Abstracts / Neuroscience Research 71S (2011) e46–e107
e97
then converted into an image using the MOS-type 100 × 100 pixel imager (MiCAM Ultima, Brainvision). The pinhole array and pixel imager correspond with each other in terms of pixels. Since the pinhole array is at a fixed position, there are no mechanical disturbances. We tested the microscope by imaging a rat hippocampal slice preparation by bulk VSD staining of the slices (Di-4ANEPPS), single-cell VSD staining, and Ca2+ signal from single cells with bulk staining with a Ca2+ indicator (Oregon green 488 BAPTA-1 AM). Research fund: JST A-STEP, HSR Grants H20-Kagaku-009, MHLW JAPAN.
For the simulation data, when the number of voxels was huge (30,000), iSLR performed better than SVM by avoiding overfitting. iSLR also drastically improved the SLR performance particularly in the case where the number of informative voxels was large. For the real fMRI data, iSLR showed higher performance as compared to the other two, both in binary classification for motor experiment, and in multiple class classification for visual experiment. These results suggest that iSLR can be a robust method for fMRI decoding, by avoiding overfitting and over-pruning, simultaneously.
doi:10.1016/j.neures.2011.07.414
doi:10.1016/j.neures.2011.07.416
O4-G-1-3 Intrasulcal ECoG approach to cortico-cortical connectivity using electrical stimulation-induced evoked potentials in macaques
O4-G-2-1 Single trial classification of phonemes for electrocorticographic brain–machine interfaces
Takahiro Osada 1,4 , Antoine J. Molcard 1,4 , Takeshi Matsuo 2,3 , Keisuke Kawasaki 2 , Yusuke Adachi 1 , Kentaro Miyamoto 1 , Tomomi Watanabe 1 , Isao Hasegawa 2 , Yasushi Miyashita 1 1
2
Dept. Physiol., Univ. Tokyo Sch. Med., Tokyo Dept. Physiol., Niigata Univ. Sch. Med., Niigata 3 Dept. Neurosurg., Univ. Tokyo Sch. Med., Tokyo 4 Equally contributed.
Tetsu Goto 1,2 , Masayuki Hirata 1,2 , Takufumi Yanagisawa 1,3,4 , Kojiro Matsushita 1 , Youichi Saitoh 1 , Haruhiko Kishima 1 , Shayne Morris 1 , Hisato Sugata 2 , Shiro Yorifuji 2 , Toshiki Yoshimine 1 1
Dept. of Neurosurg., Osaka Univ. Grad. Sch. of Med., Osaka, Japan 2 Div. of Func. Diag. Sci., Osaka Univ. Grad. Sch. of Med 3 Department of Neurosurgery, Iseikai Hospital, Osaka, Japan 4 ATR Computational Neuroscience Laboratories
To elucidate brain-wide connectivity in primates, several methods have been developed recently, such as resting state fMRI and diffusion MRI. Corticocortical evoked potential (EP) with electrocorticography (ECoG) is an in vivo method to detect connectivity through direct measurement of neural activity, where electrical pulses are applied and EPs are recorded at distant regions. However, in most previous ECoG studies, although many cortical areas are located inside sulci, probes were placed over the surface of gyri due to technical difficulty of inserting probes within sulci. Here we report a cortico-cortical connectivity study using ECoG probes that were placed both inside sulci and over gyri in macaques. We used silicone-coated electrode arrays (2.5 or 3 mm inter-electrode distance). The probes were bilaterally placed inside the arcuate sulcus (AS) and intraparietal sulcus (IPS), and over the gyri of frontal and parietal cortices in two monkeys, with minimally invasive surgery protocols (Matsuo et al., SfN 2010). Electrical stimulation experiments were conducted under anesthesia with propofol. Single-pulse stimulation was applied in a bipolar fashion to a pair of adjacent electrodes (0.3 ms constant-current pulse, <10 mA). When the anterior bank of inferior branch of the AS was stimulated, EPs were observed focally in intrasulcal and gyral sites in the parietal cortex. The response of the IPS intrasulcal sites showed a lower threshold than that of the gyral sites. Stimulation of the lateral bank of the IPS also induced EPs focally in intrasulcal and gyral sites in the frontal cortex. The response of the AS intrasulcal sites showed a lower threshold than that of the gyral sites. These findings demonstrate that this intrasulcal ECoG approach allows more precise cortico-cortical connectivity mapping, which cannot be captured by the conventional ECoG approach. Research fund: Grant-in-Aid for Specially Promoted Research from MEXT (19002010); a grant from Takeda Science Foundation; SRPBS from MEXT; JSPS Research Fellowship for Young Scientists (234682).
Since the activated brain regions in language process are broad and include common responses, it is difficult to find robust features for single trial classification. The aim of this study is to confirm the feasibility of a method for single trial classification of phonemes for an electrocorticographic brain–machine interface. We obtained electrocorticogram in seven patients with brain tumor, intractable epilepsy, or intractable pain, who were implanted with subdural electrodes on areas over the inferior part of precentral gyrus (IPrG) and/or the inferior frontal gyrus (IFG) due to the need for treatment. The subjects pronounced either five (n = 6) or nine phonemes (n = 1). In case of five phonemes, the subjects voluntarily pronounced five Japanese vowels (a, i, u, e, o) repeatedly with an interval of more than 3 s, and the trial was terminated when the least pronounced vowel was repeated twenty five times. The pronounced phonemes were inferred from the potentials or power changes in frequency bands by a support vector machine. We used a sliding period of interest in steps of 50 ms. Decoding accuracies were calculated in each period of interest and were also calculated for arbitrary selections of three phonemes. The average of the best accuracy rate in each subject was 54.9% (S.D. 10.9) for 3 phonemes, 32.3% (S.D. 8.0) for 5 phonemes, and 18.6% for 9 phonemes. Accuracy increases before pronunciation and peaks in the periods of interest including both before and after pronunciation. The electrodes of high contribution for decoding were shown to be those at the IPrG, the IFG and also dorsolateral prefrontal cortex. We examined the feasibility of single trial classification of phonemes for an electrocorticographic brain–machine interface. Our findings may also contribute to clarifying neural processes of pronouncing phonemes. Research fund: 20768.
doi:10.1016/j.neures.2011.07.415
O4-G-2-2 Development of an EEG-based brain–computer interface suitable for use during simultaneous fMRI acquisition
O4-G-1-4 Iterative Sparse Logistic Regression (iSLR): A new ensemble pattern classification method for fMRI decoding Satoshi Hirose 1,2 , Isao Nambu 3 , Eiichi Naito 3,4 1 4
ATR-CMC, Kyoto, Japan 2 JSPS, Japan 3 NICT, Brain ICT Lab., Kyoto, Japan Grad. Sch. of Medicine, Osaka Univ., Osaka, Japan
In recent human functional magnetic resonance imaging (fMRI) studies, multivoxel pattern classification is now becoming popular to predict contents of sensory and motor events from brain activity (fMRI decoding). In this approach, one of the general methodological concerns is overfitting; poor classification performance is achieved under the presence of too many voxels. Recently, Sparse Logistic Regression (SLR: Yamashita et al., 2008) algorithm has successfully dealt with this problem, but SLR has another weak point of over-pruning where many useful voxels for classification can be discarded. Here, we propose a new classification algorithm; iterative SLR (iSLR), which incorporates SLR with ensemble learning. We iterated SLR to generate subclassifiers using the voxels discarded by preceding SLR, and these classifiers were united as a meta-classifier to predict contents of sensory and motor events. We evaluated the usefulness of iSLR by applying this to simulation and real fMRI data, and compared its performance with those of SLR and Support Vector Machine (SVM).
doi:10.1016/j.neures.2011.07.417
Charles S. DaSalla 1,4 Takashi Hanakawa 1,3
, Kazumi Kasahara 2,4 , Manabu Honda 1 ,
1 Department of Functional Brain Research, National Institute of Neuroscience, National Center of Neurology and Psychiatry, Tokyo, Japan 2 Department of Radiological Sciences, Graduate School of Human Health Science, Tokyo Metropolitan University 3 PRESTO, Japan Science and Technology Agency 4 These authors contributed equally.
Brain–computer interface (BCI) technology continues to evolve, with researchers finding increasingly effective ways of interpreting brain activity in order to improve accuracy and throughput of brain-to-machine interactions. Less studied, however, are the afferent effects that BCIs incur on the user and the changes that the brain undergoes to consolidate BCI control. Through this ongoing study, we seek to investigate the role of neuroplasticity in BCIs and, specifically, how people adapt to using the technology. Understanding these effects may also provide insight into better BCI designs, tailored to the ways in which humans learn. With these purposes in mind, we performed simultaneous fMRI-EEG recordings to obtain multimodal imaging data during the control of a motor imagery-based BCI. Gradient and ballistocardiogram artifacts in the EEG data were corrected in real-time using modules designed for the BCI2000 software environment. Mu and beta rhythmic activities were extracted from the corrected EEG data stream and applied