases or with phantoms. The project is significant as the phantom may be used by other institutions, and this will lead to more easily generalizable work. Also, the weights may be determined for any multi-element phantom, not just the one to be developed by us. Consequently, existing phantoms can be used more quantitatively, leading to better standardization of image quality measurements. The proposed work is equivalent to developing a secondary standard for evaluating imaging modalities, and it is expected to complement the existing primary standard, namely, clinical cases.