Brain MR images offer an unparalleled wealth of information, and promise to enable precise measurement of changes in the brain due to disease and other conditions. Yet, most medical practitioners continue to use simple heuristic image assessments, while the research side of medical image analysis has yet to come up with a clear answer as to which image-processing pipeline is the best. In particular, we observe this for image and surface boundary registration. The essential reason is the overabundance of information, which necessitates any analysis to be model-based, relying on prior knowledge and geometric intuition. Due to a variety of ways to mathematically formalize this intuition, we observe a multitude of proposed solutions: various deformation models and registration regularizers, each offering unique advantages and making specific assumptions.
In this talk, I will showcase some of the solutions proposed by the IGC group. In particular, surface and volume registration methods, and the resulting morphometric measures will be discussed. Next, an application of the image-based measures will be shown for tracking the progression of Alzheimer’s Disease and related conditions. A spatially regularized adaptation of the LDA algorithm is applied to compute a combined MR-based biomarker of AD progression. Power estimates based on the new biomarker suggest better performance compared to traditional volumetric measures. If sufficient time remains, I will also talk about a new framework to combine cortical surface registration and structural connectivity, enabling direct registration of the continuous connectome across brains.