Grant: Discovery Grants Program - Individual

Brain morphometry is essential for the study of the brain response to development, aging, learning, disease and evolution. Thanks to recent advances in Magnetic Resonance Imaging (MRI) acquisition and computational analysis methods, the specificity and sensitivity of detecting morphometric changes is ever improving. Additionally, the adoption of open science practices has made available an unprecedented amount of public data to the community. Anatomical MRI data, in particular high resolution T1-weighted (T1w) images, are perhaps the most widely available neuroimaging data ready for analysis. This treasure trove of datasets opens new avenues for large sample, small effect size analyses, but also calls for improved processing techniques that are more accurate, reproducible and less biased. One critical step in morphometric analyses is brain segmentation, the process of partitioning an MRI of the head into neuroanatomical regions of interests. Current state-of-the-art and state-of-the-practice technology to perform segmentation rely on training (template) data that are not broadly representative of multiple demographics and are often not matched to study subjects. This is particularly problematic as the morphometry of brain structures has been associated with demographic variables such as age, sex, and ethnic background. The research program aims to address this critical challenge by developing a suite of algorithms and data for less biased morphometric analyses, in addition to further improving machine learning techniques for faster and more accurate full brain segmentation. To achieve this goal, we propose to manually annotate new more diverse training data with carefully selected demographics to better capture the general population, and specifically include race and ethnicity information. This large annotated dataset will then be used to develop a deep learning-based segmentation framework, capable of estimating volumetric and surface-based segmentation quickly, with safeguards to mitigate biased and imbalanced training datasets. Our program will improve study reliability and accuracy in neuroimaging studies, especially for underrepresented populations. It will also open new research questions on biases in neuroimaging studies, including complex interactions between sex, gender, age, race, ethnicity, ability, and more. Additionally, addressing biases in brain morphometry analyses will improve the generalizability of studies of the brain in health and disease and better support bridging the gaps between research knowledge and more equitable policy development to maximize positive health outcomes and improve the health of Canadians.