Diagnostic Data Mining for Multi-modal Brain Image Studies

Scientist
Karl Young

Abstract

Large volumes of multi-modal data are typically produced in current imaging studies of brain structure and function. A large number of diagnostic measures are available in such studies, e.g. regional volumes, metabolite concentrations, cortical thickness, white matter mean diffusivity and fractional anisotropy, perfusion levels, etc. How these measures individually provide diagnostic information is often clear, though how to utilize the available information in aggregate to provide improved diagnostic capability is generally not straightforward. A number of relatively recent algorithmic and hardware developments in the area of data mining, such as the Random Forest and Support Vector Machine algorithms and new storage topologies, provide potentially powerful new tools for generating diagnostic information from multi-modal imaging data. This project utilizes robust open source tools, such as the Weka data mining environment, to apply these methods to large data sets from current imaging studies on neurodegenerative diseases.