Utako Yamamoto

Extracting Rules for Cell Segmentation in Corneal Endothelial Cell Images Using GP

In tissue engineering of the corneal endothelium, extracting feature values of cultured cells from cell images helps us to automatically judge whether they are transplantable. To extract feature values, accurate image processing for cell segmentation is needed. We previously proposed a method that constructs a tree-structural image-processing filter by automatically combining known image-processing filters. In this paper, we propose a more accurate method that can be applied to images in which statistics differ in different regions. The proposed method prepares two types of nodes. One type of node represents known image-processing filters, and the other represents conditional branches, which determine the divergent direction using the statistics of the cell images. Moreover, the proposed method optimizes their combination by using genetic programming (GP). The proposed method is compared with the existing method using GP and specialist software for analyzing cell images. The results show that the proposed method has superior accuracy.

Gender classification of subjects from cerebral blood flow changes using Deep Learning

In this study, using Deep Learning, the gender of subjects is classified the cerebral blood flow changes that are measured by fNIRS. It is reported that cerebral blood flow changes are triggered by brain activities. Thus, if this classification has a high searching accuracy, gender classification should be related to brain activities. In the experiment, fNIRS data are derived from subjects who perform a memory task in white noise environment. From the results, it is confirmed that the learning classifier exhibits high accuracy. This fact suggests that there exists a relation between cerebral blood flow changes and biological information.

Endoscope image analysis method for evaluating the extent of early gastric cancer

In this study, a system is proposed to help physicians perform processing on images taken with a magnifying endoscopy with narrow band imaging. In our proposed system, the transition from lesion to normal zone is quantitatively analyzed and presented by texture analysis. Eleven feature values are calculated, i.e., six from a co-occurrence matrix and five from a run length matrix with a scanning window. Integrating these feature values formulates an effective and representative feature value, which is used to draw a color map, so the transition from lesion to normal zone can be visibly illustrated. In this paper, the proposed method is applied to images, and the efficacy is considered. This method is also applied to some rotated images to examine whether it could work effectively on such images.

A Preliminary Study of Interaction Effects on Brain Activity during Cooperative Work Using fNIRS

The primary goal of our research is to investigate brain functions during human-human cooperative work. For this preliminary study, a human-machine system to investigate cooperative work was developed. Blood flow changes in the brain were examined using functional near-infrared spectroscopy (fNIRS) and a region of interest (ROI) was determined. The relationships between the difficulty in cooperation and brain activity were assessed. A tapping task between a person and a machine was performed as a cooperative task. The machine provided a sound stimulus and the person tapped in correspondence to this stimulus. The stimulus interval was fixed, but stimuli were presented with disturbances. We assumed that degree of this disturbance was equal to the difficulty of cooperation. Our results demonstrated that cerebral activation was observed near the inferior frontal gyrus when the stimulus disturbance increased. Thus, the inferior frontal gyrus was the cerebral region associated with cooperation.

Investigation of regions of interest (ROI) through the selection of optimized channels in fNIRS data

We have proposed a method for extracting optimal regions of interest (ROI) through the selection of optimized channels, using machine learning classifiers and genetic algorithms in relation to functional near-infrared spectroscopy (fNIRS) data. Classifiers in machine learning have been used for determining labels belonging to test data. By using classifiers in the proposed method when determining object functions through optimization of existing discriminant functions, identifying the brain function area related to a particular subject is possible. In feature extraction, dynamic time warping (DTW) is used to extract any similarity in fNIRS data, and brain function areas are identified for a certain subject through classification by the support vector machine and feature extraction using the genetic algorithm. We confirmed the extraction of the areas related to working memory and results related to the brain function network by applying the proposed method to a time series of cerebral blood flow during a reading span test.

A feature transformation method using multiobjective Genetic Programming for two-class classification

In this paper, we investigate a method of performing feature transformation on input data in a 1-dimensional space in order to increase the accuracy of classifiers. Through optimized feature transformation, it is possible to create data which generate the models with high accuracy than the original data. We use Genetic Programming (GP) to find a feature transformation function. We proposed evaluation functions using GP and have been successful in finding transformation functions with a high degree of accuracy. On the other hand, where there is a deviation in the number of data items belonging to multiple classes, classes with a large number of data items are more accurate than those that do not. In order to resolve this, referring to existing research, we examined a method of handling the problem of improving accuracy and correcting class imbalanced accuracy from the generated models based on multi-purpose optimization. We then investigated the method of multi-purpose optimization and how to determine the threshold for classification. The results of the investigation were that we could obtain a transformation function that was more accurate and could consider the accuracy of multiple classes simultaneously.

Working Memory Training Strategies and Their Influence on Changes in Brain Activity and White Matter

In this study, we investigated whether different working memory training tasks influence brain activity and white matter changes. Thirteen participants were involved in our interventional study over a period of one month. During pre- and post-training, brain activity and structural integrity were measured using functional magnetic resonance imaging and diffusion tensor imaging. The reading span task was used to measure working memory capacity in participants performing different strategies. Participants were classified into a training group (10 participants) and a control group (4 participants). The training group was further divided into the imagery strategy group and rehearsal strategy group. Only the imagery strategy group improved working memory capacity, showing significantly increased activation in the anterior cingulate cortex and fractional anisotropy adjacent to the right temporal gyrus. Consequently, adopting the appropriate strategy is important for improving working memory capacity as different strategies affect brain activity and white matter to different degrees.

A feature transformation method using genetic programming for two-class classification

In this paper, a feature transformation method for two-class classification using genetic programming (GP) is proposed. GP derives a transformation formula to improve the classification accuracy of Support Vector Machine, SVM. In this paper, we propose a weight function to evaluate converted feature space and the proposed function is used to evaluate the function of GP. In the proposed function, the ideal two-class distribution of items is assumed and the distance between the actual and ideal distributions is calculated. The weight is imposed to these distances. To examine the effectiveness of the proposed function, a numerical experiment was performed. In the experiment, as the result, the classification accuracy of the proposed method showed the better result than that of the existing method.

Electroencephalographic method using fast Fourier transform overlap processing for recognition of right- or left-handed elbow flexion motor imagery

Recently, systems using motor imagery (MI) have been developed as practical examples of brain-computer interface (BCI). Electroencephalography (EEG) was used to generate an electroencephalogram of elbow flexion. In addition, a method was proposed to extract the feature values that would enable the recognition right- or left-handed elbow flexion MI. In the proposed method, fast Fourier transform overlap processing was used to determine the time period required to extract feature values. In this study, the following two experiments were performed. 1) the recognition of right- or left-handed elbow flexion by analyzing only the MI time period and 2) recognition of the right- or left-handed when the MI time period was presumed. In the first experiment, right- or left-handed elbow flexion MI was processed for 20 subjects using support vector machine and the proposed method was used to extract the feature values. In the second experiment, the presumed MI time was determined using the channels in which the highest accuracy was obtained in the first experiment, and then, right- or left-handed recognition was processed for the time period presumed. In the first experiment, the recognition accuracy of the proposed method was superior to that of the previous method in 15 of 20 the subjects. In the second experiment, the mean accuracy was 7.2%. Therefore, the recognition accuracy can be improved by improving the MI detection method.

Construction of an Interactive System Aims to Extract Expert Knowledge about the Condition Cultured Corneal Endothelial Cells

We aim to construct an expert system for diagnosing the health of corneal endothelial cells. To construct the proposed system, we first constructed a system that confirms whether experts use the same criteria to diagnose the condition of cells. In the constructed system, an expert interacts with a computer that generates images of cells by simulation. These images describe cells that are in the best condition, according to expert diagnosis. By comparing the results from multiple experts, we can elucidate whether experts use the same criterion for diagnosis. The proposed system is composed of an interactive genetic algorithm (IGA) and involves the simulation of cells. We confirmed the system operated normally through operational experiments. In another experiment, conducted with no experts, we confirmed that this system could generate images demonstrating a predetermined feature value.