T.-S. Kim

Accelerometer signal-based human activity recognition using augmented autoregressive model coefficients and artificial neural nets

Automatic recognition of human activities is one of the important and challenging research areas in proactive and ubiquitous computing. In this work, we present some preliminary results of recognizing human activities using augmented features extracted from the activity signals measured using a single triaxial accelerometer sensor and artificial neural nets. The features include autoregressive (AR) modeling coefficients of activity signals, signal magnitude areas (SMA), and title angles (TA). We have recognized four human activities using AR coefficients (ARC) only, ARC with SMA, and ARC with SMA and TA. With the last augmented features, we have achieved the recognition rate above 99% for all four activities including lying, standing, walking, and running. With our proposed technique, real time recognition of some human activities is possible.

A standardized method for brain-cutting suitable for both stereology and MRI-brain co-registration

We have developed an agar-embedding method for brain-slicing that minimizes the geometrical distortions which arise from handling and slicing the fixed postmortem brain. To facilitate postmortem brain-magnetic resonance imaging (MRI) co-registration, each hemisphere is processed separately. We embed the fixed brain hemisphere with reference markers in agar. The block containing the brain and markers is sliced at a fixed interval using a rotary slicer. Each slice is photographed with a high-resolution digital camera. The digital images are realigned as a 3-dimensional volume via a control point-based registration method for multi-slice registration. The realigned multiple slices of the reconstructed postmortem hemisphere are then co-registered to corresponding slices of an in vivo reference MRI-volume. We illustrate these postmortem MRI-brain co-registration methods to correlate in vivo T2-weighted MRI hyperintensities in gray and white matter with underlying pathology. For design-based stereology, the volume of interest (VOI) is defined using reproducible anatomical boundaries. This method is suitable for stereologic measures of structures ranging from defined nuclei to whole brain.

Influence of White Matter Anisotropy on the Effects of Transcranial Direct Current Stimulation: A Finite Element Study

Although the use of transcranial direct current stimulation (tDCS), a noninvasive brain stimulation technique, has become popular in cognitive neuroscience research and clinical applications, there still exists limited knowledge on how to optimally use tDCS for the brain stimulation. To understand the underlying principles and effects of tDCS, finite element analysis (FEA) has been successfully applied due to its truly volumetric analysis and capability of incorporating the anisotropic tissue property. However, there are still many factors to be considered in stimulation: the influence of the white matter (WM) anisotropy is one of them which has not been considered in tDCS. In this study, we have examined the influence of the WM anisotropic conductivities on tDCS via high-resolution FE head models of the whole head. The effects of the WM anisotropy has been assessed by comparing the current density maps computed from the anisotropic FE model against those of the isotropic model using the similarity measures. The results show that there are significant differences caused by the tissue anisotropy, indicating that the anisotropic electrical conductivity is one of the critical factors to be considered during tDCS for accurate and effective stimulation of the brain.

Diffusion PDE-based denoising technique for magnetic resonance electrical impedance tomography

Recent progress in magnetic resonance electrical impedance tomography (MREIT) research has shown that conductivity images with higher spatial resolution and accuracy are achievable. One of the most important remaining problems to be solved in MREIT before we can apply the technique to human subjects is how to reduce the amount of injection current. Since we use an MRI scanner to measure the induced magnetic flux density data subject to an injection current, the data is contaminated with random noise. In order to obtain enough signal-to-noise ratio (SNR), we need to inject a large amount of current into the subject. However, it is obvious that we must comply with the electrical safety regulations and this means that we should deal with noisy data having a low SNR due to the limited amount of injection current. Furthermore, in the developed reconstruction algorithms, the required numerical differentiations of the noisy data may result in deterioration of the reconstructed conductivity image leading to a loss of important information. We propose a PDE-based denoising technique that diminishes the degradation of reconstructed conductivity images due to the noise in measured data. The proposed PDE-based technique is advantageous in reducing the random noise while preserving useful features in MREIT.

Content-Adaptive Finite Element Mesh Generation of 3-D Complex MR Volumes for Bioelectromagnetic Problems

In studying bioelectromagnetic problems, finite element method offers several advantages over other conventional methods such as boundary element method. It allows truly volumetric analysis and incorporation of material properties such as anisotropy. Mesh generation is the first requirement in the finite element analysis and there are many different approaches in mesh generation. However conventional approaches offered by commercial packages and various algorithms do not generate content-adaptive meshes, resulting in numerous elements in the smaller volume regions, thereby increasing computational load and demand. In this work, we present an improved content-adaptive mesh generation scheme that is efficient and fast along with options to change the contents of meshes. For demonstration, mesh models of the head from a volume MRI are presented in 2-D and 3-D

Microelectrode array analysis of hippocampal network dynamics following theta-burst stimulation via current source density reconstruction by Gaussian interpolation.

BACKGROUND Multielectrode arrays (MEAs) have been used to understand electrophysiological network dynamics by recording real-time activity in groups of cells. The extent to which the collection of such data enables hypothesis testing on the level of circuits and networks depends largely on the sophistication of the analyses applied. NEW METHOD We studied the systemic temporal variations of endogenous signaling within an organotypic hippocampal network following theta-burst stimulation (TBS) to the Schaffer collateral-commissural pathways. The recovered current source density (CSD) information from the raw grid of extracellular potentials by using a Gaussian interpolation method increases spatial resolution and avoids border artifacts by numerical differentials. RESULTS We compared total sink and source currents in DG, CA3, and CA1; calculated accumulated correlation coefficients to compare pre- with post-stimulation CSD dynamics in each region; and reconstructed functional connectivity maps for regional cross-correlations with respect to temporal CSD variations. The functional connectivity maps for potential correlations pre- and post-TBS were compared to investigate the neural network as a whole, revealing differences post-TBS. COMPARISON WITH EXISTING METHOD(S) Previous MEA work on plasticity in hippocampal evoked potentials has focused on synchronicity across the hippocampus within isolated subregions. Such analyses ignore the complex relationships among diverse components of the hippocampal circuitry, thus failing to capture network-level behaviors integral to understanding hippocampal function. CONCLUSIONS The proposed method of recovering current source density to examine whole-hippocampal function is sensitive to experimental manipulation and is worth further examination in the context of network-level analyses of neural signaling.

Mesh Quality Analysis of MRI Content-adaptive FE Head Models for Neuro-Electromagnetic Imaging

Realistic finite element (FE) head models for neuro-electromagnetic imaging are getting more attention due to their analytic advantages over conventional models. To improve the numerical efficiency, we have previously developed a novel mesh generation scheme that produces FE head models automatically that are content-adaptive to given MR images. MRI content-adaptive FE meshes (cMeshes) represent the electrically conducting domain more effectively with less number of nodes and elements, thus lessen the computational loads. In general, the cMesh generation is affected by the selection of feature maps derived from MRI. In this study, we have tested the effects of various feature maps on the generation of cMesh FE head models. Also we have evaluated the quality of cMesh FE head models to check their suitability for neuro-electromagnetic imaging using EEG and MEG. The results suggest that the cMesh FE head models with properly selected feature maps do show acceptable quality to be used in neuro-electromagnetic imaging.

Extraction and Localization of Alpha Activity of the Brain in EEG and fMRI Using Constrained ICA

In order to extract only the alpha activity related EEG signals of the brain, we have applied constrained independent component analysis (cICA), a new extension of ICA in which some a priori knowledge of alpha activity is utilized to extract desired components. Its performance has been compared to that of conventional band-pass filtering via the scalp power maps of alpha activity. The preliminary results show that cICA-extracted alpha signals produce the alpha power maps which spatially match the activated regions of the brain mapped using fMRI more closely than those from the band-pass filtered alpha EEG signals. The use of cICA might provide a more effective means of extracting EEG alpha activity of the brain.

Radix Polygalae Extract Attenuates PTSD-like Symptoms in a Mouse Model of Single Prolonged Stress and Conditioned Fear Possibly by Reversing BAG1

Radix Polygalae (RP) has been used to relieve psychological stress in traditional oriental medicine. Recently, cell protective, antiamnestic and antidepressant-like effects were disclosed but the possible application of RP to post-traumatic stress disorder, in which exaggerated fear memory persists, has not yet been explored. For this purpose, the effects of RP on fear behavior was examined in a mouse model of single prolonged stress and conditioned fear (SPS-CF), previously shown to mimic key symptoms of post-traumatic stress disorder. Male mice received daily oral dose of RP extract or vehicle during the SPS-CF procedure. Then fear-related memory (cohort 1, n=25), non-fear-related memory (cohort 2, n=38) and concentration-dependent effects of RP on fear memory (cohort 3, n=41) were measured in 3 separate cohort of animals. Also working memory and anxiety-like behaviors were measured in cohort 1. RP-treated SPS-CF mice exhibited attenuated contextual but not cued freezing and no impairments in the working memory and spatial reference memory performances relative to vehicle-treated SPS-CF controls. RP-treated SPS-CF and naive mice also demonstrated no difference in anxiety-like behavior levels relative to vehicle-treated SPS-CF and naive controls, respectively. In the hippocampus of SPS-CF mice, expression of BAG1, which regulates the activity of GR, was decreased, whereas RP increased expression of BAG1 in naïve and SPS-CF mice. These results suggest that RP exerts some symptomatic relief in a mouse with exaggerated fear response. RP and its molecular components may thus constitute valuable research targets in the development of novel therapeutics for stress-related psychological disorders.

Magnetic Resonance Electrical Impedance Tomography of the Breast: A Simulation Study on Basic Imaging Setup

Magnetic resonance electrical impedance tomography (MREIT) has been developed as a new medical imaging modality providing high-resolution conductivity and current density images. This paper is about MREIT of the breast. To show the feasibility of breast MREIT, we carried numerical simulations and breast phantom experiments. We found that an anomaly with 4 mm diameter can be visualized in a reconstructed conductivity image using 5 mA injection current if the SNR of the corresponding MR magnitude image is at least 150. We propose a desirable electrode configuration and show our first experimental results of the breast MREIT. Developing an RF coil for the breast MREIT, we plan to conduct various experimental studies including tissue phantoms