Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
488407 | Procedia Computer Science | 2016 | 7 Pages |
It is widely known that task-specific analyses are used to understand human brain functioning while performing cognitive tasks. Here, time-series of 3D volumes of Functional Magnetic Resonance (fMR) scans of subjects performing well defined cognitive tasks are utilized. We report a framework for classifying between two distinct cognitive tasks, (a) Viewing picture (b) Reading sentences. In the first phase, the classification ability of each voxel is computed and the best-performing voxels are identified based on an empirical threshold, labeled here as pivotal voxels. In the second phase, voxels that belong to the anatomical regions which lead to the discrimination between the tasks are identified, labeled here as subtle voxels. Active voxels for the respective cognitive tasks are obtained using a t-test; Intersecting active voxels are eliminated in order to obtain discriminating voxels. In our experiments, 80 time-series were used, equally representing the two cognitive tasks. Classification using the Support Vector Machine yielded classification accuracy of 98% using pivotal voxels and 92% using subtle regions, on Leave-One-Example-Out validation scheme.