| dc.description.abstract | The amount of radiological imaging data being generated is growing at an unprecedented pace. Concurrently, analytical tools in image processing and big data are making it feasible to study large amounts of data to better quantify and understand aging and its associated diseases. Aging is largely considered a complex, multi-systems process by which bodily functions change over time. Advances in imaging data acquisition and processing allow us to directly study the effects of aging on brain anatomy and quantify structural changes associated with diseases of aging. In this work, we present the following contributions to the study of aging in neuroimaging through data-driven metrics:
First, we generated a normative metric of brain aging by predicting chronological age from brain magnetic resonance imaging (MRI) in healthy individuals. We showed improved accuracy and robustness by incorporating anatomical context in age prediction. We then showed a clinical applicability of this metric in a geriatric population with depression. Second, we showed a technique to improve the generalizability of whole brain segmentation on heterogeneous MRI acquisitions, including imaging with intravenous contrast, resulting in more accurate volumetric estimates. Lastly, we use this robust segmentation method to study the brain changes associated with cardiovascular disease on clinically-acquired neuroimaging.
Together, this work expands and contributes to the development of data-driven metrics on neuroimaging to study aging and its associated diseases. These contributions additionally provide tools capable of improving generalizability and robustness of medical image analysis algorithms on heterogeneous clinical imaging. | |
