Under this CAREER Award, a new set of computer-assisted analysis techniques will be developed to improve the noninvasive diagnosis of brain cancer by integrating biochemical and morphological markers from MRSI (magnetic resonance spectroscopy imaging) and MRI (magnetic resonance imaging). MRSI, which allows for characterization and quantification of biochemical metabolites and the construction of metabolite intensity images, combined with MRI provides a biochemical and morphological view of the disease. Using short MRSI echo time techniques, 10-20 dimensional multi-variant feature space will be studied to uncover specific signatures for characterizing cancer. Specific aims include: develop "semi-blind" source separation using a maximum a posteriori framework for recovery of metabolite intensity images in MRSI; characterize the correlations and dependencies between metabolite intensity images and morphological information derived from MRI; develop a hierarchical probabilistic model for integrating metabolite intensity images with MRI for the joint biochemical/morphological characterization of brain tumors; and assess the performance of the models within the context of computer-assisted diagnosis, making comparisons to traditional methods that have relied on fairly elementary relationships, such as the ratio of two metabolite concentrations.
The educational component of the proposal focuses on a program in machine learning for biomedical engineering, including a new course and computer laboratories and efforts that would serve as a basis of an industrial internship program. The course will introduce students to the mathematical theory behind machine learning and probabilistic models, their application to the biomedical sciences, and techniques for evaluating and validating their performance.
