畑 豊

This article describes a personal identification method using dynamic foot pressure change while walking. This system acquires foot pressure change by wearable pressure sensor. The sensor has four sensing tips for the each foot, and these tips are fixed on shoes. In the experiment, we acquire eight steps pressure change data. The system extracts gait features from every step. The features are based on peak pressure value of each sensing tip. For all steps and all registered subjects, we calculate Euclidean distance between the feature and template of subject. The template is made from learning data of each registered person. For each step, the system chooses a registered person with the shortest Euclidean distance as a candidate of walking person. Then, the system counts selected number of the steps of a person. Finally, we identify the walking person by the larger number. The number less than threshold, identification of this walking person is failure. We employed 10 volunteers and identify them. In the experiment, we took pressure data 10 times for each volunteer. We used 3 data for learning and used the other 7 data for test data. The proposed method obtained 0.83% in FRR and 0.02% in FAR, when threshold parameter less than 3. © 2012 IEEE.

This paper proposes a body weight prediction method using Fuzzy prediction model. Fuzzy prediction model is constructed by an autoregressive (AR) model based on body weight data and linear prediction models based on biological data. The biological data are obtained by pedometers such as number of steps, calorie consumption and so on. The Fuzzy prediction model is fixed by solving Yule-Walker equation and minimizing the Akaike's Information Criterion. In our experiment, the model predicts body weight change for next p days where p is the order of AR model. Then, four linear prediction models related to the biological data are constructed by linear regression analysis. We make a fuzzy membership function based on mean absolute error between body weight data and predicted value of each prediction model. Furthermore, these models are optimized for each subject in prediction models which add the biological data to AR model based on the mean absolute error. We employed 452 volunteers, and collected their body weight time-series data and the biological data during 730 days. We use these data from 1st to 365th day as learning data to determine the Fuzzy prediction model. As the result, the Fuzzy prediction model obtained higher correlation coefficient between predicted and truth values than the AR model on most subjects. In addition, the Fuzzy prediction model obtained smaller mean absolute prediction error than the AR model. © 2012 IEEE.

This paper describes human brain ultrasound-mediated diagnosis in emergency medicine and home health care. The ultrasonic equipment has many advantages, for example, the very simple operation to touch to the body surface diagnosis enables real-time visual recognition for heart beat and unborn baby moving, and so on. The ultrasonic diagnosis is safety to human body and many repetitions. The goal of our research is the portable and real time brain diagnosis under the thick-skull. In our experiment, we employ two ultrasonic array probes with the center frequency of 1.0MHz and 0.5MHz. The choice of ultrasonic frequency is a trade-off between spatial resolution of the image and imaging depth. We perform the experiment with a cow scapula as a skull and a steel sulcus as a lateral cerebral sulcus. As the results, the synthesized image in Wavelet transform has higher efficiency than the other synthesizing on the images for the bone and the sulcus. © 2012 IEEE.

Asthma causes the bronchus inflammation, and makes breathing impossible. In worst case, asthma causes death by dyspnea. If we can predict cause asthmatic attacks, they can prevent from asthmatic attacks. Therefore, asthmatic attacks prediction system is desired. As a prediction system using time series data, there is Fuzzy-AR model that can consider multi factors. In this paper, we propose a prediction method of the number of asthmatic attacks on next month based on Fuzzy-AR model. The proposed method considers weather factors, temperature, atmospheric pressure and humidity data. This method is applied to asthmatic attacks data from Himeji city Medical Association. As a comparison method, AR model is applied to same data. The experimental results shown that the proposed method predicts the number of asthmatic attacks better than AR model. © 2012 IEEE.

This paper describes a health care service system to care disaster survivors. The Tohoku area in Japan was hit by a big earthquake and tsunami. There were many fatalities, with survivors forced into temporary housing. First, health care service management system for them is introduced. In it, primary service consists of mental health, physical health care systems, and body and mental check systems for caring the disaster survivors in temporary housing. Second, two practical systems such as a human behavior detection using 3D infrared camera, and biosignal monitoring by air mat sensor are described. They employed for mental care and locomotive syndrome prediction. Finally, a conclusion and future system are discussed. © 2012 IEEE.

This paper describes an ultrasonic image synthesis in Fourier transform for visualizing a human brain image under a skull. We employ two ultrasonic array probes with the center frequency of 1.0MHz and 0.5MHz. We perform the experiment with a cow scapula as a skull and a steel sulcus as a lateral cerebral sulcus. We propose two methods to synthesize the ultrasonic data. The first method is a way which synthesized a 1.0MHz data and a 0.5MHz data in the condition of setting a cow scapula and a steel sulcus. The second method is a way, with using the Fourier transform which never changes each frequency component, which synthesized a 1.0MHz data in the condition of setting a cow scapula and a 0.5MHz data in the condition of setting a steel sulcus. As the results, the synthesizing in Fourier transform has higher efficiency than the normal synthesizing on the images for the bone and the sulcus. © 2012 TSI Press.

The neonatal cerebral disorders might deform the brain shape, and reduce the cerebral function. For the diagnosis of cerebral disorders, it is effective to measure cerebral volume and surface area using head magnetic resonance (MR) image. The measurement should require a brain segmentation process. However, there are few studies for neonatal brain. This study proposes a brain segmentation method for a neonatal brain. In this study, we propose a shape and appearance knowledge based brain segmentation (SABS) method. SABS method segments a brain and cerebrospinal fluid (CSF) region by using a brain atlas model. Next, it classifies a brain and CSF region into some classes by using Bayesian classification with Gaussian mixture model, and optimizes the brain surface by using fuzzy rule-based active surface model method. Experimental results in 14 neonatal subjects (revised age between 1 month and 1 month) showed that the proposed method segmented the brain region with higher accuracy than the conventional methods. © 2012 TSI Press.

The article proposes the new notion of Systems Health Care which is systems approach for more effective and efficient health care. Recent developments of sensor, information, and communication technology have been realizing powerful and evidence-based solution using low invasive and continuous measurements of life style habits and vital signs. The measured indices are variously valuable to be used by diagnosis criterion, prevention, and treatments of metabolic syndrome and life style disease. Besides the ideas, the authors point out the importance of cyclic and continuous development of valuable solutions according to health dynamics and individual dependency. The solution should be suitably provided by coevolutionary integration of smart devices and smart services. Smart device realizes smarter service. Through providing services, smarter devices will be newly developed based on the experience and evidence. In the article, some discussions on health issues and systems approach for health care are organized in the first part. Some case studies and discussions of smart devices and smart services follow them. © 2012 IEEE.

This paper describes current state of the art of medical image processing and surgery support system. In it, typical medical image processing processes such as segmentation, registration and enhancement are described as an application of fuzzy logic. Second, orthopedic surgery and surgery support systems are introduced. In it, we meet the fuzzy logic application to ultrasonic system. Third, a bed-side health monitoring system is described. Fuzzy signal processing is employed in this system. Finally, a future direction of medical and health technology are discussed. © 2012 TSI Press.

Marine radar systems have been equipped by almost sea vehicles. However, radar image quality can be easily deteriorated by signal strength decay with increasing distance, blurring due to side lobes, signals of behind objects and of shaded area. And it is difficult to recover the artifacts. This paper proposes a new approach to improve radar image quality by image fusion. The fusion method generating high quality radar images is based on EM algorithm. Performance was validated using radar images acquired from a single radar system at multiple locations. The experimental results showed that the proposed method constructed high quality radar images, and recovered signal decay and blurring due to side lobes. © 2012 TSI Press.

This paper describes a fuzzy system of stem implantation on total hip arthroplasty by an ultrasonic device. The system can perform automatic and accurate assessment in the surgery. In this system, we employ a single ultrasonic probe whose center frequency is 1,000 Hz. We detect the acoustic signals when knocking the inserted stem with a hammer. We then have a correlation between the degree of tightening and the attenuation time of acoustic signal. That is, the higher tightened degree implies shorter attenuation period. The support system selects the most suitable stem size by fuzzy inference with respect to the attenuation time and its difference time from correct stem to one larger size stem which dynamically adapts to each patient. As a result, we successfully determined the suitable stem in comparison to the results of the practical surgery. © 2012 Naomi Yagi et al.

This paper describes a trans-skull ultrasonic Doppler system for measuring the blood flow direction in brain under skull. In this system, we use an ultrasonic array probe with the center frequency of 1.0 MHz. The system determines the fuzzy degree of blood flow by Doppler Effect, thereby it locates blood vessel. This Doppler Effect is examined by the center of gravity shift of the frequency magnitudes. In in-vitro experiment, a cow bone was employed as the skull, and three silicon tubes were done as blood vessels, and bubble in water as blood. We received the ultrasonic waves through a protein, the skull and silicon tubes in order. In the system, fuzzy degrees are determined with respect to the Doppler shift, amplitude of the waves and attenuation of the tissues. The fuzzy degrees of bone and blood direction are calculated by them. The experimental results showed that the system successfully visualized the skull and flow direction, compared with the location and flow direction of the phantom. Thus, it detected the flow direction by Doppler Effect under skull, and automatically extracted the region of skull and blood vessel. © 2012 SPIE.

A system of ambulatory, halter, electrocardiography (ECG) monitoring system has already been commercially available for recording and transmitting heartbeats data by the Internet. However, it enjoys the confidence with a reservation and thus a limited market penetration, our system was targeting at aging global villagers having an increasingly biomedical wellness (BMW) homecare needs, not hospital related BMI (biomedical illness). It was designed within SWaP-C (Size, Weight, and Power, Cost) using 3 innovative modules: (i) Smart Electrode (lowpower mixed signal embedded with modern compressive sensing and nanotechnology to improve the electrodes' contact impedance); (ii) Learnable Database (in terms of adaptive wavelets transform QRST feature extraction, Sequential Query Relational database allowing home care monitoring retrievable Aided Target Recognition); (iii) Smartphone (touch screen interface, powerful computation capability, caretaker reporting with GPI, ID, and patient panic button for programmable emergence procedure). It can provide a supplementary home screening system for the post or the pre-diagnosis care at home with a build-in database searchable with the time, the place, and the degree of urgency happened, using in-situ screening. © 2012 SPIE.

Health is quite important to be realized in our daily life. However, its idea covers wide area and has individual dependency. Activities in health care have been widely developed by medical, drag, insurance, food, and other types of industries mainly centering diseases. In this article, systems approach named Systems Health Care is introduced and discussed to generate new and precious values based on measurements in daily life to change lifestyle habits for realizing each health. Firstly, issues related to health such as its definitions are introduced and discussed by centering health rather than disease. In response to the discussions on health, Home and Medical Care is continuously introduced to point out the important role causality between life style and vital signal such as exercise and blood pressure based on detailed sampling time. Systems approaches of Systems Health Care are discussed from various points of views. Real applications of devices and services are used to make the studies and discussions deeper on the subjects of the article. © 2012 SPIE.

This paper proposes a heart rate monitoring system for detecting autonomic nervous system by the heart rate variability using an air pressure sensor to diagnose mental disease. Moreover, we propose a human behavior monitoring system for detecting the human trajectory in home by an infrared camera. In day and night times, the human behavior monitoring system detects the human movement in home. The heart rate monitoring system detects the heart rate in bed in night time. The air pressure sensor consists of a rubber tube, cushion cover and pressure sensor, and it detects the heart rate by setting it to bed. It unconstraintly detects the RR-intervals; thereby the autonomic nervous system can be assessed. The autonomic nervous system analysis can examine the mental disease. While, the human behavior monitoring system obtains distance distribution image by an infrared camera. It classifies adult, child and the other object from distance distribution obtained by the camera, and records their trajectories. This behavior, i.e., trajectory in home, strongly corresponds to cognitive disorders. Thus, the total system can detect mental disease and cognitive disorders by un-contacted sensors to human body.

This paper describes a method of average difference imaging and applying it to ultrasonic nondestructive evaluation of wind turbine blade. Average difference image is made from the means of intensity difference among several images. The image demonstrates the difference between an interested image and the other images. In performance test of this method, we employ a specimen of wind turbine blade. This specimen has artificial damages. We acquire ultrasonic waveforms from scanning lines on specimen using an ultrasonic single probe. To extract the damage, we make average difference images for every line. In the images, we could successfully find the line image with damage portions, and we estimated depth of damage portions with high accuracy. Thus, average difference imaging effectively led the line image with the damages. We then found the depth of damage portions from the line image. © 2012 IEEE.

This paper proposes a classification system for human activity using a multi-sensor system with a built-in electrocardiograph and 3D accelerometers. The multi-sensor system unconstraintly measures biological information, and provides these data to personal computer by wireless communication. We classify human activity by the biological information. The sensor detects the electrocardiogram and triaxial acceleration data of subject. The subject has several activities such as "Walking", "Walking Stairs", "Rest" and "Strength training". The proposed system classifies these activities by a decision tree. Branch conditions of the decision tree are defined by fuzzy membership functions. These fuzzy membership functions are constructed by exercise intensity, distinction frequency and postures. We compared our proposed method with a method using only acceleration data to show the effectiveness of the multi-sensor system. As the results, the proposed method obtained high classification accuracy. © 2012 IEEE.

The importance of measuring vital signs and life style activities in ordinary life besides in medical field has been realized more and more. This is because the affect of life style disease in super aging society has been strongly associated with long term nursing care. Additionally, sensing and information technology has been developed for realizing ease of use, cost reduction, and low intrusion. In this article, systems approach to health care is developed by centering home and medical care. The essence of the notion is using sensory data of vital signs and life style activities in both medical field and home by bridging between them to make medical treatment and home self-care efficient and effective. Systems Health Care mainly composes of Health Management and Knowledge Harvesting technologies. First one is designed for continuous health care and improvement by applying index, criterion, and causality. The second is for causal knowledge extraction process from sensory database, which is used in Health Management. The notion and the technologies are studied and discussed in the article. Application studies follow them by centering health care supporting system and employing both vital signal and life style activities monitoring. Blood pressure analysis program used in medical field as vital signal monitoring is employed. Regarding life style activities, active mass monitoring, non-contact sleep monitoring, and weight-loss programs are introduced. © 2012 IEEE.

This paper proposes a human movement trajectory recording method by a thermopile array sensor. In the system, the sensor is attached to the ceiling and it acquires place-dependent temperatures, which is called thermal distribution. The system obtains 4 x 4 pixels thermal distributions from the sensor. The distributions are analyzed to extract human movement trajectory. First, human candidate pixels are detected by background removal with three fuzzy rules on temperature. Second, those pixels are separated by Connected Component Labeling. Then, each human area is identified. Third, human centroid is calculated. Human movement trajectory is defined as a path of those centroids and recorded at all times. In the experiment, subjects performed 15 motion patterns such as standing and walking for an adult male. The system measured each motion during 1∼3 minutes. As the experimental result, the system successfully recorded accurate human movement trajectories in the all. © 2012 IEEE.

This paper describes a biometric personal authentication method, using a pair of right and left sole pressure distribution changes, while walking. This system acquires sole pressure distribution changes via a mat type load distribution sensor, and does personal authentication. We employ twelve features based on the shape of a footprint, and twenty seven features based on weight movement for sole pressure data. Fuzzy if-then rules for each registered person are introduced, within which, their parameters are statistically determined in the learning process. We calculate the fuzzy degree of a pair of right and left sole pressure data for any registered person, and identify the walking person as the person with the highest fuzzy degree; the fuzzy degree being higher than a threshold. We employed 90 volunteers and authenticated them. We evaluate the proposed fuzzy method by five hold cross validation on which low false rejection and false acceptance rates are achieved. Thus, this fuzzy logic approach is precise for this biometric system. © 2011 Sharif University of Technology. Production and hosting by Elsevier B.V. All rights reserved.

Cerebral surface extraction from neonatalMR images is the basic work of quantifying the deformation of the cerebrum. Although there are many conventional methods of segmenting the cerebral region, only the rough area is given by counting the number of surface voxels in the segmented region. This article proposes a new method of extraction that is based on the particle method. The method introduces three kinds of particles that correspond to cerebrospinal fluid, gray matter, and white matter; it converts the brain MR images into the set of particles. The proposed method was applied to neonatal magnetic resonance images, and the experimental results showed that the cerebral contour was extracted with a root-mean-square-error of 0.51 mm compared with the ground truth contour given by a physician.

There are various cerebral diseases that deform the cerebral shape with region specificity. So it is effective to quantify the deformation change of cerebral gyri. This study introduces new index called gyral deformation index (GDI) that is defined as a ratio of area of gyrus of interest to area of cerebrum in the defined projection plane. To calculate the gyral areas, this paper proposes a gyral labeling method in the projection plane using magnetic resonance images. The new method finds the boundaries between the gyri by optimizing deformable boundary models aided by fuzzy logic. The proposed method was applied to quantify, the cerebral deformation of infants on a plane which is perpendicular to the longitudinal fissure. The comparison results with the manual delineation showed that the method labels gyri with a mean sensitivity of 92.8% and a mean false positive rate Of 0.1% for 14 infantile subjects (3 weeks - 4 years 3 months old).

In this paper, we propose an approach to extract features of center of foot pressure (COP) obtained by a load distribution sensor and apply this method to develop a biometrics personal identification system. Biometrics technology, as a method of personal identification, plays an important role in our daily lives. In our experiment, we have a user stand on load distribution sensor with slipper, and acquire pressure data during a simple motion, as touching a bell nearby by one hand but without movements of feet. We propose a biometrics personal identification system with less information, time and low space. First, we calculate the site of COP from the obtained pressure data. Features for identification are extracted from the position and the movement of COP. Second, we built a k-out-of-n system and a neural network (NN) model with the feature parameter. Third, we input test data to the two systems. Finally, we give a comparison of these two methods. We employ 11 volunteers. The experimental result reveals that the proposed identification method can achieve an accuracy of 12.0% in FRR (False Rejection Rate) and 1.0% in FAR (False Acceptance Rate).

This report describes an eddy current system for locating distal transverse screw holes of an intramedullary nail. In fracture surgery, screw holes on an intramedullary nail are in invisible situation. Although conventional X-ray methods can visualize screw holes in bone, they pose a danger due to X-ray exposure. Therefore, a system to fmd screw holes without X-ray exposure is required. We solved this problem using an eddy current system and applying a method that calculates local minimum positions of a searching signal obtained by the eddy current system. Shape characteristics of screw holes were revealed in the waveform and amplitude of the searching signal. As a result, the screw hole positions could be identified with the smallest error of 0.99 mm when the distance between the probe and the intramedullary nail was 8 mm. The screw was successfully inserted. to the screw holes in this accuracy. © 2007, TSI® Press Printed in the USA.

Finding intracranial aneurysms plays a key role in preventing serious cerebral diseases such as subarachnoid hemorrhage. For detection of aneurysms, magnetic resonance angiography (MRA) can provide detailed images of arteries non-invasively. However, because over 100 MRA images per subject are required to cover the entire cerebrum, image diagnosis using MRA is very time-consuming and labor-intensive. This article presents a computer-aided diagnosis (CAD) system for finding aneurysms with MRA images. The principal components are identification of aneurysm candidates (= ROIs; regions of interest) from MRA images and estimation of a fuzzy degree for each aneurysm candidate based on a case-based reasoning (CBR). The fuzzy degree indicates whether a candidate is true aneurysm. Our system presents users with a limited number of ROIs that have been sorted in order of fuzzy degree. Thus, this system can decrease the time and the labor required for detecting aneurysms. Experimental results using phantoms indicate that the system can detect all aneurysms at branches of arteries and all saccular aneurysms produced by dilation of a straight artery in I direction perpendicular to the principal axis. In a clinical evaluation, performance in finding aneurysms and estimating the fuzzy degree was examined by applying the system to 16 subjects with a total of 19 aneurysms. The experimental results indicate that this CAD system detected all aneurysms except a fusiform aneurysm, and gave high fuzzy degrees and high priorities for the detected aneurysms.

Finding intracranial aneurysms plays a key role in preventing serious cerebral diseases such as subarachnoid hemorrhage. For detection of aneurysms, magnetic resonance angiography (MRA) can provide detailed images of arteries non-invasively. However, because over 100 MRA images per subject are required to cover the entire cerebrum, image diagnosis using MRA is very time-consuming and labor-intensive. This article presents a computer-aided diagnosis (CAD) system for finding aneurysms with MRA images. The principal components are identification of aneurysm candidates (= ROIs; regions of interest) from MRA images and estimation of a fuzzy degree for each aneurysm candidate based on a case-based reasoning (CBR). The fuzzy degree indicates whether a candidate is true aneurysm. Our system presents users with a limited number of ROIs that have been sorted in order of fuzzy degree. Thus, this system can decrease the time and the labor required for detecting aneurysms. Experimental results using phantoms indicate that the system can detect all aneurysms at branches of arteries and all saccular aneurysms produced by dilation of a straight artery in 1 direction perpendicular to the principal axis. In a clinical evaluation, performance in finding aneurysms and estimating the fuzzy degree was examined by applying the system to 16 subjects with a total of 19 aneurysms. The experimental results indicate that this CAD system detected all aneurysms except a fusiform aneurysm, and gave high fuzzy degrees and high priorities for the detected aneurysms. Copyright © 2006 The Institute of Electronics, Information and Communication Engineers.

In this paper we propose a learning algorithm to enhance the fault tolerance of feedforward neural networks (NNs for short) by manipulating the gradient of sigmoid activation function of the neuron. We assume stuck-at-0 and stuck-at-1 faults of the connection link. For the output layer, we employ the function with the relatively gentle gradient to enhance its fault tolerance. For enhancing the fault tolerance of hidden layer, we steepen the gradient of function after convergence. The experimental results for a character recognition problem show that our NN is superior in fault tolerance, learning cycles and learning time to other NNs trained with the algorithms employing fault injection, forcible weight limit and the calculation of relevance of each weight to the output error. Besides the gradient manipulation incorporated in our algorithm never spoils the generalization ability.

This paper proposes an on-line testing procedure for fuzzy controllers. A stuck-at-0 fault and a stuck-at-1 fault in the membership function are assumed as the fault models. In the proposed method, up to 7 extra functions are considered for each input variable in the antecedent part. The antecedent part and the consequent part can be tested in parallel. In the testing of the antecedent part, the sum of function values (degrees) derived based on the matching between the current measured input and the functions is observed. In the testing of the consequent part, the sum of the area of the output fuzzy sets, the sum of the degrees of applicability, and the sum of 7 degrees in the output fuzzy sets are observed. These procedures make it possible to handle simultaneously a fault in the antecedent part and a fault in the consequent part. Through comparisons with other methods, the proposed method is shown to be useful in terms of the number of faults that can be handled and the ease of application to multi-input controllers.

This paper proposes an automated procedure for discriminating tissues using ultrasonic waves. Generally, the property of the echoes varies at each tissue. It is useful to discriminate the tissues according to difference of the property for diagnosis of diseases such as cancer, But, it is difficult to discriminate the tissue from the property because of its inter- and intraobserver variability. Therefore, the proposal method is aided by fuzzy inference and consists of two stages. The first stage constructs the expert system. In this stage, the fuzzy membership functions are automatically constructed at each characteristic value from pre-experimental data. The second stage predicts the tissues from another waves using the constructed expert system. Finally, the method was applied to six tissues. The experimental result indicated that all the tissues were discriminated (96.7%) with high accuracy than "C4.5" (93.3%).

In this paper, we propose an automated method for segmenting the cruciate ligament and the meniscus from CT knee images. The method first finds the candidate region of interests (ROIs) and the bone region by using intensity thresholding. The obtained bone region is decomposed into the femur and the tibia by watershed segmentation. To eliminate the cartilage and the cortical bone from the candidate region we can express these tissues by using the fuzzy if-then rules. To segment the ROIs we employed physician's knowledge; 'the cruciate ligament and the meniscus are wedged between the femur and the tibia', 'the shape of the meniscus is half-moon' and 'the cruciate ligament are located near the center of the knee'. The knowledge is converted to fuzzy if-then rules, and then the rules can compute the fuzzy degree for ROIs. To evaluate our method, it was applied to 5 normal subjects. Quantitative evaluation of the resultant images by a physician shows that our method can give interesting 2D reconstructed and 3D surface rendering images. These results would help us to understand 3D shape and to evaluate the condition of the cruciate ligament and the meniscus.

Clustering methods, such as k-means, Fuzzy C-Means (FCM), and others have been developed. However, they are only partitioning a database, so it is difficult to discover the reason why each cluster is formed. This paper proposes a method to discover the knowledge of how the clusters are derived. To select the center vector of each cluster, we employ an unsupervised clustering method based on Self-Organizing Map (SOM) without giving the number of clusters, We define the degree of contribution calculated from the weights of the neural network which learned the center vectors. We then describe the knowledge discovery method from the degrees. We applied our method to the artificial data and the clustering problem. The results show that the degree of contribution is an efficient indicator to represent the knowledge of how the clusters are formed.

This paper proposes a method for extracting the cholecyst and the bile duct from magnetic resonance cholangiography (MR-C) volumetric images. We propose weighted fuzzy c-means clustering to classify an MR-C image into some clusters in which voxels have similar intensity and similar position. Then, the method finds the clusters corresponding to the cholecyst and the bile duct by evaluating the center vectors. Our experimental result on six subjects showed that this method could extract both the cholesyst and the bile duct.

Image segmentation is one of the fundamental techniques to develop a computer-aided diagnosis (CAD) system in the medical field. This paper first introduces rough sets into image segmentation method. In this method, attribute values of each pixel of an image of interest are given by using K-means clustering, and the attribute values divide the image into many regions. By applying value reduct, which is one of the typical concepts of rough sets, to the attribute values, dissimilarities between regions are calculated. Final clustering result is obtained by merging similar regions. To evaluate the performance of the proposed image segmentation method, it was applied to an artificial generated image, and a human brain Magnetic Resonance (MR) image. The results were also compared with convention K-means clustering.

To enhance fault tolerance ability of the feedforward neural networks (NNs for short) implemented in hardware, we discuss the learning algorithm that converges without adding extra neurons and a large amount of extra learning time and cycles. Our algorithm modified from the standard backpropagation algorithm (SBPA for short) limits synaptic weights of neurons in range during learning phase. The upper and lower bounds of the weights are calculated according to the average and standard deviation of them. Then our algorithm reupdates any weight beyond the calculated range to the upper or lower bound. Since the above enables us to decrease the standard deviation of the weights, it is useful in enhancing fault tolerance. We apply NNs trained with other algorithms and our one to a character recognition problem. It is shown that our one is superior to other ones in reliability, extra learning time and/or extra learning cycles. Besides we clarify that our algorithm never degrades the generalization ability of NNs although it coerces the weights within the calculated range.

This paper proposes a method for extracting the human hippocampus based on multiscale structure matching scheme. Focusing on the feature that an overextraction occurs on anatomically specific place, the method detects the redundancy by comparing with given desired models. Since each of the desired models has information about locations of their redundant segments, the place of corresponding redundancy can be specified on the overextracted object. Then, subtle intensity difference around their connecting place is investigated to separate the hippocampus and redundancy. The matching process can proceed in parallel for various types of redundancy and individual variances. Qualitative evaluation of a physician shows that our method can detect the redundancies and extract hippocampus correctly.

This paper presents an automated method for segmenting CT images of the fractured foot. Segmentation boundary is determined by fuzzy inference with two types of knowledge acquired from orthopedic surgeons. Knowledge of joint is used to determine the boundary of adjacent normal bones. It gives higher degree to the articular cartilage according to local structure (parallelity) and intensity distribution around a joint part. Knowledge of fragment is used to find a contact place of fragments. It evaluates Eucledian distance map (EDM) of the contact place and gives higher degree to the narrow part. Each of the knowledge is represented by fuzzy if-then rules, which can provide degrees for segmentation boundary. By evaluating the degrees in region growing process, a whole foot bone is decomposed into each of anatomically meaningful bones and fragments. An experiment was done on CT images of the subjects who have depressed fractures on their calcanei. The method could effectively gives higher degrees on the essential boundary: suppressing generation of useless boundary caused by the internal cavities in the bone. Each of the normal bones and fragments were correctly segmented.