The research proposed herein aims at obtaining robust estimates of diffusion representations (images, tensors, spectra) from diffusion-weighted magnetic resonance (MR) data, by compensating for the high levels of noise and distortions in the data. Although the algorithms will be widely applicable to diffusion MRI, the application of interest is the imaging of cerebral white-matter structures. The proposed approach is that of a penalized likelihood (PL) framework, where the diffusion representations are estimated by maximizing an objective function that consists of a likelihood term that fits the solution to the raw MR data plus a regularization term that penalizes overly noisy solutions. The algorithms will utilize the raw time-domain data from the scanner, avoiding the oversimplified Fourier transform data model. The first components of the framework, involving a PL approach to tensor estimation with magnetic field inhomogeneity correction, are being prototyped and will be completed during the mentored phase of the award. In later stages, these components will be incorporated in diffusion spectrum estimation. In parallel to development, high-resolution ex vivo data will be used as a gold standard to evaluate the methods and optimize the relative weighting of the likelihood and regularization terms, i.e., the amount of smoothing. The project fits the candidate's long-term career goal of establishing a high-quality independent research program on inverse problems in medical imaging that spans different modalities. It will also facilitate the candidate's immediate goals of becoming an expert in diffusion MR data analysis and advancing this field by translating the skills acquired in her previous work in statistical reconstruction for emission tomography. The mentored phase will be performed at the MGH/Harvard/MIT Martinos Center for Biomedical Imaging. The candidate will take advantage of the cutting-edge MRI facilities and expertise at the Center, as well as the world-class educational opportunities at its collaborating institutions. Her career development plan includes training in MR data acquisition; consultations with experts of the field; coursework in MR physics and neuroscience; seminars and scientific meetings. As part of launching her own independent research program, the candidate will mentor a graduate student who will be expected to contribute to this project. Relevance: Information extracted from diffusion-weighted MR data is used in medicine, e.g., to monitor brain function in stroke patients; to detect the effects of diseases such as schizophrenia, multiple sclerosis and Alzheimer's; to assess newborn brain development; and to research connectivity of brain regions. The long- term objective of this work is to develop algorithms that enhance the quality of the measures estimated from diffusion-weighted MR data. As such, it has the potential to benefit this wide and growing range of medical applications and promote important areas of public health.
