Self-Organizing Maps (SOM) method which was developed by Kohonen [1, 3] was introduced. In the basic form, SOM is nonlinear projections of high dimensional data space, usually on a two-dimensional display space. These possibilities of the applications of SOM to medical purposes are discussed.

View full abstract

A technique is presented for modeling 3-D shape of rigid as well as non-rigid objects employing multiple video images. Several feature points are initially specified on the object interested. The image coordinates of the feature points on the video images are collected into a single matrix called an extended measurement matrix, to which singular value decomposition is applied to obtain the 3-D locations of the feature points. These recovered feature points provide us a 3-D model of the object. Advantages of the presented technique over others include that neither camera calibration nor registration of recovered feature points is required and that it can be applied to any object only if feature points can be defined on it. Experimental results are shown on some objects including human motions.

View full abstract

The SAT (SATisfiability problem) is a problem to find a solution, namely, a 0,1 assignment to each variable which satisfies the given conjunctive normal form(CNF). The SAT is a fundamental and important NP complete problem in the field of information science. We proposed a neural network called LPPH for solving the SAT. The LPPH is known to be able to solve the SAT efficiently compared with conventional algorithms. The MAX SAT is a problem to find a solution which maximizes the summation of the importance of satisfied clauses, where the importance is given for each clause. In this paper, in order to solve MAX SAT, we propose coefficients which reflect the importance of clauses into the dynamics of the LPPH, and study the behavier of the dynamics experimentally.

View full abstract

The Satisfiability (SAT) problem is one of the most important problems in the computer science and it is well known as famous NP complete problem. In the past years, some efficient techniques for solving the SAT have been developed, and we have proposed a technique called LPPH. For the SAT, only solutions which satisfy the given CNF are required. However, in practice, there exists some objective function which represents the cost of solutions. In this paper, we study an optimization problem OSAT defined as follows: for a give CNF, find a truth assignment that minimize a given linear objective function. For solving the OSAT we propose a technique which is an extended version of the LPPH.

View full abstract

The satisfiability problem (SAT) is one of important problems in the field of the information science. However, no general deterministic polynomial time algorithms are known yet. The Hopfield type neural networks for solving difficult combinatorial optimization problems have used the gradient descent algorithms to solve constrained optimization problems via penalty functions. However, it is well known that the convergence to local minima is inevitable in these approaches. We proposed dynamical differential equations called LPPH for the SAT. The LPPH is based on the Lagrangian method and it is proved that every equilibrium point of the dynamics is a solution of the SAT and vice versa. Hence it is never trapped by any point which is not the solution of the SAT. From experiments it is known that the LPPH outperforms already proposed combinatorial algorithms even if it is executed by numerical simulations on conventional computers. In this paper we propose a method which combines heuristics to the LPPH dynamics.

View full abstract

Satisfiability problem (SAT) of the prepositional calculus is a problem to find a variable assignment which satisfies the given Boolean expression in conjunctive normal form (CNF). The SAT is one of the most fundamental and important problems in the field of information science. We proposed a neural network called LPPH (Lagrange Programming neural network with Polarized High-order connections) for solving the SAT. We also proposed several improvement for the LPPH , such as LPPH with attenuation coefficient and double weighted LPPH. In this paper we propose a method to add new clause to the given CNF to speedup the LPPH.

View full abstract

Hypothetical reasoning is one of important reasoning frameworks because of its usefulness for some practical problems such as diagnosis,design,etc. Given a set of observations, hypothetical reasoning is an abductive reasoning mechanism to find consistent hypotheses. We consider cost-based hypothetical reasoning. Cost-based hypothetical reasoning associates a cost with each hypothesis, then the best explanation is defined as an explanation which has the least sum of the cost corresponding to each hypothesis. We have proposed the Lagrange programming neural network with polarized high-order connections (LPPH) as an effective technique for satisfiability problem (SAT). In this paper, at first we describe a method to accomplish deductive reasoning approximately using LPPH with bias. Then, we discuss cost-based hypothetical reasoning with restrictive background knowledge. One of the restrictions used is that background knowledge is represented by prepositional horn clauses

View full abstract

This paper deals with a construction of an expert system. In Our Expert System,the diagnosis rules are generated on the basis of fuzzy grouping method doctor' knowledge. We use fuzzy theory as measure of which expresses the ambiguity which is included in normal diagnosis. As an example, we consider the prognosis of patients with gastrointestinal cancer.

View full abstract

This paper proposes an interactive instructional method of human-body circulatory physiology including artificial heart control based on intelligent tutoring architecture. We use Coleman's dynamic model Human, which consists of 25 physiological modules and covers standard circulatory physiological knowledge. This tutoring system deals with causal relation among variables in the Human model, and discusses with student about the reason for increase or decrease of variables in accordance with dialogue strategy and student model.

View full abstract

In this paper, we describe a development of equational edit function and graphical simulator for living-body circulatory system model HUMAN. This edit support tool enables modification of model equations at on-line interface, so when user modifies and rebuilds the model, he/she can access the model on tool without programming knowledge. After the reconstruction user can smoothly make an experiment of diagnosis and control for the modified model. Moreover, we develop the graphical simulator that is effective when user traces dynamic response of the model.

View full abstract

Fuzzy inference has been applied for various types of clinical diagnosis, such as computer-aided diagnosis. Most of clinicians, however, tend to hesitate to adopt it, because some difficulties are inevitable to apply fuzzy inference actually. Although there are some certain ways of thinking in clinical diagnosis, clinicians cannot transform them to fuzzy production rule in most cases. For general clinicians, it is very complicated to generate and prepare the membership functions. Consequently, computer-aided diagnosis using fuzzy inference is hardly to be applicable in clinical field. Therefore, the authors would like to propose the adaptive neuro-fuzzy inference system (ANFIS) for the construction of computer-aided diagnosis system. Using ANFIS, the membership functions can be adjusted adequately in neural-network principle from retrospective clinical data. Then, clinicians do not have to generate and verify the membership functions critically. In this paper, we report the attempt of ANFIS to ultrasonographic diagnosis and indicate its validity for diagnostic aid.

View full abstract

When we evaluate an infomation system by fuzzy inference, the number of fuzzy rules becomes a lot in general. In this paper, we apply the concept of rough sets to diabetes data for the sake of the solution on this problem and indicate that we can get good results by fuzzy rules obtained by using rough sets.

View full abstract

Because the effectiveness of rehabilitation by horseback riding has already been shown in the medical science, the horseback riding exercise is used wide in Europe. However, the cost of the amount of height is needed because of securing the place and breeding of the horse, etc. Therefore, the purpose of this research is to manufacture the horse riding robot and to undergo clinical testing actually. In this paper, horseback riding robot system is introduced, and the effectiveness of this system is shown by the clinical trials .

View full abstract

Present association result is not independent, when the human carries out the association, and there is the relation in some forms in association result in the last time. In short, the association is dynamically carried out. hi the previous paper, authors proposed rubber string type associative memory algorithm using the physical model, however, it did not have the dynamics like the human. Therefore, in this paper, the dynamics is given to the conventional associative memory algorithm.

View full abstract

Similar cases retrieval from a database is useful, when a patient with an unfamiliar disease is treated. A database management system, therefore, for clinical cases on the WWW with self-organizing maps has been developed. Keywords are selected from case documents by natural language processing. As the same keywords are usually selected from similar cases, this database is especially available to retrieve similar cases. Similar cases, however, are not always found in case of the same few keywords. To overcome this problem, a new method has been proposed with which to modify input vectors by quantifying relationship between keywords using correlation coefficient. Computer simulations illustrated usefulness of the method in spite of the same few keywords. The future direction of this study will be an application to real clinical cases.

View full abstract

In this article, we propose a document clustering method with Natural Language Processing and Self-Organizing Maps (SOM) on our Web based DBMS for clinical cases. The system is implemented in Java language on the Internet server, and all documents of clinical case in the database are written in Japanese. We tested the system ability to evaluate for correct clustering of clinical cases, that each document has 30 attributes in an input-vector for SOM-learning. For the result of this ability test, it shows that 73.3% of clustering datasets are correct answers. Thus, it seems that our method is available to search the similar cases in the database.

View full abstract

In this paper, we suggest a distributed database management system for clinical cases with Java Servlet. In this system, cases were put two or more servers, and case index informations were put all servers. Communication between each servers were used Servlet-Servlet Connection. With this method, system can handle various formats of data and system were more extensive and more reliable.

View full abstract

It should be important to represent a fuzzy graph as an approximate graph and clarify its relational structure. We propose a methodology to solve this problem using a personal computer, which we have developed. The system could analyze the problem easily, speedly and exactly. But the system had several points to be desired. So, we have partially some performances of it, and inplemented on a personal computer. In this article, we describe the functions, performances and a case study.

View full abstract

A project is widely employed in the modern organization of management. The project organization is appropriate to a speed and agile management in uncertain and most changeable economic circumstance today. We propose the method to select members of a new project and distribute budget to each of members based on their past achievement and competencies. The method is named project portfolio management because it takes its risk into consideration.

View full abstract

In our study, we have analyzed the trend of the isolation of anaerobic bacteria from 1995 to 1998 in clinical laboratory. To statistically analyze the relationship between the patients' clinical backgrounds and the anaerobes isolated, we used DATA MINING TECHNIQUES. The smaller usage of drugs resulted in the decrease of anaerobes detection. In conclusion, the Lactbacillus Theraphy balanced the normal bacteria flora, and then, the anaerobes were not detected from the blood culture and other samples.

View full abstract

Pneumoconiosis analysis in chest radiographs using neural network is presented. The rounded opacities in the pneumoconiosis X-ray photo are picked up by a back propagation (BP) - neural network with several typical training patterns. The training patterns from 0.6 mm0 to 30 mm0 are made by simple circles. The neck problem for an automatic pneumoconiosis diagnosis has been to reject the unnecessary part like ribs and vessel's shades. In this paper such an unnecessary part is rejected well by adding several output neurons for own presenting neural network. These neurons are used only for picking unnecessary parts up. The input for the neural network is 30 x 30 pixel image which is quarried succeedingly from the bi-level ROI (region of interest) image with the size 500 x 500 pixel. The new technique called moving normalization is developed here in order to made an appropriate bi-level ROI image. The total evaluations is done from the size and figure categorization, many simulation examples show that the proposed method gives much reliable results than traditional ones.

View full abstract

Recently,healing pet robots are developed in application of AI, especially natural language processing. The psychotherapist ELIZA is a famous in early AI research but has not Japanese voice communication as human interface. We developed Japanese ELIZA system without translators,Japanese to English Englist to Japanese and with IBM Via Voice 98,speach recognition and speach synthesis system. Robot chat Eliza system by Peri on the network is discussed.

View full abstract

The purpose of this study was to predict lymph node metastasis in tongue cancer. The subjects were 137 patients with squamous cell carcinoma of the tongue who had been treated at our department during the 20-year period from 1971 through 1990. We first used multivariate analysis and then fuzzy inference to assess the accuracy of these methods for predicting lymph node metastasis. Based on the results of the chi-square test and a correlation matrix, T category, N category, keratinization, mitosis and mode of invasion were selected as parameters for multivariate analysis. For the prediction using fuzzy inference, the patients were first divided into two groups : those treated surgically and those given radiotherapy. Three items, i.e., tumor size, keratinization and mode of invasion were used for analysis. Then, an "If-Then rule" was applied to each of the two groups. It was found that fuzzy inference has a higher accuracy for predicting lymph node metastasis than multivariate analysis. Fuzzy inference appears to be useful for predicting lymph node metastasis.

View full abstract

We produced a tongue diagnosis supporting system by applying the syndrome determination process on a discriminating system based on the fuzzy theory. In this study, the results of the tongue diagnosis of coldness-heat were compared with those of the diagnosis using this supporting system. To evaluate the diagnostic ability of our tongue diagnosis supporting system, the results of tongue diagnosis was compared with those of the diagnosis using this system. The diagnosis using the supporting system was consistent with the tongue diagnosis in 70%, showing results close to clinical diagnosis.

View full abstract

To investigate an interaction between lymphocytes and adrenal medullary cells, we examined the effect of bovine lymphocytes or their conditioned medium on catecholamine efflux from bovine adrenal medullary cells. 1) Co-culture of adrenal medullary cells with lymphocytes caused an increase in catecholamine efflux from the cells. 2) Treatment of adrenal medullary cells with the conditioned media prepared from lymphocytes also increased the efflux of catecholamines. 3) Heat treatment of the conditioned media by boiling for 10 min reduced the stimulatory effect to 12 % of control. The stimulatory effect of the factor still remained even after dialysis using a < 8kDa MW cutoff tube. 4) The lymphocyte-derived factor was partially purified from conditioned media using a DEAE-cellulose ion exchange column. Three peaks of the active fraction were collected and pooled. Two antisera against the peak 1 and peak 2 were prepared and named as fuzzy antibody 1 and fuzzy antibody 2. Using the fuzzy antibody 1, 20 positive clones were isolated from bovine spleen cDNA library and partially sequenced. The present findings suggest that lymphocytes secrete factors which stimulate catecholamine efflux from adrenal medullary cells. The lymphocyte-derived factor(s) which seems to be heat-sensitive and have molecular mass of more than 8kDa will be cloned.

View full abstract

[in Japanese]

View full abstract

Epidemiological study has been performed on the relation between three cancers and five genes under the influence of tobacco smoking. The three cancers are stomach cancer, colorectal cancer and urothelial cancer, while five genes are glutathione S-transferase Ml, Tl, PI and N-acetyltransferase 1 and 2. We clarify the genetic polymorphisms increase the susceptibility to the cancers, especially by smoking tobacco.

View full abstract

The purpose of this study was to investigate the correlation of subjective ratings and fuzzy ratings based on cephalometric soft tissue analysis. Materials were lateral cephalogram traces of 20 females. Rators were 21 OMF surgeons and 20 orthodontists. The result of multiple regression analysis showed the form of nasolabial and labiomental region more influenced the subjective evaluation than soft tissue facial convexity.

View full abstract

The main purpose of this paper is to introduce the basic concept of our Artificial Emotion model so as to make a computer which deal with human sensibility. The study of emotions in engineering field has become an active recently. To construct a standard emotional database has long been a major focus of our study. But standardisation of emotional database is contrary to emotional character. Strict definition of the numerical data that express emotions is very difficult. Emotional data is strongly depend on a group or method of the questionnaire, Consequently the output data of emotion processing system strongly depends on a group or style of the questionnaire. First and foremost, we should construct a standard database of this emotion in conformity with the basic knowledge of emotional psychology. We construct a formal emotion database from this point of view. We call such a formal emotion database "Artificial Emotion". And processing system should have a function that can treat Subjectivity or vagueness. In particular taking subjectivity into account Prof. Yaiiaru Proposed the theory for Subjective observation model. The theory is based on a methodology of affine mapping between Euclidean spaces. This model stands on philosophy that all cognition and understandings can be done only on the observation space. We discuss here the mapping function that forms our model, and show some emotional maps which were drawn by computer simulation, and seem like "Screen of Consciousness in our mind".

View full abstract

As computer is making remarkable progress, it is hoped to construct the human friendly system taking parson's individual character into account. Facial expression is the most important individual character. In this paper, for a questionnaire, giving a face pictured on CRT to the testee, he is required the answer on what expression the given face is. As the result of analysis, we could give possibility to construct a system to recognize the facial expression for a certain given face by the some words inferrenced.

View full abstract

There was a problem of the frequency band widely crossed, when it intended to be made to realize the large-scale neural network hardware using Folthret as the fundamental element. Then, in this paper large-scale Folthret network would be able to be realized in group modulation and the multiple super heterodyne technology which are using on communication field.

View full abstract

On designing a human interface, we should aware of that a interface without considering skill of the user is difficult to To overcome this problem, we have employed a human interface adapting to user's characteristic. State transition model is used for the human interface. User-level is defined as a measure of user's skill, and it is changed according to information obtained from user's operation. State transition is changed based on the user-level. Evaluation function is explicitly denned and is used to change the user-level. Simple example is demonstrated.

View full abstract

[in Japanese]

View full abstract

In the analysis of medical and economic data, it is important to precisely grasp the present state of their systems. These human systems such as economical system or human bodies are influenced by many uncertain factors. Therefore, the systems are modeled as a complex one. It is not sufficient to understand time-series data by means of statistical methods and most important to analyze and understand the data in terms of the concept of fuzziness. In this paper, we employ the fuzzy concept into modeling time-series data and formulate fuzzy AR regression analysis.

View full abstract

This paper discusses fuzzy reasoning method by Shu et al. in which the consequence B' for a fuzzy rule A⇒B and a given fuzzy input A' is inferred according to the spatial relationship of fuzzy sets A, A' and B. Especially, the multifold fuzzy reasoning is investigated, and the improvement method for the fuzzy reasoning method is proposed by introducing a scale factor as the parameters characterizing a fuzzy rule.

View full abstract

To understand the anatomy of the nervous system, the neuron (nerve cell soma, dendritic processes, andaxon), and synoptic terminals must be studied. In the present study a freeze-drying device was applied to t-butyl alcohol substituted nerve cells, fibers and synoptic terminals in the human cerebellar cortex. Two types of nerve cells were observed: a flat polygonal cell, and many small round granule cells. A flat polygonal cell was surrounded with many granule cells. This cell soma with many synoptic terminals was revealed and micrographed using a scanning electron microscope (SEM). The results of measurements of these fuzzy microfeatures using an image analyzing software were reported. The granule cells varied in shape, but with many fibers. These microfatures showed fuzziness in shape, and differences between the flat polygonal cell and granule cells.

View full abstract

These days, multimedia data has been closer to computer users caused by development of information processing technology and computer user explosion. Therefore many kind of multimedia database system are published. But conventional multimedia database system is established as an annex, so it cannot be called them as real multimedia database system. In Uchinami laboratory at Kyushu Institute of Technology, we have proposed and developed techniques for multimedia database system aggressively. This paper reports these proposals and experimental results of its application.

View full abstract

In this research, the system to which an informational complement which is performed in the dialog with man is made is proposed, and it aims at raising the interface capability of a computer. This time, the fundamental mathematical technique introduced into the system and its meaning are reported.

View full abstract