Integrating Histology and Microarchitecture Modeling with Deep Learning for Diffusion-Weighted Magnetic Resonance Imaging

Abstract

The human brain is one of the most complex organs to understand in terms of anatomy and microstructural tissue properties of the brain’s white matter. There is still a lack of fundamental understanding for various neurological disorders in terms of the microstructural tissue properties of the brain. A critical imaging modality: Diffusion-Weighted Magnetic Resonance Imaging (DW-MRI) has become a key insight for probing the in-vivo tissue organization of the underlying microstructure and microarchitecture for the human brain. DW-MRI was proposed in the early 1980’s and has rapidly grown as a field; due to it being the only imaging modality that can provide information about tissue microstructural properties in-vivo. In brief, contributions are briefly summarized: 1.) We characterize the empirical reproducibility of microstructural measures; due to the existence of a large number of approaches that reconstruct the microstructural measures and white matter nerve tract reconstructions as there was a lack of intra-method reproducibility validation. 2.) The empirical validation led us to propose data-driven novel methods using rare animal study datasets. Validation of microstructural measures has traditionally been approached using phantom based studies. 3.) Inter-site harmonized reconstruction of microstructure measures was proposed by the addition of auxiliary data to the rare animal dataset so that effective statistical analysis can be conducted on inter-site data. 4.) Promising results on the prior contributions led us to the proposition of recovery of 3D microstructural measures from 2D microscopic histology slides. 5.) Data-driven modeling in general from the prior contributions (2,3 and 4) led the motivation for learning across the manifolds of single and multi-shell DW-MRI as the acquisitions vary by multiple parameters and the capabilities of the scanning hardware.