Shuichi Matsuzaki

Adaptive classification for brain-machine interface with reinforcement learning

Brain machine interface (BMI) is an interface that uses brain activity to interact with computer-based devices. We introduce a BMI system using electroencephalography (EEG) and the reinforcement learning method, in which event-related potential (ERP) represents a reward reflecting failure or success of BMI operations. In experiments, the P300 speller task was conducted with adding the evaluation process where subjects counted the number of times the speller estimated a wrong character. Results showed that ERPs were evoked in the subjects observing wrong output. Those were estimated by using a support vector machine (SVM) which classified data into two categories. The overall accuracy of classification was approximately 58%. Also, a simulation using the reinforcement learning method was conducted. The result indicated that discriminant accuracy of SVM may improve with the learning process in a way that optimizes the constituent parameters.

Independent component analysis of multi-channel near-infrared spectroscopic signals by time-delayed decorrelation

Multi-channel near-infrared spectroscopy (NIRS) is increasingly used in empirical studies monitoring human brain activity. In a recent study, an independent component analysis (ICA) technique using time-delayed decorrelation was applied to NIRS signals since those signals reflect cerebral blood flow changes caused by task-induced responses as well as various artifacts. The decorrelation technique is important in NIRS-based analyses and may facilitate accurate separation of independent signals generated by oxygenated/deoxygenated hemoglobin concentration changes. We introduce an algorithm using time-delayed correlations that enable estimation of independent components (ICs) in which the number of components is fewer than that of observed sources; the conventional approach using a larger number of components may deteriorate settling of the solution. In a simulation, the algorithm was shown capable of estimating the number of ICs of virtually observed signals set by an experimenter, with the simulation reproducing seven sources where each was a mixture of three ICs and white noises. In addition, the algorithm was introduced in an experiment using ICs of NIRS signals observed during finger-tapping movements. Experimental results showed consistency and reproducibility of the estimated ICs that are attributed to patterns in the spatial distribution and temporal structure.

Temporal representation of arm force direction using fNIRS signals

We investigated the possibility of creating a temporal representation of brain activity from fNIRS signals. In an experiment, subjects performed isometric arm movements in four directions, and fNIRS signals were measured over the primary motor area in the left hemisphere of their brain. We estimated the direction of the arm force from the fNIRS signals by using two classifiers: sparse linear regression (SLR) and support vector machine(SVM). Classification accuracy was approximately 70% with SLR. The temporal distribution of the features selected with SLR was the same as those selected with SVM. The results indicated that the fNIRS signals possibly included information about arm force direction in 4–6 [s] after stimulus onset and offset.

Investigation of brain-computer interfaces that apply sound-evoked event-related potentials

Assessing brain wave functions that are evoked by auditory stimuli is an important area of study that may lead to the development of brain computer interface (BCI) systems that incorporate natural features of auditory perception such as tone, pitch, and sound-source locations (e.g. direction). We analyzed event-related potentials (ERPs) evoked by auditory stimuli that are applicable to BCI systems. In recent studies, sound localization systems have been intensively studied in order to enhance BCI system development in a way that reproduces a virtual 3D auditory environment, applicable to human-machine communications. We conducted experiment using a sound localization system in which subjects were instructed to listen to a sound cue and answering the relative direction (i.e. the direction to which the sound cue is emitted from an observer) of the sound source. For each trial, a target direction was indicated by the experimenter, although the direction of the sound cue emitted during the trials was not necessarily the target direction. Changes in brain activity were measured using an electroencephalogram (EEG) . Experimental results showed that prominent excitations in EEG signals were observed during a trial where the target direction corresponded to the sound source direction, by subtracting the mean EEG signal of the non-target trials from that of the target trials.

Identification of sound-source direction using auditory evoked event-related potentials

The cerebellum has been implicated to form an inverse dynamic model of a control object. Purkinje cells (PCs), the sole output cell type of the cerebellum, are thought to encode motor commands for their control object that are conformed by cerebellar synaptic plasticity to minimize errors in motor performance. Thus a question arises: Can the PCs that control a biological motor system, such as the eyes, acquire appropriate motor commands to drive a non-biological motor system, such as a direct current (DC) motor? The objectives of this study are to answer the foregoing question, and further to evaluate the capability of a single PC in adaptive control of a non-biological system. Conventional single unit recording provides a correlation between unit activities and behavior, but this study addresses direct causality of single unit activities on the resultant behavior. We utilized a vestibulo-ocular reflex (VOR) motor learning scheme. Single unit recording of vestibulo-cerebellar PCs was conducted in goldfish. The desired trajectory of a DC motor was given to the fish in the form of head rotation. The PC firing rate modulated by the vestibular stimulus was converted into a pulse width modulation signal in real time to drive the DC motor. The produced DC motor trajectory was subtracted from the desired trajectory to calculate a performance error. The error signal was fed-back onto the goldfish retina as whole-field visual image motion. This activates VOR motor learning, and lets us evaluate if the PC under recording can improve the performance of the DC motor. The performance error of the DC motor under the VOR motor learning scheme was reduced adaptively. The result suggests that the cerebellum can acquire an internal model of a non-biological motor system, and a single PC is capable of adaptive motor control.

Brain activation during finger tapping and circle-drawing tasks measured by functional near-infrared spectroscopy (fNIRS)

P3-r19 Changes in cerebral blood flow during singing: preliminary fNIRS study Mone Tsukimoto 1 , Yasunori Shiono 1, Tsuyoshi Matsumoto 1, Yasuhiro Kawano 1, Yasushi Fuchigami 1, Kunio Sato 2, Yoko Hoki 1 1 Dept. of Neurophysiol, Division of Neuroscience, Mie Univ Grad School of Medicine, Tsu, Japan 2 Dept. of System Engineering, Division of Environmental Science & Technology, Mie Univ Grad School of Bioresources, Tsu, Japan

Task-related brain activation during finger tapping and circle drawing monitored by fNIRS

The present study assessed brain activation measured by functional near-infrared spectroscopy (fNIRS) in which subjects perform two types of hand motor tasks: a finger tapping task and circle drawing task. For both experiments we employed 24-channel NIRS system covering the left hemisphere (Fig. 1(a)) The typical time course in oxygenated hemoglobin concentration (oxyHb) and deoxygenated hemoglobin concentration (deoxyHb) are shown in Fig. ​Fig.2.2. In finger-tapping tasks, four right-handed subjects performed right hand finger tapping at a rate of approx. 3 Hz paced by a metronome. Subjects performed 12 or 15 trials in which a trial consists of 10-s rest, 20-s finger tapping movement, 30-s rest and about 180-s of inter-task intervals. Changes in oxyHb and deoxyHb were measured at a sampling period of 130-ms. Also changes in the angular velocity of finger tapping, as a measure of the movement smoothness, were measured. Significant decrease in oxyHb and the angular velocity over time were observed, while those transitions further exhibited significant correlation at the channels covering SMA, the pre-motor area (PMA) and PFC (Fig. 1(b)). In circle drawing tasks, six right-handed subjects (males from 22 to 24 years old) performed circle drawing with right hand. Subjects were instructed to trace around a template circle at the frequency of 0.667 Hz. All subjects performed four trials where each consists of 10-s rest, 30-s circle drawing, 30-s rest and about 180-s intervals. We calculated a spatial error as a measure of task performance which is derived from the trajectory deviated from the template circle. We found that oxyHb as well as error rate were decreased over time, with those transitions further be significantly correlated at the channels covering PFC and M1 (Figure 1(c)). Figure 1 (a) Channel setups in fNIRS measurements. (b) Regional activations during finger tapping tasks. Colored area represents where changes in oxyHb were significantly correlated with the task-performance (i.e. tapping smoothness) in all subjects (red), all ... Figure 2 Typical time course of the hemoglobin concentration changes at channel 18 over the prefrontal cortex. (a) The result of finger tapping task at the first trial period (top) and the 15th trial period (bottom). (b) The result of circle drawing task at the ... In summary, this paper reports our empirical results using the two types of hand motor tasks. The experimental results indicated the task-related activity observed in NIRS signals. A previous fNIRS study using hand motor tasks showed similar results including decreased pattern of hemodynamics in SMA [1], whereas further investigation will be desired that may allows to estimate precise cortical functions reflecting motor behaviors.

Insight into Kansei Color Combinations in Interactive User Interface Designing

Color has a major impact on Human Computer Interaction. Although there is a very thin line between appropriate and inappropriate use of color, if used properly, color can be a powerful tool to improve the usefulness of an interactive interface in a wide variety of areas. On the contrary the excessive or inappropriate use of color can severely hinder the functionality and usability of an interface accordingly. A good visual design provides higher level of user satisfaction and further aids with conveying the intended message to its audience. In this paper we focus on one requisite aspect of visual design as such the Color, revealing one hidden dimension of color; Affectivity, by acquiring prospective users concealed color aesthetic preferences, employing Kansei Engineering Assessing System with respect to interactive Interfaces.

Synaptic Plasticity as Agent Communication

Neurons are considered as main computational units of the human brain, are working together with millions of synapses to convey information. The processes of information decoding and neurons’ communication mechanisms are still in a debate. Apart from the numerous researches into those areas, significant attention has given to the synaptic plasticity, which is suspected to have direct relationship with information processing of neurons. As per the biology, synaptic computation can be mainly divided into three plasticity processes, homeostasis, short-term and long-term. The long-term plasticity is considered as the main phenomena related to learning and memory formation; the roles of short-term plasticity and homeostasis plasticity have direct influences to synaptic efficacy and thereby to long-term plasticity. A few researches are being carried out to in cooperate the homeostasis plasticity to Artificial Neural Networks, are still unable to find real integrated mechanism without damaging to learning process. This paper proposes a new model for synaptic computation. In our approach, we understand the neurons as agents consisting of large number of constituent agents those play the roles of synapses, as transmitters or receivers. The statuses of these constituent agents are subjected to homeostasis and short-term plasticity. The number of active transmitters is an in-parameter for the learning processes. With the proposed model, through the active number of transmitters, learning can be explained as integrated process of three plasticity processes.

Decision Making Model Supporting Emergency Medical Care

Accuracy in a pre-hospital trauma triage plays a critical role in reducing trauma mortality in a way that appropriately chooses a patient with severe injuries. Although various triage criteria have been devised and tested, there is no computer-based system developed that helps ambulance teams can make a decision in an appropriate manner. This research proposes to develop Expert Helper, an Expert System which provides a user-friendly environment for the paramedics to decide whether patient to be posted to Critical Care Medical Centers (CCMC).