Observation of daily human activity and status is important from the viewpoints of maintaining health and preventive medical care. In this study, we describe a system for monitoring human activities and conditions that uses microelectromechanical systems (MEMS) sensors. The system contains four MEMS sensors for environmental monitoring-3-axis acceleration, barometric pressure, temperature, and relative humidity -as well as the peripheral circuitry for each sensor. Measured human activity data are stored in a memory via an on-board microprocessor. We measured environmental data for a subject's daily life. To estimate the subject's activity and his condition from a huge volume of data, we applied a soft computing technique to machine learning for the automatic extraction of human-activity classification.
The nursing-care quality improvement is very important in the medical field. Currently, nursing-care freestyle texts are collected from many hospitals in Japan through the Internet by using Web applications. Some nursing-care experts evaluate the collected data for improving nursing-care quality, and then return a recommendation for improvement to each nurse. Since this Web based nursing-care quality evaluation system is a novel system in Japan, participation of many hospitals is desired. However, it is very difficult to increase the number of hospitals because the only few experts evaluate the collected nursing-care texts. In this paper, for evaluating many nurses, we develop a support vector machine (SVM) based classification system. And we propose a new attribute definition by using the expert's knowledge such as the length of a text and words or expressions which are focused by experts. First, for generating numerical training data, we extract attribute values from the nursing-care freestyle texts. And then, we classify the nursing-care numerical data using the SVM. Computer simulation results show the effectiveness of our proposed attribute definition. From the result of this paper, since many nurses can be evaluated by our proposed system, we can expect the quality improvement of the nursing-care in Japan.
The quantitative evaluation of heart motional function is important for diagnosis of heart disease. Doctors evaluate the motional function using indexes such as the changing rate of myocardial thickness and the changing rate of ventricular volume. Our research group has developed “Correlation Method Weighted with Confidence (CMWC)”. CMWC is a tracking method for regional heart muscle from ultrasonic radio frequency (RF) signals in order to measure the motion in the myocardium correctly. Using the tracking method, we were able to measure the motion quantitatively from RF signals with low noise. In this study, first, we statistically investigated the motion of normal heart muscle using successful results of CMWC. At the next, we proposed three indexes for evaluation of heart motional function. The first index was based on the sift velocity of normal heart muscle. The second index was based on the changing rate of myocardial thickness. The third index was the linear combination between the first index and the second index. At the last, we compared the accuracy of diagnosis between proposed indexes and changing rate of myocardial thickness by ROC analysis. As a result, the area under the ROC curve was 0.95 at the first index, 0.91 at the second index, 0.97 at the third index, and 0.93 at the changing rate of myocardial thickness. We were able to propose an index with higher accuracy than the changing rate of myocardial thickness.
Measurement of cortical thickness using human brain magnetic resonance (MR) imaging can assist physicians in quantifying cerebral atrophy. Most of the conventional measurement methods assign the same class to all pixels with a similar MR signal independent of their locations, and are therefore unsuitable for MR images that have strong intensity nonuniformity (INU) artifact. We propose an automated method that locally segments the cerebral cortex using an adapted fuzzy spatial model representing the transit of MR signals from the cerebral cortex to the white matter. This method assigns fuzzy degrees belonging to brain tissues using the adaptive fuzzy spatial model for local intensity transition from the cerebral cortex to inside the cerebrum. We also introduce an evaluation method of cortex segmentation algorithms that consists of reproducibility, quantitative, and qualitative tests; we use this method to evaluate and discuss the proposed segmentation method in comparison with the conventional method.
In this paper, we propose a computer-aided orthognathic surgery planning method which aggregates satisfying and unsatisfying examples obtained from past cases. In orthognathic surgeries, from a viewpoint of facial aesthetic improvement, unsatisfying cases sometimes occur unfortunately. In order to accomplish appropriate surgery planning, information from not only satisfying cases but also unsatisfying cases should be utilized effectively. To realize this, the modified version of the Self-Organizing Relationship Network (SORN) is employed. The network extracts a desirable input/output relationship from both satisfying and unsatisfying examples. Furthermore, some post-processing (e.g. clustering method by image processing) are implemented for effective visualization of the learning results. The proposed method is combined to profilograms for constructing a surgery planning system, and the system is qualitatively evaluated.
ACL reconstruction is performed to recover the injured knee which often happens in sports activity such as football, and basketball. The ACL reconstruction makes two bone tunnels in the femur and tibia. Then, the harvested graft passes the bone tunnels so that the femur and tibia are connected. However, the bone tunnel is enlarged with time. The bone tunnel enlargement can cause instability of the knee. To examine relationship of the bone tunnel enlargement and clinical results, the bone tunnel should be evaluated quantitatively and three-dimensionally. In MDCT image, intensity distribution around the bone tunnel is unclear so that it is difficult to extract the volume of the bone tunnel with error by simple thresholding method or region growing method. Then, manual extraction needs a lot of time and energy for examiners, because the data size of MDCT image is huge. Therefore, this study proposes an automated extraction method of the bone tunnel from MDCT image by using active contour model. In the experiment, the proposed method and the manual extraction method were applied for six patients who were performed ACL reconstruction. As a result, the proposed method could extract the bone tunnel with the error of 3.5 % in comparing to the manual extraction method.
In this paper, a scheme of recognizing hematopoietic tumor patients is presented, using self-organizing maps constructed by fast block-matching-based learning. This fast learning is referred to as T-BMSOM leaning. To classify the patients, screening data of examinees are presented to a constructed map. In T-BMSOM learning, a set of neurons arranged in square is regarded as a block, and one of the blocks is chosen as a winner per the presented data. It is assumed that members of a training data set to construct the map never change in static environments, whereas the data set is suddenly updated during learning in dynamic environments. While adopting the concept of blocks makes it possible to construct well-organized maps in dynamic environments, it lengthens the time for learning. To overcome this issue, T-BMSOM learning is based on a decision-tree-like winner search and a batch process. The screening data of an examinee frequently lacks several of the item values, and hence the data is presented to the map after averages of non-missing item values substitute for items with no values. The class of the data to be classified is basically judged by observing the label of a winner block. Simulation results establish that the proposed scheme achieves high accuracy of correctly recognizing the data of hematopoietic tumor patients, even if training the map is conducted in a dynamic environment.
Diagnostic imaging system is a necessity for brain diagnosis. Transcranial Ultrasonography can noninvasively image the intracranial blood flow and brain tissue in real time from only temple area of human head. However, the ultrasonic wave causes attenuation, decentration, and refraction in the skull, so the ultrasonography can not provide the transcranial brain surface image from arbitrary place. In this paper, we propose an imaging system of brain surface and skull from arbitrary places by considering the ultrasonic refraction of the skull. We do an experiment by using a cow scapula to imitate the skull bone and a biological phantom to imitate the cerebral sulcus. We first visualize the shape of scapula, and grasp the shape of scapula surface. We second remove the delay and the multi echoes of refracted wave. We third calculate the thickness of the scapula by using fuzzy inference. In the inference, we employ amplitude, correlation coefficient and the elapsed time. Finally, we calculate the refractive angle of ultrasonic wave and visualize the image referring to the refraction of ultrasonic wave. In the result of applying our method, we can estimate the thickness of scapula at all points, and successfully visualize the phantom surface image.
The method of bioelectrical impedance-based visceral fat estimation is in advance of other methods such as CT or MRI viewed from points of cost and safety. However, the method requires complex signal analysis and modeling to realize its high estimation accuracy factoring in individuality. In response to the requirement, the assurance of the feature attributes' variation and the selection of them to develop the estimation model have been proposed in this paper. The assurance of the feature attributes' variation is realized by employing the idea of cardinality as a quantitative evaluation index. The selection of the feature attributes is realized by employing the Akaike information criterion. The experiments were conducted to evaluate the proposed method. The results prove that the estimation model developed by the proposed method provide high estimation accuracy and stability.
We estimated the cardiac risk in non-cardiac surgery. We developed predictors for perioperative cardiac accidents, including hard events (cardiac death and myocardial infarction), using a total of 22 clinical properties such as surgical risk, patient's clinical information, and results from nuclear scanning. A total of 1351 surgery records including intermediate and low risk surgery, the risk prediction of which is often difficult were used and we analyzed them using linear and support vector machine (SVM) classification models. Our linear and SVM models achieved superior prediction results to conventional ones; 80% in sensitivity (SE) and 66% in specificity (SP) for all cardiac accidentsand 85% in SE and 81% in SP for hard events with a leave-one-out cross-validation strategy. Several parameters measured by nuclear scanning were commonly selected and we can therefore conclude that the nuclear scanning is effective for estimating perioperative cardiac risk in advance.
In this paper, we propose a new diagnosis method of pulmonary nodules in CT images to reduce false positive(FP) rate under high true positive (TP) rate conditions. An essential core of the method is to extract two novel and effective features from the raw CT images: One is orientation features of nodules in a region of interest (ROI) extracted by a gabor filter, while the other is variation of CT values of the ROI in the direction along body axis. Simulation results show that the discrimination rate of the proposed method is extremely improved compared with that by the conventional method.
In fractal analyses of EEG (Electroencepharography) and MEG (Magnetoencepharography) signals, the correlation dimension based on the correlation integral has usually been applied so far. In this research, we use the graph dimension obtained by the time-series fractal analysis by Higuchi's algorithm, and evaluate the MEG recordings for mentally stable 8 subjects and unstable 8 ones chosen among 23 subjects based on the result of psychological test. The fractal dimensions to the total 64 channels were determined every subject based on the relation between the coarsened time scale and the corresponding coarsened time-series length for the MEG data. In this result we found a multiple overlap structure of fractal dimensions (from the minimum dimension Dmin to the maximum one Dmax), so that we introduced its measure, ΔD= Dmax-Dmin. MEG recording during 3 sessions (10 minutes each) was executed for each subject. In session 1 a subject had remembered his/her daily events for 10 minutes. Continuously in sessions 2 and 3 the subject had remembered the contents of video after watching a cheerful video footage and a horrible and frightening video footage (2 minutes each), respectively. Mutual subtraction among three ΔD values by these 3 sessions enables to eliminate the noise under measurements and the personal equations. Owing to this method we could catch the variance of fractal structure in MEG data induced by different emotional stimuli, and revealed the difference between the mentally stable and unstable groups through the statistical examination. Not only the above discrete (scalar) application for each channel, the application of channel vector which consists of plural channels was examined. As the selection way of plural channels, two cases were considered, one is intra-area selection and the other is inter-area selection from the areas in cerebral cortex. As a result we could improve the discrimination ability of our method for the mentally stable and unstable groups by integrating the features of selected channels. Furthermore, we divided the MEG data in four time stages and evaluated the time dependency of fractal structure. With this analysis, there seemed that video's influence on the stable group subjects keeps fading away in time, but that on the unstable group subjects is long remaining.
We are studying travel time measurement method by vehicle identification using height data of vehicle provided from ultrasonic vehicle detector. And we proposed not one-to-one vehicle matching method but each vehicle sequence (two ormore vehicles) matching method, because many similar vehicles exist on the road. In addition, we developed this method by Genetic algorithm operations corresponding behavior of vehicle between upstream and downstream. Moreover, we describe improvement of evaluation method of vehicle sequence by introducing Levenshtein distance to express changing, inflow and outflow of vehicle. According to the experiment of proposed method, good precision was provided.
Kansei words are adjectives that mean psychological feeling or image of a product. Usage of kansei words depends on individuality. In this paper, by assuming that there is a hierarchy of kansei words, we propose a model of the mechanism that we assign kansei words given physical sensation. Then we can extract the users' preferences as numerical values. In addition, we present a product color coordinate system that reflects user's preference. Finally, we attempted to extract users' preferences and inspect the usefulness of the system.