Deep-Learning-Augmented Quantitative Gradient Recalled Echo (DLA-qGRE) MRI for in vivo Clinical Evaluation of Brain Microstructural Neurodegeneration in Alzheimer Disease

Washington University, Saint Louis MO

Investigators

Linked publications, trials & patents

Abstract

Alzheimer Disease (AD) is one of the major health problems in the US and worldwide; it is a neurodegenerative disorder that is characterized clinically by progressive dementia caused by pathological changes in brain tissue preceding clinical symptoms by 15-20 years. Clinically-accessible methods are critically needed to screen for early AD pathology and monitoring it over time, as well as for outcome measures in clinical drug trials. The goal of this grant application is to establish an MRI-based technique, Deep-Learning-Augmented quantitative Gradient Recalled Echo (DLA-qGRE), as a platform for quantitative clinical evaluation of brain tissue microstructural neurodegeneration at early preclinical stages of Alzheimer Disease (AD). DLA-qGRE is a combination of qGRE MRI technique and Regularization by Artifact REmoval (RARE) deep learning (DL) methodology, both developed by our team. qGRE data obtained from a well-characterized cohort of patients revealed the existence of brain regions with low R2t* values (Dark Matter), representing tissue essentially devoid of neurons. These data show that Dark Matter can be identified already in people with preclinical stages of AD (amyloid positive but without clinical symptoms) and also has a predictive power of future AD progression. While qGRE sequence can be implemented on any commercial MRI scanner, the data analysis currently requires hours of computing time, tempering clinical applications. To significantly accelerate and improve data analysis, as well as data acquisition, in this proposal we will use innovative RARE technique, a DL approach that explicitly accounts for the physical models of specific imaging systems and biophysical models of biological tissues. Preliminary data show that DL has a potential for reconstructing qGRE metrics in a matter of seconds with improved image quality and reduced noise. This opens opportunity for implementing DLA-qGRE as a widely available tool for clinical applications. Based on this approach, we plan to achieve the following Specific Aims: In Aim 1 we will develop DLA-qGRE data processing pipeline, compatible with MRI protocols of commercially available GRE sequences, for fast and reliable detection of microstructural pre-atrophic neurodegeneration. In Aim 2 we will optimize k-space sampling strategy for developing qGRE imaging protocol with increased isotropic resolution and simultaneously decreased MRI acquisition time. Reducing scan time will significantly help with patient comfort, be much less susceptible to motion, and reduce costs of the MRI exam. In Aim 3 we will demonstrate that in a clinical neuroradiology setting DLA-qGRE compatible with MRI protocols of commercially available GRE sequences (developed per Aim 1), and accelerated DLA-qGRE (developed per Aim 2), can reliably detect microstructural neurodegeneration in preclinical and early symptomatic AD. In Summary, successful completion of the aims of this proposal will open doors for using DLA-qGRE in clinical settings as novel and more sensitive and specific MRI-based diagnostic measure of the neurodegenerative aspects of early AD pathology as compared with current measurements of tissue atrophy.