Classification of cognitive state using statistics of split time series

Functional MRI (fMRI) data comprises of a set of trials, each trial is described in terms of a group of 20 to 25 anatomical Region Of Interests (ROI). Each ROI consists of neuroimage sequence information in terms of a set of voxels. Extracting features from ROIs and classifying cognitive states is a challenging task. In this work, average of voxel time horizon for each ROI is considered as an input to the Support Vector Machine (SVM) classifier, which gives the classification performance of each ROI. The proposed method is discussed in the context of ROIs in Starplus fMRI data, and classifier performance of each ROI has been analysed. The ROI, CALC and the combination of ROIs CALC & LIPS provides a good classification accuracy compared to remaining ROIs in Starplus fMRI data. Features formed using second half of time horizon of voxel time series gives a good classification accuracy and it is consistent over six subjects in Starplus fMRI data.