In this project, we examined several methods for supervised classification. The goal was to distinguish between two classes: (1) the "active" brain and (2) the "resting" brain, based on the fMRI scan.
The preprocessing stage is crucial in such a task. Series of fMRI images of the brain should be converted to the statistical object reflecting the dynamic between the spatial areas. We examined the correlation, partial correlation, and tangent correlation matrices to serve as such an object.
To estimate these matrices, we utilized the Ledoit-Wolf and OAS estimators.
For the active-resting classification task, we compared a support vector machine (SVM) implementation to a fully connected neural network (NN).
Our findings show that NN was able to classify between active and resting state brain with very high accuracy with respect to SVM.
These results can be used for future studies to unveil which areas of the brain are most influential when the subject is performing a task.
Prof. Tsipi Horovitz Kraus
Education, Science and Technology