This research plan envisages a methodological advance in functional MRI (fMRI) to allow for adaptive stimulus presentation derived directly from the acquired image data. Adaptation of stimuli will be accomplished by modeling fMRI to classify brain states during the image reconstruction process, and subsequently modulating a visual display. This emphasis on image-based prediction constitutes a fundamental shift from the conventional approach of using temporal changes in images to detect spatial """"""""hot spots"""""""". This research will generate significant insights and development of capabilities for adaptive fMRI experiments using prediction of brain states. This has several significant applications. Primary among these is the potential contribution to designing much more flexible experiments to enhance our basic understanding of brain function. Also relevant are biofeedback rehabilitation, therapeutic meditation, learning studies, sports therapy or other virtual reality-based training, and lie-detection. Moreover this approach will provide spatially resolved data that complements ongoing EEG-based brain computer interface (BCI) research. The experimental plan incorporates a constructive progression that first develops offtine predictive algorithms to a range, of fMRI experminents, secondly treats the case of measurable human learning characterized by offline analysis, and ifinally utilizes these initial studies to characterize system comprising a real-time machine learning algorithm coupled with a responsive human volunteer. Long-term goal: Initiate a research program that will enhance current spatial mapping studies by allowing for temporal classification of brain states based on image data and biofeedback capabilities for adaptive fMRI experiments.
Specific Aims : 1) Characterize the relationship between choice of fMRI task and choice of predictive technique to examine the importance of the particular predictive model, the connection between task difficulty and modeling accuracy, and the amount of training data required to build accurate predictive models. 2) Analyze fMRi data from a motor-learning task to study how behaviorally demonstrated learning by a subject corresponds with changes in the image data, and if this effect is directly observable using predictive models. 3) Develop capabilities to perform real-time feedback of stimulus based on interaction between a predictive algorithm and subject adaptation.
| LaConte, Stephen M; Peltier, Scott J; Hu, Xiaoping P (2007) Real-time fMRI using brain-state classification. Hum Brain Mapp 28:1033-44 |
