"Fuzzy" is defined as "blurred" or "not clear" (Webster's New Twentieth Century Dictionary, 1957). In Japanese, "Fuzzy" is "曖昧" (a-i-ma-i). These Chinese characters originally mean that a substance can not be seen at dawn or in shadows. However, when once the sun shines, it can be clearly noticed. It has a definate core even though it is vague or ambiguous. Natural languages (categorical information) have their definate meanings ("langue" by Saussure), but the meaning must be variously changed in their contexts ("parole" by Saussure). Measured informations as clinical loboratory data are expressed by definate numbers, but they have also different meanings depending on their backgrounds (contexts). When we process the categorical informations in biomedical systems, we usually transform them into the numerical data. The words could not be used directly. We should not use blurred or false data but the correct and most suitable data. Fuzzy inference works most efficiently to discriminate the members and to make a decision. Association of the fuzzy inference in the neural network could afford the most human-like way to process the biomedical data.

View full abstract

In the framework of inflammatory protein variations, it is illustrated a methodology based on fuzzy logic for diagnosis assistance. The pattern of medical knowledge involves five proteins and eleven groups: eight inflammatory syndromes, two non inflammatory syndromes and the normal condition. Most of the inflammatory syndromes are not typical for protein variations can be dissociated, so that a global analysis of the problem has to be considered. Choosing as an example, Vasculitis (three protein levels are increased, while two others are decreased or normal), it is shown how the biomedical knowledge can be expressed linguistically to be then translated into possibility distributions. In the case of poor diagnostic classifications, it is introduced appropriate ponderations, acting on the characterizations of proteins, in order to decrease their relative influence. As a consequence, when pattern matching is achieved, the final ranking of inflammatory syndromes assigned to a given patient might change to better fit the actual classification. In some cases, influence of proteins in terms of importance operations is not symmetrical. To face such situations, a calibration of soft-AND operators is performed. These aggregation procedures are illustrated with an example. Defuzzification of results (i.e. diagnostic groups assigned to patients) is presented as a non fuzzy sets partition issued from a "separating power", and not as the center of gravity method commonly employed in fuzzy control.

View full abstract

The authors have developed a diagnosis system with a Fuzzy inference on a personal computer. In this system, we use the membership functions for evaluating data and can get the excellent detecting ability in borderline cases. One of the characteristrics of the new system is a multi-layer structure. As we can examine one case from many point of view, we can often find combined disease. Moreover, We can easily use the information which denies the probability of some disease, and become able to rule out a lot of disease more exactly.

View full abstract

We have developed an Expert System (ES) which assists the Doctors in the Automated Multiphasic Health Testing and Service (AMHTS), specially in Diabetes Mellitus. In the AMHTS the Doctors classify the patient conditions to 6 special categories (A, B, BF, C, D, G) difined in AMHTS. Almost of all ES can infer the Diagnosis only by using Rule Base with crispy threshold. These are not so sufficient for Doctors, because Doctors have ambiguous sense. We have developed ES, applied to DM, utilizing two types of methods indicating continuous certainty. One is the Index method similar to FUZZY integration theory. The other is the ES with Neural Network (NN). In the Index method, we have newly introduced a Diabetes Mellitus Index (DMI) which indicates the degree of DM. DMI is calculated by converting the 3 blood sugar (FBS, 1hGTT, 2hGTT) using non-linear scale, and multiplying the weightings corresponding to the each blood sugar value. Subsequently, inference is carried out using a rule base and DMI. Consequently, the categories from the ES upon 1000 patients data in Toshiba Rinkan Hospital have demonstrated a high degree of coincidence with the categories by the Doctors employed in the AMHTS. NN is constructed with 3 layers as the input layer, the hidden layer and the output layer. The input layer has 3 units assigned to 3 blood suger. The hidden layer has 50 units. The output layer has 5 units assigned to 5 categories except "A" category. "A" category is infered by the rule base. Consequently, the coincident ratio between the categories from the ES with NN and ones by the Doctors upon 3684 patients including above 1000 data has been sufficient. It is, however, less than the coincident ratio from the ES with DMI. Each of the 4 methods (FUZZY, Index, NN, Rule) is suitable and effective to the Medical Application, and so it is important to apply and combine the 4 methods to the appropriate problems.

View full abstract

Cell differentiation of Pleodorica californica was analyzed and a model was made depending on some assumptions. The most important assumption in our model is the two types of assumed fate of differentiation. One type is a fixed fate cell which differentiate into somatic or reproductive cell depending on the spatial arrangement in the cell sheat. The other type is a "fuzzy cell" of which fate of differentiation is not fixed. High coincidence between the measured and theoretical distribution of the number of reproductive cell in a cology indicates that the existence of "fuzy cell" is important in cell differentiation of Pleodorina californica.

View full abstract

This paper describes a design method of a fuzzy control system for artificial hearts. Artificial heart pump outputs and dose frequencies are controlled by the system for mainaining the circulatory system physiologic. To obtain control laws of these variables, the fuzzy inference is based on results of real time computation with circulatory models as well as on empirical knowledge about artificial heart control.

View full abstract

Medical images such as X-ray images, ultrasonic images and CT-images provide the phisician with a great deal of valuable information that is useful in the making of a diagnosis. It is often unclear, however, which facets or items of these images are most valuable, how much weight should be given to each item, and how diagnostic logic can be used with these items. In this paper, fuzziness in imaging data and in the diagnostic process are studied, and then diagnostic logic based on imaging data using fuzzy inference is discussed. Fuzzy chracteristics are first encountered in imaging data in the process of scaling the degree of the imaging features. The criteria for evaluation of each item differ from doctor to doctor. Judgement becomes subjective. Fuzziness also exists in the diagnostic logic for imaging because its framework cannot be clearly defined, since it is changed by the images chosen and these choices are somewhat subjectively made. Moreover, the weights of these items do not always remain constant.

View full abstract