UCSF

Diffusion magnetic resonance imaging (dMRI) allows us to noninvasively investigate the microstructural properties of brain tissue, and reconstruct the axonal pathways that connect distant brain regions. This enables us to infer the biological processes that give rise to thought and consciousness. However, despite significant advances in both imaging technology and computing power, our ability to estimate connectivity in a single subject using dMRI data remains quite limited. Barriers to accurate single subject estimates include poor accuracy and reproducibility of both fiber tracking and diffusion modeling results, and a difficulty in reproducing the methods of other researchers in this field. As a result, studies using different dMRI methods have drawn conflicting conclusions about the same biological systems. To overcome these barriers, I first present a technique to estimate the noise in dMRI data and show that this measure is a strong indicator of the reproducibility of dMRI measurements. Software engineering principles, such as modularization and thorough testing, were implemented and made publicly available in an open source library called Dipy. By providing a single platform where tools and methods from different developers can be implemented using shared constructs and made publicly available to users, Dipy aims to help the community more easily reproduce the findings of other researchers. In the last section of this work, I use these modeling and fiber tracking tools to reconstruct whole brain networks for individual subjects in a large population. The white matter tissue properties projected onto these networks show that regional differences in white matter integrity are strongly associated with body mass index in young, healthy individuals. This association helps explain the reduced cognitive ability in individuals with higher BMI. This study demonstrates the power of using single subject connectivity networks when studying the human brain and its role in health outcomes. In order to fully unlock the potential of dMRI imaging, methods development needs to continue to focus on improving the reproducibility and accuracy of dMRI studies.