Classifying Cognitive Tasks Based on Brain Activity
Our brains enable us to interact with the outside world. Although the question of the underlying mechanism still remains unsolved, it is clear that the brain can give rise to diverse mental states when dealing with different real-world situations. With the advent of neuroimaging techniques, these latent cognitive constructs can be observed directly by measuring cortical activity patterns. Therefore, the research question is now turned into how elements of cognition are associated with the neural activities inside the brains. I employed support vector machine and neural network approaches to classify the task labels based on the fMRI scans and identify multi-variate patterns of brain activities underlying each cognitive task. The SVM analysis achieved good ability in classifying 12 cognitive tasks based on the fMRI scans and confirmed the grouping of cognitive tasks with four broader previously identified “reference abilities” (Salthouse et al. 2008). The classification accuracy for a single task varied from 0.42 to 0.72, and the reference ability level classification achieved high accuracies over 0.8. Our work confirmed past evidence for distinct cortical coding of distinct cognitive reference abilities, and further showed evidence of distinct cortical coding of some nominally ‘same-ability’ tasks. Novel use of anatomical constraints and dimensionality reduction in a two-layer neural network allowed for the identification of new factors in cognition and helped identify new brain activity patterns for each cognitive task. Two feature reduction methods were applied to deal with the overfitting problem. I developed a new searchlight-based feature reduction method that improved the performances of both classifiers and showed an exciting potential promise for more stable cortical learning results. The t value-based feature reduction revealed the inconsistency between the statistical parameter values and the discriminative information stored in voxels.
Zhang, Tianyu, "Classifying Cognitive Tasks Based on Brain Activity" (2020). ETD Collection for Fordham University. AAI28000076.