Mind reading: An automated classification of thought processes from imagery fMRI data

Abstract

Automated interpretation and classification of functional MRI (fMRI) data is an emerging research field that enables the characterization of underlying cognitive processes with minimal human intervention. In this work, we present a method for automated classification of human thoughts reflected on an event-related paradigm using fMRI modality with significantly shortened data acquisition time (less than a minute). Six distinct thoughts (right-hand motor imagery, left-hand motor imagery, right-foot motor imagery, mental calculation, internal word/speech generation, and visual imagery) were chosen as target tasks. Five healthy volunteers performed the tasks. The regions-of-interest (ROIs) were delineated from the activated regions that were consistently and exclusively activated during the training phase. Extracted feature vectors of activations were recognized using a support vector machine (SVM) classifier. With a parameter optimization using a cross-validation scheme, the classifier successfully recognized the six different categories of the given thought tasks with above 80% average accuracy from four participants. The proposed automated processing method of short-time fMRI data can be utilized for the further investigation of monitoring/identifying of human minds and their possible link to the hardware and computer control.