[ 戻る ]
Neuro Feedback fMRI
Neurofeedback (NF) training (NFT) using motor imagery (MI) tasks has been recently developed as a rehabilitation tool after cerebrovascular disorders [1]. NFT is a self-modulation method to control mental training more efficiently. NFT is also an indispensable part of a brain-computer interface (BCI) to improve its efficacy. It will be useful to match the body representation of movements with the target outputs to better manipulate machines or devices as a part of one's own body.
Real-time functional magnetic resonance imaging (fMRI) is an emerging technique for the assessment of the dynamic and robust changes in brain activations during an ongoing experiment. With real-time fMRI, several processes within the brain could be measured as they happen. The extracted information could then be used to monitor the quality of the acquired data sets, serve as the basis for neurofeedback training, and manipulate scans for interactive paradigm designs, among others. Although more work still needs to be done, recent results have demonstrated real-time fMRI’s potential applications both for research and clinical use. In this review, we will discuss these developments and focus on the methods enabling the real-time analysis of fMRI data sets, novel research applications arising from these approaches, and the potential use of real-time fMRI in clinical settings.
The basic technique for NR-fMRI is real-time fMRI refered above, which supplies activation status during task performance. We developed rt-fMRI using a general linear model, which is the standart method to compute activation maps, and a PC cluster.
1.4 Development of NF-fMRI
Neurofeedback (NF) training (NFT) using motor imagery (MI) tasks has been recently developed as a rehabilitation tool after cerebrovascular disorders (Naros 2015). NFT is a self-modulation method to control mental training more efficiently. NFT is also an indispensable part of a brain-machine interface (BMI) to improve its efficacy. It will be useful to match the body representation of movements with the target outputs to better manipulate machines or devices as a part of one's own body. In this study, we designed a BMI model using a MI task to manipulate a small humanoid which gives visual representation of MI as a NFT. By using fMRI, the brain activations with and without NFT were compared to investigate the performance to discriminate brain maps to send commands to the output device.
In the NF experiment, the performance rates of generating robot arm movement (correct volume classification / total number of volumes) were 70 to 80%. These rates did not significantly change across the three sessions on each side (Table 1). Conjunction analysis showed activation in the bilateral supplementary motor area (SMA; [-6 -4 62]) in both NF and N-NF experiments. Activations detected during NF session included: NF1) right middle temporal gyrus (MTG [50 -64 -2], T = 5.56,), left putamen ([-22 -2 4] , T = 4.66), inferior parietal lobule (IPL) and right superior temporal gyrus (STG [48 -32 22], T = 4.03); NF2) right STG, IPL ([62 -32 18], T = 5.74) and right IFG ([54 2 44], T = 4.84); and NF3) right IPL, STG ([58 -30 16], T = 4.91) and left putamen ([-18, -8, -4], T = 4.99)(Figure 1). Comparison between NF and N-NF (Rt and Lt) showed greater activation in the left superior temporal gyrus (STG) during the NF1 (Figure 2). No significant difference in grasping power was observed before and after the fMRI sessions in both experiments (paired-t test).
Functional MRI represents a new modality for NF training (Chapin 2012, Weiskopf 2012) and could play an important role in exploring the underlying neuronal principle of NFT. A necessary condition for NFT is the real-time, frame-by-frame analysis of functional images during tasks. Here, we used SVM to classify functional images and to generate the corresponding BMI commands during NF scans. Our result showed that SVM's classification performance was consistent across the three feedback runs. Moreover, we observed consistent activations in the STG throughout the NF sessions, which may suggest the stronger involvement of the salience network in generating MI by integrating sensory representation of body and movements. In conclusion, we have demonstrated the potential use of SVM to translate MI brain states to a corresponding small humanoid movement for NF training.
[1] Bagarinao E. (2015) ‘Real-time classification of brain activation patterns using support vector machine and its potential application as a tool for cognitive training’, The 38th Annual Meeting of the Japan Neuroscience Society, #1O 07-2-3.
[2] deCharms, R.C. (2004) ‘Learned regulation of spatially localized brain activation using real-time fMRI’, NeuroImage, 21:436-443.
[3] Chapin, H., Bagarinao, E., Mackey, S. (2012) Real-time fMRI applied to pain management. Neuroscience Letters, 520:174-181.
[4] Naros, G. (2015) ‘Reinforcement learning of self-regulated β-oscillations for motor restoration in chronic stroke’, Frontiers in Human Neuroscience, 9:391.
[5] Weiskopf, N. (2012) ‘Real-time fMRI and its application to neurofeedback’, NeuroImage, 62:682-692.