News & Events

Artificial Intelligence in Brain Micro-Architecture Investigation Using Clinical Diffusion MRI

Abrar Faiyaz

Supervised by Marvin M. Doyley and Giovanni Schifitto

523 Computer Studies Building

Join Zoom Meeting
https://rochester.zoom.us/j/96908763343?pwd=azVFY1lqU1huQ2JMWVRHU3lsVmtZUT09

Meeting ID: 969 0876 3343
Passcode: 068027

This thesis presents a comprehensive study aimed at overcoming significant limita- tions in clinical diffusion MRI (dMRI), with a specific focus on enhancing the resolu- tion of Q-space and the effective use of single-shell data for improved characterization of brain microstructures. The research addresses the problem of inadequate angular resolution in standard clinical diffusion MRI, which often leads to missing crucial clin- ical details. Additionally, it tackles the difficulty of characterizing neurite orientation dispersion and density, which is challenging without multi-shell data but is essential for probing biologically meaningful parameters in the brain.

Central to this research is the hypothesis that clinical dMRI can recover lost mi- crostructural details from limited Q-space resolution in single-shell protocols. This recovery is achieved by integrating Diffusion Tensor Imaging (DTI) and Neurite Ori- entation Dispersion and Density Imaging (NODDI) with advanced Artificial Intelli- gence (AI) techniques, along with the use of relevant multi-modal clinical priors. The study aims to accomplish two primary objectives: first, to develop a sophisticated Q-space up-sampling technique that improves angular resolution using DTI without compromising clinical details; and second, to effectively address the ill-posed problem of single-shell NODDI, culminating in its reliable reconstruction validated through clinical applications.

The research methodology involves a combination of theoretical and practical approaches, including the simulation of the single-shell ill-posed problem of NODDI and the identification of a key parameter, f_ISO, that is instrumental in resolving this issue. Additionally, the study explores the application of multi-modal MR priors for the estimation of f_ISOand investigates the feasibility of applying NODDI in both single- and multi-shell settings. This includes clinical validation in contexts of aging and cognitive performance in a cohort of HIV and Cerebral Small Vessel Disease (CSVD).

In conclusion, this thesis makes a remarkable contribution to the field of clinical diffusion MRI by proposing innovative methodology for enhancing Q-space resolu- tion and effectively utilizing single-shell data, leading to more accurate and detailed characterization of brain microstructures. The findings and methodologies developed have the potential to influence future research and clinical practices in the realm of neuroimaging.