Shape analysis to assess neurodevelopment and neurodegeneration
Guido Gerig
University of North Carolina
Computer Science
Analysis of brain structures using high-resolution MRI is a technique commonly used to assess morphologic change due to illness, aging, or treatment. However, the timing of these changes, their association with treatment, and an intuitive explanation of morphologic alterations are not fully understood. Imaging studies using structural MRI have most often focused on volumetric assessment of anatomical structures or on voxel-based analysis following image registration. With increasing evidence for structural changes in small subregions and portions of structures, shape analysis may provide important additional information, as morphologic assessment via shape analysis offers a rich set of features.
Phenotyping via noninvasive imaging and statistical shape analysis applied to clinical studies may provide improved insight into pathological processes and drug effects. A central emphasis of our research is to describe shape changes in intuitive terms—such as local growth, atrophy, or bending—in order to explain the type, magnitude, and locality of changes and thereby facilitate associations between pathomorphological alterations and findings at the cellular level.
This lecture discusses surface-based and medial-based shape representations of brain structures (hippocampus, ventricles, and caudate) applied to clinical studies of schizophrenia and autism. In particular, we address statistical issues related to the common problem of high-dimensional samples with low sample size (HDLSS), and compare surface and medial shape representations applied to the same data sets. We also demonstrate shape change over time and over duration of illness by combining shape parametrization with patient variables within a statistical framework. Several exploratory studies indicate that shape measures discriminate patients and controls with greater power than volume alone.
