April 2: Vince Calhoun, PhD
Brain-Based Biomarkers: A Focus on the Heterogeneous by Enhancing Sensitivity to Brain Disorders and Change
Brain-Based Biomarkers: A Focus on the Heterogeneous by Enhancing Sensitivity to Brain Disorders and Change
Background: The use of neuroimaging to study mental and neurological disorder has shown to be a powerful tool to capture information on the underlying brain changes. However diagnostic heterogeneity is a major issue as the field struggles to learn from the brain imaging data. One important aspect which has not been well explored is the use of rich, high-dimensional brain data to guide us through this complex territory. We show that by focusing on more similar subsets of data, identified via advanced algorithmic strategies, we can facilitate an apples-to-apples comparison, enhance sensitivity to mental illness, and provide a framework for improved stratification.
Methods: We focus on several examples using multiple large N data sets. Our first approach highlights a novel approach which identified ‘statelets’ or homogeneous temporal primitives of transient connectivity patterns in fMRI data. The second approach captures homogeneous subsets of individuals within data driven subspaces in multimodal brain imaging data. We also show that we can leverage this information to refine grouping of individuals, essentially showing where the biological data is pushing against a pre-defined category. And finally, we present some results based on visualization of deep learning approaches that provide insight into possible biological subclusters within existing clinical categories.
Results: Results show that leveraging advanced ‘clustering’ like approaches to identify subsets of data which are more homogeneous within and between subjects groupings enhances our ability to capture neural data which is linked to unique patterns of symptoms. We can also capture new information about how brain disorders impact brain dynamics, for example showing that patients with schizophrenia show much shorter statelet behavior than do controls. And finally, we show that such strategies, perhaps counter-intuitively, enhance our sensitivity to uncover changes in brain that may inform our approach to nosology as well as prove useful targets for future treatment studies.
Conclusions: The brain imaging field has largely focused on group studies, or more recently on individual subject classification using existing categories. We show that a focus on identifying unique subsets of data and subjects which exhibit homogeneity can leverage the benefits of both approaches, increase sensitivity to unique clusters of symptoms, and potentially help refine our understanding of diagnostic categories.
Dr. Vince Calhoun is a recognized expert in developing algorithms to strengthen understanding of brain function, structure and genomics, and how each is affected during various tasks or by mental or neurological illness. He also works to develop neuroinformatics tools that enable experts to use larger data sets and improve efficiency in data capture, management, analysis and sharing.
Dr. Calhoun is the founding director of the Center for Translational Research in Neuroimaging and Data Science, a joint effort between Georgia State, Georgia Tech, and Emory University, which is focused on improving our understanding of the human brain using advanced analytic approaches with an emphasis on translational research such as the development of predictive biomarkers for mental and neurological disorders.