There is a substantial range of intra-specific variation in the structure of the human brain. However, the variation of different regions of the brain is not independent, and studies have demonstrated that there are significant correlations in the shape and structure of various regions. This phenomenon is called “structural covariance” and has been increasingly studied in recent years. It has been hypothesized that these correlations are (among other considerations) the basis of functional connections between regions. For example, if several regions are known to be linked to memory, then their volume might be correlated. It has also been shown that these correlations are affected by several neurological disorders such as Alzheimer’s disease, Schizophrenia and Autism.
Until now, most research into structural covariance has been carried out using scalar values such as cortical thickness or grey matter density. Representing the intricate structure of the brain with a single value inherently ignores some of the underlying complexity of its structure. On the other hand, defining and quantifying such correlations between brain regions without using scalar values is difficult. Thus we propose a novel way to study structural covariance, using deep learning models. Our goal is to train deep learning models to predict the structure of one region from the structure of other regions. This approach would allow us to study what information the structure of one region of the brain potentially contains about other regions. In this manner we can model complex relations that are not as easily captured by standard statistical techniques.
We hope that this research will improve our understanding of the phenomenon of structural covariance, which could lead to improvements in our understanding of the constraints on brain development, evolution, and the effects of certain diseases on brain structure and connectivity. We request access to the imaging data for a period of 2 years.
