Neurofeedback of two motor functions using supervised learning-based real-time functional magnetic resonance imaging.

This study examines the effects of neurofeedback provided by support vector machine (SVM) classification-based real-time functional magnetic resonance imaging (rt-fMRI) during two types of motor tasks. This approach also enables the examination of the neural regions associated with predicting mental states in different domains of motor control, which is critical to further our understanding of normal and impaired function. Healthy volunteers (n = 13) performed both a simple button tapping task, and a covert rate-of-speech counting task. The average prediction accuracy was approximately 95% for the button tapping task and 86% for the speech task. However, subsequent offline analysis revealed that classification of the initial runs was significantly lower - 75% (p<0.001) for button and 72% (p<0.005) for speech. To explore this effect, a group analysis was performed using the spatial maps derived from the SVM models, which showed significant differences between the two fMRI runs. One possible explanation for the difference in spatial patterns and the asymmetry in the prediction accuracies is that when subjects are actively engaged in the task (i.e. when they are trying to control a computer interface), they are generating stronger BOLD responses in terms of both intensity and spatial extent.