脳神経系,拍尖脈波,心臓にみられるカオスがどのような働きをしているかを情報論的観点から解説する.カオス概念に不慣れな人のために,初めにカオスの入門的解説を行う.最大の目的は,生体の中でカオスが,様々な計算論的レベルのプロセッサーとして働いていることを理解して頂くことである.

抄録全体を表示

本論文ではまずパーセプトロンネットワークの一種である分散型単一層パーセプトロンネットワーク(DSPN)の入力層において、各種のあいまいさをもったファジィデータとあいまいさをもたないクリスプデータをレベル集合表現を用いて統一的に表現し、学習・テストデータとして処理するためのファジィニューラルネットワークを示す。また、このファジィニューラルネットワークを不完全な教師信号あるいは主観的に与えられたあいまいな教師信号を用いて学習する方法および出力層の構成を示す。基本的な学習アルゴリズムはポケットアルゴリズムである。次に上記のファジィニューラルネットワークおよびあいまいな教師信号を用いた学習法の医療診断への応用として、肝胆道系疾患の診断を行うファジィニューラルエキスパートシステムを開発した。GOT、GPTなど9種類の血液生化学検査成績と性別だけから一般に識別が困難な4種類の肝胆道系疾患(アルコール性肝障害、肝癌、肝硬変、胆石)の診断を行った。データは大学病院に入院した患者の症例の中から専門医によって確定診断された536症例を用いてあいまいな教師信号を含む学習・テストデータを作成した。上記の内からランダムに選択した373症例を学習データとして用い、163症例をテスト(未学習)データとして用いた。本論文で新たに開発したシステムの診断精度はテストデータに対して77.3%、学習データに対して100%であった。一方、同一のデータに対して判別関数分析を行うと、診断精度はテストデータに対して63.2%、学習データに対して67.0%であった。後者の結果は既往の肝胆道系疾患に対する判別関数分析による診断結果と同様である。以上の結果から本論文で提案しているファジィニューラルネットワークとあいまいな教師信号を用いた学習法はあいまいさを含む各種のデータの識別・診断に非常に有効であることがわかる。

抄録全体を表示

In this paper the general statement of the problem of medical diagnosis as well as its main featers have been presented. Then some fuzzy sets methods for diagnosis in medical systems are conceptually explained and analyzed by small illustrative examples. They include the methods using fuzzy relational equations as well the methods of fuzzy rule based models and fuzzy reasoning procedures. A brief comparison between crisp (traditional) and fuzzy diagnosis is also made.

抄録全体を表示

It is well known that the Chinese medical knowledge is a very complex system which is not only related to human body, human illness but also human activity, the environment of human living, the weather, even the states of other stars (including sun and moon) in the universe. Since there are so many factors should be considered for a doctor, it is almost impossible to use only the knowledge from text books and general logical inference, and usually is strongly dependent on doctors' experience. The associative inference for a Chinese medical expert system is based on fuzzy logic, neural networks, associative memory, concept structure and self-learning approaches. It can learn the Chinese medical knowledge from the supervises of expert doctors and from its working experience as a Chinese medical doctor in hospital.

抄録全体を表示

Structured neural network system was developed for diagnosing hepatobillary diseases. Structure was given to the neural network by using the pre-existing medical knowledge before learning the neural network. Comparisons was made of diagnostic accuracy between the ordinary and structured neural networks. The error back propagation was used as the learning algorithm both neural networks. The diagnostic accuracy of the structured neural network was higher than that of the ordinary one. The structured neural network had some advantages over the ordinary network concerning rapidity and convergence of the learning. Structured neural network might be effective tools for acquiring the knowledge from the medical database.

抄録全体を表示

Medical information based on natural languages such as "terrible fatigue" takes an important role for medical diagnosis. Though, such information suggests some of diseases, we have difficulty in the measurement of degree, the reliability of information and so on. Becuase, the medical information by natural language rises from the patients pain or feeling. So, moreover, the evaluation of this information is made by the doctor's criteria. So, information on natural language is subjective. The purpose of this paper is to make clear the fuzziness of medical information based on natural language and to propose a medical diagnostic system based on natural language using fuzzy inference.

抄録全体を表示

Ultrasonography is indispensable in diagnosis of various malignant tumors. Large variances in distinguishing between benign and malignant tumors are due to the fact that description evaluations and diagnostic method/expertise vary widely among individual physicians. Variable description evaluation is mainly due to image diagnosis fuzziness. We describe an ultrasonographic-based (ultrasonic tomogram) cancer diagnosis system which takes into account image diagnosis fuzziness and applies fuzzy theory. The system is used to distinguish between benign and malignant thyroid nodules and to diagnose prostatic cancer. We also discuss application of fuzzy theory to medical image diagnosis.

抄録全体を表示

Diagnosis of bone tumors is often difficult for even Orthopaedic surgeon, because they've shown various X-ray's appearances and clinical findings according as their growth. Through an experience of writing a program of Diagnostic Expert System for Bone Tumors using a Fuzzy Method, We considered the utilization of Fuzzy Method for Orthopaedic Surgery. Expert system permitting to input a vague expression had many advantages over another system to reject such an expression in it had been able to input the symptom and X-ray's appearance of bone tumors by natural expression for clinician.

抄録全体を表示

We have developed a new computerized system that can aid in the detection and classification of coronary artery disease. This system employs fuzzy theory that can analyze the data of stress myocardial scintigrapy in cardiac neuclear medicine. This system requests 4 units of input data such as, stress electrocardiogram(ECG), existance of chest pain (CP), parcentageof lung uptake(LU) and double product(DP). And output data of this system (diagnosis) consists 3 units, normal, single vessel disease (SVD), multivessel disease (MVD). The results in 500 cases of coronary artery disease show that sensitivity of this system is 83% and specificity is 62%. Our study also demonstrates that the sensitivity and specificity in this system depends on the variety of the membership function. We can quantitate the division process of medical docter (Radiologists) that usually depends on the experiencre and intuition. The preliminary study suggests that usefulness for the computer-aided diagnosis of coronary artery disease in stress myocardial scintigrapy.

抄録全体を表示

The activity sensing rate-responsive pacer incorporates a sensor as a piezo-cristal or accelerometer, bounded to the pacemakers. Body vibration which (is produced by physical activity effects the sensor. Rate regulation system cannot detect physiological parameters, because the system is open loop. Therefore, rate responsibility is to depend upon sensor geometry and system algorithm. Rate regulation of activity sensing pacer is linearly regulated by signal strength of the sensor, because the signal strength is associated with strength of activity. However the relationship of workload and sinus rate isn't linear, especially on intermittent or decreasing load. This isone of activity sensing pacer limitation in physiological pacing. Neural network system has a superior capacity of pattern recognition. In this article, we propose a new rate regulation algorithm using neural network system which learned patterns of relationship between sinus rate and workload. The Kohonen's Feature Map was chosen as a network model and the extended Learning Vector Quantization (LVQ) was adopted as a learning algorithm. Relationship between sinus rate and workload was examined using multiple exercise circuit test.

抄録全体を表示

An automatic, on-line and real-time monitoring and estimation system (TOTOMES; Tohoku University and Toyohashi University of Technology, monitioring and estimation system) for the cardiovascular dynamics under left ventricular assist device (LVAD) pumping has been developed to unload the operator who has to pay constant attention to the LVAD operation. The TOTOMES has the function which detects the presence and origin of accidents happening in the cardiovascular system, the pneumatic drive unit as well as the automatic controller. Detection of the malfunction is realized by using fuzzy reasoning. The combination of several measurements and estimates included in the antecedent of the fuzzy rules enables us to distinguish the origin of the malfunction. In in vivo experiment using an adult goat, it was ascertained that the proposed system successfully worked when the inlet cannula resistance was suddenly changed by clamping and when the blood withdrawing was carried out.

抄録全体を表示

ECC is an extremely important technique used during operative cardiac surgery. Its automatization was first attempted during the latter half of the 1950s. However, at that time there were many factors which hindered the development of an auto-regulation system. Recently, remarkable developments in ECC techniques have been attained through improvements such as increased performance and decreased size of the apparatus. In addition, sufficient processing speeds and performance are possible at a relatively low cost because of the epoch-making progress in computers over the past decade. Under these circumstances, we previously attempted the development of a microcomputer-based auto-regulation system for ECC. However, with respect to domain specifications our static regulation method lacked smoothness. This time we have attempted to develop a preliminary experimental model of an auto-regulation system for ECC, using fuzzy theory to improve the insufficiencies of the static regulation method. In this paper we describe the control theory of the artificial heart-lung circuit, the operating environment of the regulatory apparatus, and the performance evaluation results of the system.

抄録全体を表示

In the present report we propose a method which gives us the results of analysis with sufficient accuracy about Hellman's dental age by means of inputted dental formula by Fuzzy logic. It was found that the deductive results of approximated knowledge was gained even if there are few informations and it possessed a high practical value compared with previous methods. Though this study is new in progress, we have arrived at certain conclusions and outline of findings.

抄録全体を表示

Although rating of perceived exertion (RPE) is used to estimate exercise intensity, it has become evident that if fact there is only slight relationship between RPE and heart rate(HR). This study examined variability of HR in response to each RPE in healthy male students. One hundreds and five subjects walked and ran on the tread mill by progressively increasing workloads. HR and RPE measurements were recorded every two minutes. The frequency distribution of HR in response to RPEs were shown in 10 bpm increments. The median of distribution frequancy of RPE at very hard stage was shown at the HR of 180 bpm and variability was less than other RPEs. However, the widest range of HR, 100-200 bpm, was that of RPE at somewhat hard stage, which covered all the HR during exercise. These data suggeste that the estimation of exercise intensity by use of RPE not reliable, especially in the case of the range of somewhat hard exercise.

抄録全体を表示

There are people acute to some material. This symptom is what we call allergy. Allergy is haunting in many people, and is severly bothersome. There are many sources to invoke allergy. The most prevailing way of allergy test is skin test. That is, a nurse drops test materials(more than 60 kinds) on the human back. And pricks thru the material. So the material permiates into the skin. In 15 minutes, a doctor checks the reaction area. During that 15 minutes, a patient fills out a questionaire. The questionaire consists of 147 items: 16 items about symptom, 121 items about patienthistory, and 10 items about skin test. A doctor diagnoses a patient based on this questionaire. But, there are many items on it, so it is difficult to give considerations onto the whole items. Besides, the reaction area has very ambiguous boundary, the doctor decides the area ad hoc. Surveying these environment, we decided to automatize the process of diagnosis on allergy. First, Automatize acquiring the reaction area. Take pictures using CCD camera and Processing the image using computer. Second, Expert systemize the surveying of the questionaire. Using neural networks, automatizing the process from questionaire to diagnosis on allergy.

抄録全体を表示

In most of the Fuzzy Controllers or Fuzzy Inference systems, the center of gravity method is widely used as a defuzzifier method. We can occasionally find that several conclusion patterns in Fuzzy Inference are obviously different each other, although the results from defuzzification show middle and almost equal. Some of their outlines are classified to uni-peak, bi-peak or flat pattern for example, and these three types suggest different meanings respectively. In this study, we discuss new approach in which Neural Networks classify several conclusion patterns into three groups. This method will probably present additional information in medical decision making.

抄録全体を表示

DSA(digital subtraction angiography) is a valuable method in ragiology. But, the diagnostic worth of DSA may easily be spoiled by misregistration artifact. For this reason, a procedure to reduce the artifact is needed. We propose a new method for the reduction. In the method, a motion vector detection technique to find slippage between two images in used. Motion vectors include the vector that is derived from subject and that from flow of contrast medium. To distiguish the two, we utilize a fuzzy inference. The usefulness of the misregistration artifact reduction method was confirmed by a simulation test.

抄録全体を表示

We propose a new Neural Network (NN) system named Fuzzy Filtered Synapse (FFS) by which information is filtered actively by fuzzy membership function before input to Synapse. With the FFS-Network complicated and/or vague problems can be resolved easily, since non-linear and non-monotonic fuzzy membership functions are learned directly and the network is organized automatically. The usefulness of FFS has been confirmed through some applications. In this paper, we introduce an application to a scheduling problem of menu. In the learning, an example of menu list made by a dietitian is given to the network, and filter functions and weights of each synapse are determined. From the result of reasoning that all generated menu lists with various restrictions satisfy adequate balance, it has been confirmed that the network can learn the hidden knowhow of menu ordering.

抄録全体を表示