Grant: Discovery Grants Program - Individual

The brain is an organ that is capable of extremely complex functions and behaviours, in large part due to its underlying structure. Magnetic resonance imaging (MRI) provides a unique ability to peer inside and visualize the structure of the brain, and with computational methods, we can examine macro-scale structure with estimates of size and shape of brain structures (computational anatomy), as well as micro-scale structure with estimates of the how microscopic compartments are organized (quantitative MRI, or qMRI). Despite the many advances in macro-scale and micro-scale MRI methods, there are still many applications where abnormal structure in the brain goes undetected. Improved computational neuroimaging methods, especially those that build upon advances in artificial intelligence (AI), could help detect what is currently invisible, but are faced with a number of significant challenges. For one, sample sizes with neuroimaging are relatively small, especially qMRI, and often lack diversity and representation. This leads to models that are difficult to reproduce, or don't work when applied to new, unseen, datasets. Secondly, obtaining the "truth" we need to train AI methods methods is difficult, since that requires examining the tissue under a microscope slide. Finally, the problems we are tackling with computational neuroimaging are complex and require the consideration of structure at multiple scales at once in order to resolve ambiguities. My research program is tackling these issues by proposing new ways to link micro-scale and macro-scale structure, and take advantage of the complementary information from each. In the first theme, Micro2macro, we are building the capability to use microscopy images for training AI models for MRI. In the second theme, Macro2micro, we are making use of population-level data to overcome issues of bias and diversity, by synthesizing new qMRI images using this information. This research will produce new open-source computational tools for interrogating structure of the brain, as well as open datasets that will allow other researchers to benchmark, validate, and build-upon our innovations. Building the "Micro2macro" bridge will enable a vast range of cellular and molecular techniques to inform MRI and enable basic and translational discoveries. The "Macro2micro" bridge will bring significant added value to the many small qMRI datasets that exist and continue to be produced, by transforming them into synthetic population-level datasets. The long-term vision of this program is towards transforming MRI to become a "smart microscope", that is able to see what is hidden to the naked eye, and collectively process this information to make decisions. This proposal will make a significant step towards this vision by providing the means to integrate the micro-scale information from the "few" datasets available, along with macro-scale information from the "many" datasets available.