B.I.G. Summer 2020 Symposium

Daniel J. Tward, in undergraduate student

Ophelia and Jaqueline presented their abstract, Computational Algorithms for Revealing Microstructure in Brain Images with Deformable Registration and Deep Scattering Networks.

We aim to quantify patterns of cell distribution in the brain, by building brain atlases from multiple neuroimages. Because the brain contains information at multiple spatial scales, atlases require alignment of high resolution data using deformable image registration. This calls for downsampling techniques that preserve information while decreasing image size for faster computations. Using novel methods based on the scattering transform, we extracted information from microstructures to produce low resolution images with high feature counts at each voxel. We examined how our downsampling method preserves information by predicting anatomical structures at each location using machine learning algorithms (LDA and random forests). Aligning these images requires a new approach to cross-modality image registration. We developed a method for working with this data, and also tested its performance on single-modality benchmark datasets. These techniques are being used to build better brain atlases, to study diseases and quantify variation in populations..