Faculty of Environmental and Information Sciences
Ph. D / Professor
Department of Fundamental Science and Materials, The Graduate School of Science and Technology, Kobe University (Doctoral Program)
Researcher of Center of Excellent, Osaka University, Researcher of RCNP, Osaka University, Lecture, Fukui University of Technology, Assistant Professor/Associate Professor, Fukui University of Technology / Branch chief of Hokuriku area, SICE (2018)
Nonlinear classification by machine learning, Mathematical modeling and quantitative analysis on nonlinear phenomena, Data analysis using methods of complex network
What is our brain doing when we see things or think about something ? The activity of the cerebrum is performed under the network through many neurons, and the electroencephalograms(EEG) and/or magnetoencephalograms (MEG) that are the outputs of activities include complicated nonlinearities. These complicated signals can be considered to reflect the state of as cognition, thinking, memory, and brain diseases. In recent years, it has become clear that in neuroscience, not only the firing rate of conventional neurons but also collective activities of neurons such as synchronous oscillation, complexity, and firing correlation are very important as information from the brain.
Meanwhile, in the field of information science, the algorithms that used to take a long time to analyze can now be calculated in a short time with the high performance of computers. In addition, the machine learning has made dramatic progress, and the feature extraction and classification of data can now be performed with high accuracy.
Here, for example, let us evaluate the severity of cognitive function by using EEG on dementia from Alzheimer‘s disease. So, the phase synchronization of EEG is calculated as it is regarded as a disorder of the brain network by considering Alzheimer’s disease from the neural basis. Figure 1 implies the classification for the degree of cognitive function progression from Alzheimer‘s disease by dimensionally compressing phase synchronization index on the α band of the EEG using the t-distributed stochastic neighborhood embedding (t-SNE) method, which is one of the machine learning methods. It can be seen from Fig. 1 that cognitive decline can be divided into mild and bad regions.
Furthermore, we find that the proposed method is powerful not only for diseases but also for quantitative evaluation of external features in the brain such as the correlation between the score value of the creativity test and the phase synchronization index of EEG.
Considering the cerebral activity in a hierarchy, each neuron is fired in the micro, and the set of firings in the network formation of neurons expresses electroencephalograms and/or magnetoencephalograms as brain waves in the macro. In the development of the brain, abnormal movement and placement of neurons, or miswiring of neural circuits due to network abnormalities are known to be a cause of brain abnormalities, because they inhibit various types of neurons mixing well in the cortex and interfering with balanced placement. So, we investigate the behavior of collective firing by using a model of neurons and connections between ones with a small world-like structure for complex network in cerebrum. The cerebrum is divided into 11 areas with physiological functional connections, and 200 neurons (excitatory neurons/inhibitory neurons = 4/1) are arranged in one area so as to form a small world. Figure 2 shows firing results of our model. By adjusting the strength of the stimulus and the values of several parameters required for the model, and by imposing a delay in the firing transmission between regions, we are able to reproduce the rhythm of slow collective firing around 0.1 Hz shown in spontaneous activity (default mode) and the α wave (8Hz-13Hz) that occurs in a relaxed state in humans.
S. Nobukawa, T. Yamanishi, et al., “Classification methods based on complexity and synchronization of electroencephalography signals in Alzheimer’s disease,” Frontiers in Psychiatry, 11 (2020) 255.
S. Nobukawa, H. Nishimura, and T. Yamanishi, “Temporal-specific complexity of spiking patterns in spontaneous activity induced by a dual complex network structure,” Scientific Reports, 9 (2019) 12749.
T. Yamanishi, J.-Q. Liu, H. Nishimura, and S. Nobukawa, “Low- frequency in the default mode brain network from spiking neuron model,” GSTF International Journal on Computing, 3 (2013) 8-16.