Data Mining from Functional Brain Images

Recent advances in functional brain imaging enable identication of active areas of a brain performing a certain function. Induction of logical formulas describing relations between brain areas and brain functions from functional brain images is a category of data mining. It is di cult, however, to apply conventional mining techniques to functional brain images due to several reasons, such as the di culty of reducing images to symbolic data, possible existence of correlations between adjacent pixels in a image and the limited number of samples available from a single subject. Tsukimoto and Morita presented an algorithm for data mining from functional brain images and showed that the algorithm works well for arti cial data. The algorithm consists of two steps. The rst step is nonparametric regression. The second step is rule extraction from the linear formula obtained by the nonparametric regression. The authors have applied the algorithm to real f-MRI images. This paper reports that the algorithm works well for real f-MRI data and has led to the discovery of certain rules for a nger tapping action and a speech-related action.