Graphical data mining of human cortical surface morphometry

• Published in: 2013 IEEE 10th International Symposium on Biomedical Imaging pp. 194 - 197
• Main Authors: Van Horn, John Darrell, Joshi, Shantanu H., Bowman, Ian
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.04.2013
• Subjects: Brain; Data mining; dimension reduction; Feature extraction; Magnetic resonance imaging; Neuroimaging; supervised classification; Visual knowledge discovery; Visualization
• ISBN: 1467364568; 9781467364560
• ISSN: 1945-7928
• DOI: 10.1109/ISBI.2013.6556445

This paper illustrates a novel visualization technique for the graphical exploration of large feature-rich brain imaging datasets. An interactive and dynamic OpenGL/Qt-built user interface has been designed for domain experts and students who are non-specialists in informatics, analytics, or data mining. Multi-dimensional scaling projects a full collection of cortical surface representations into three-dimensions, where surface location proximity is proportional to mutual information-based feature similarity. Users can also search over subject meta-data and navigate in the 3D space to group clusters to explore possible trends across data types. This enables users to easily and rapidly generate hypotheses relating cortical surface features and meta-data values. We showcase the usefulness of this novel neuroimaging data-mining approach with an application to data drawn from large-scale MRI archives.