This subproject is one of many research subprojects utilizing the resources provided by a Center grant funded by NIH/NCRR. The subproject and investigator (PI) may have received primary funding from another NIH source, and thus could be represented in other CRISP entries. The institution listed is for the Center, which is not necessarily the institution for the investigator. Over the past decade, multi-center clinical trials utilizing diagnostic imaging modalities have been conducted and sponsored by the National Institutes of Health. Collaborative research is a key part of the 'Roadmap' for the future. When analyzing large diagnostic imaging trial data, assigning voxel labels to different tissue classes or anatomical structures is an important goal. From these studies, apart from binary and categorical data, continuous data are increasingly available, for example, probabilistic segmentation, lesion volume, distance between volume surfaces, percentage of overlapping voxels, and percentage of highly discrepant voxels. Thus, current standards for assessment of imaging systems require task-dependent measures. In this R01 grant proposal, we propose to develop a novel and general statistical validation strategy for evaluating multi-center diagnostic imaging trial data, illustrated on two completed prospective studies previously conducted by the Radiological Diagnostic Oncology Group and the Biomedical Informatics Research Network, respectively.
We aim to validate the accuracy and reliability of the previous studies, with the presence of multi-level factors derived from clustered multi-center diagnostic data. Hierarchical methodology is developed by incorporating the effects of 'spatial' (voxels), 'individual' (patients), 'clusters' (clinical centers), and 'risk strata' (covariates). Receiver operating characteristic analysis, mutual information, overlap index, and the expectation-maximization algorithm will be employed to evaluate diagnostic classification accuracy. These methods may be generalized to many problems related to the analysis of prospective diagnostic trials. The short-term goal of this research is to develop informatics tools to validate diagnostic systems such as breast cancer mammography and functional magnetic resonance imaging. The long-term goal is to develop efficient ways for better analyzing clustered data and utilizing prior knowledge in multi-center clinical trials.
Showing the most recent 10 out of 261 publications
