The authors have developed a computer-assisted diagnostic system for ultrasonograpny using fuzzy reasoning. In this system, we adopted the centroid method of defuzzification for fuzzy reasoning. There have been reported, however, several methods for defuzzidication. Although the computer-assisted diagnosis obtained from fuzzy reasoning might varies with different defuzzification methods, there were few literatures concerning this subject. Hence, the authors considered other defuzzidcation methods;centroid(Cent), bisector(Bis), middle of maximum(MOM), largest of maximum(LOM)and smallest of maximum(SOM)methods. Diagnostic accuracy, sensivity and specificity acquired from these defuzzifications were compared each other. Consequently, Cent, Bis and MOM are adequate procedures. On the other hand, diagnostic accuracy acquired by LOM or SOM is worse than other procedures. However, sensitivity of LOM and specificity of SOM are better than others. Therefore, these methods are useful when sensitivity or specificity is clinically important.

View full abstract

This paper presents on expert system for an acupuncture diagnosis using fuzzy based knowledge representation and fuzzy reasoning. Furthermore, this paper gives a method to learn rules by genetic algorithms, and a technique to optimize fuzzy membership functions.

View full abstract

This paper presents an application of fuzzy integral model for clinical diagnosis. This model is constructed with fuzzy measures to the attributes of objects and Choquet integral. The fuzzy measures are obtained by genetic algorithm. A clinical application model to diagnose depressed type early gastric cancer was constructed and the structure of this evaluation model was analyzed with fuzzy measures. As the results, the attibutes with high degree of this model using possibility measure as fuzzy measure corresponded with the attributes a subject estimated at high degree.

View full abstract

The relationship among fractal dimension, noise and spatial resolution was investigated. Fractal dimension images were generated by Boxcount method for simulation data sets(FBM:Fractional Brownian Motion)and clinical data sets(MRI and PET glucose metabolic rate and cerebral blood flow images of the brain surface). In the image with low fractal dimension, the fractal dimension was markedly overestimated by noise. However, by using appropriate smoothing to reduce noise, the texture of the images can be evaluated by fractal dimension. In clinical applications, the difference of fractal diension between patients and a control subject was evaluated.

View full abstract

Many clinical cases, which include their diagnoses and treatments, are discussed at least once a week at conferences in hospitals. However, many of them are not published in any journals but collected in papers in each hospital. If many medical doctors could refer to these cases, they would be useful in clinical medicine. It is important to study similar cases when a patient with an unfamiliar disease is treated. The purpose of this study is to develop a database system by which unpublished cases, as mentioned above, can be utilized. From clinical cases transmitted through the Internet to be system, keywords are selected automatically by natural language processing, which is called automatic indexing. These keywords are added to the case document to make a full text database. Also, an inference engine using fuzzy measure allows efficient retrieval from the database. Trials have shown that successful applications to real cases can be inferred by using the system with a full text including keywords. This confirms that a database can be constructed from clinical cases by the system.

View full abstract

An application of fractal theory was studied for evaluating quantitative expression of the optokinetic nystagmus pattern(OKP). Fourteen ocular movement patterns(OM)and eye speed patterns(ES)of the OKP were analyzed using the box-counting method of the fractal theory to evaluate any fractal dimensions(FD). Almost all values of correlation coefficient both OM and ES were over 0.99. The values of standard error of OM were smaller than that of ES(U test, p<0.01). The FD of both patterns were also different from each other. These findings indicate that the FD of OM can express OKP quantitatively rather than FD of ES.

View full abstract

The patients, old age persons and physical handicapped persons need both of wheelchair and helpers in the hospital. When a helper carries one these persons, he or she must be attend on a person ridden on a wheelchair. The helpers are busy and have many works. Also, other persons with no helper are very tired to navigate their wheelchairs. Therefore, we propose a fuzzy navigation system for the wheelchair. This system consists of a navigation planning part, a navigation control part and an environment recognition part. In this paper, we report about a planning method for wheelchair's moving route.

View full abstract

This paper proposes a distance-type fuzzy reasoning method based on the calculated distance between two fuzzy sets. At present, Mamdani's fuzzy reasoning, functional fuzzy reasoning and simplified fuzzy reasoning are in widespread use for fuzzy control and fuzzy expert systems. These fuzzy reasoning methods are generally called direct approach. Many problems have arisen in the direct approach because height of a product of fact and antecedent is adopted for the compatibility degree. In this paper, a new method based on the distance between two fuzzy sets is proposed to solve these problems. The proposed distance-type fuzzy reasoning method has the lowing advantages over the conventional method:(1)reasoning consequence also becomes convex fuzzy sets if fuzzy sets are convex for the then part of the conditional proposition;(2)modus ponens is strictly satisfied;and(3)precise consequence can also be reasoned in the case that the antecedent is set of singleton type.

View full abstract

In almost of conventional image enhancement methods, the intensity mapping curve based on the intensity histogram is generated, and the intensity of each pixel in an original image is transformed. In the process, the evaluation function is defined, and the intensity mapping curve is tuned in order to satisfy the function. However, user impression for an image is based on his subjectivity, it can not be represented by the objective evaluation function. In this paper, the Self-Organizing Relationship(SOR) network which can construct the input / output relationship based on the subjective evaluation, is proposed, and it is applied to obtain the relationship between the intensity histgram and intensity mapping curve base on the user subjectivity. The images transformed by the conventional methods and the proposed method are showed to seven users, almost users answered that the images transformed by the proposed method are suitable for their preference.

View full abstract

The present paper newly proposes multilayered neural networks with two kinds of intermediate elements, that is, feature detection element and categorizing element, to substitute the conventional neural networks. Furthermore, computer simulation is performed fot three kinds of three-layered networks. The simulation results show that leaning capability in the case of using intermediate elements is improved compared to the case of increasing the number of the second layer elements of the conventional three-layered networks and that categorizing element is more effective than feature detection element.

View full abstract

In this paper, multiplication-type learning threshold element and addition-type learning distance element are newly proposed in addition to fundamental and multiplication-type learning distance elements presented already. Furthermore, computer simulation was performed for three layered network with two inputs and one output. Simulation results show that the most excellent is multiplication-type learning distance element in learning stability, lerning accuracy and learning rate, and is fundamental learning distance element in generalization abolity.

View full abstract

In this paper, we propose a fuzzy diagnosis system that is a pregnant prognosis using an artifical intelligent system with a learning function, wichi is called "FAIS" short for. The FAIS has functions that acquire new fuzzy diagnosis knowledge from clinical experiences under an unknown situation and solve some problems by deducing a new state through fuzzy inference. The purpose of constructing this system "FAIS" is also to make a system that can solve many kinds of problems in various medical fields. As an example, the FAIS was applied to a pregnant prognosis for some cases after treating the infertile women and the pregnant prognosis rules were obtained by an unification(a heuristic clustering)on the basis of measurement values for a progesterone and an estradiol. Furthermore from results by simulating a diagnosis using these rules, the usefulness of the FAIS is discussed.

View full abstract