NSF/FDA SIR: Numerical Model Observer for Multiple Abnormalities

点击翻译按钮获取中文摘要

PI: Markey, Mia K.Proposal: 1445713Title: Numerical Model Observer for Multiple AbnormalitiesSignificanceThe proposed model observer would enable more rapid optimization of medical imaging systems, especially during the early stages of the technology development. The underlying paradigm is to quantify the quality of a medical image by its effectiveness with respect to its intended clinical task. This aspect of is aligned with the mission of the FDA to facilitate new, safe, and effective medical devices. Of further importance is the development of a new research collaboration between Prof. Mia K. Markey and her graduate students at The University of Texas at Austin and researchers in the Division of Imaging and Applied Mathematics at the FDA, especially Dr. Subok Park. In addition to integrating research and education through the experiences of the graduate student supported by this project, the PI (Markey) anticipates that this work will inform her teaching such that she can improve biomedical engineering course offerings at The University of Texas at Austin with respect to topics that are particularly relevant to the Division of Imaging and Applied Mathematics at the FDA.that this work will inform her teaching such that she can improve biomedical engineering course offerings at The University of Texas at Austin with respect to topics that are particularly relevant to the Division of Imaging and Applied Mathematics at the FDA.Technical DescriptionThe proposed collaborative research is designed to provide a structured opportunity for a graduate student to conduct engineering and scientific research at the FDA focused on improving the design of numerical model observers for medical imaging research. As a surrogate of human observers, numerical model observers provide an objective method for assessing medical image quality, and, thus, play an important role in the optimization and assessment of imaging devices. The FDA needs model observers in order to develop regulatory criteria and identify scientific issues that imaging device manufacturers must address. In clinical practice, there may be zero, one, or more than one abnormality that a radiologist needs to find on a medical image. The objective of the proposed study is to develop a model observer to characterize the detection of multiple abnormalities in a medical image and to estimate the properties of the abnormalities. This is critical for enhancing the generality of model observers and their practical application in medical imaging since current state-of-art model observers are unable to handle multiple abnormalities in a single image. The main idea of the proposed model is to compute the likelihood of an ensemble of suspicious locations, estimate the parameters of each detected signal, and then assemble all the information into a final score that reveals the most suspicious scenario.