Statistical Inference for Imaging and Disease

PI: Polina Golland, PhD


The Statistical Inference for Imaging and Disease TR&D will develop machine learning and statistical inference methods to capture spatial patterns of cerebrovascular pathology in neuroimaging. To date, most large-scale studies of neurodegenerative and cerebrovascular disease have employed images collected for the purpose of a research study and have focused on volumetric measures as a coarse phenotype to be analyzed and tracked. We aim to overcome these limitations and to develop methods that will produce detailed spatial descriptors of disease from images acquired as part of routine clinical practice. Our approach is to perform image imputation, i.e., reconstruction of anatomically plausible images that are consistent with low-resolution clinical scans. Image imputation will immediately enable applications of state-of-the-art image processing pipelines originally developed for high resolution research scans, leading to accurate spatial normalization and segmentation of the clinical images. We will extract spatial phenotypes of disease by characterizing individuals relative to the patterns observed in the clinical cohort. We will validate and deploy our proposed methods in in the context of a research study that aims to identify patterns of white matter disease in stroke patients that are predictive of the post-stroke outcomes.