Modern decision making support systems for physi cians of various specialties exert a substantial effect on quality of medical service. These systems provide consul tation in prognostic, diagnostic, prophylactic, and thera peutic activities.Many commercial systems of this class are presently available for specific medical purposes. Such commer cially available systems are based on expert databases that are tunable to user problems [1 5].In this work, design of network based fuzzy knowl edge bases (NBFKB) for medical decision making sup port systems is discussed. Such bases capable of tuning to user problems were developed at the Department of Biomedical Engineering, Kursk State Technological University.Selection of network systems for medical NBFKB is based on the following factors. Diagnostic and therapeu tic processes contain several stages and human health hypotheses. At each stage of diagnostic and therapeutic processes the hypothesis is supported, rejected, or modi fied. Parallel hypotheses can be put forward in case of combined pathology. The tactics of monitoring methods can also be changed. This change can be fairly sharp. If a stage of decision making procedure is represented as a node in the model of a subject area, branched transitions to other nodes are possible. Many researchers use net work models of a subject area to implement such mecha nisms [1,3,6,7].Decision making models are based on various math ematical methods and information knowledge databases. The NBFKB quality depends on model adequacy.Many medical problems (particularly problems of early diagnosis) use different structures of fuzzy knowledge bases, whereas class structure is not clearly determined.Fuzzy logic with decision rules tuned to a training procedure has been used as the main mathematical appa ratus [5,8].This apparatus does not exclude other approaches (discriminant analysis, group argument method, refer ence methods, etc.). The reference methods solve the problem of scale adaptation to decision making.With due regard to variety of decision rules in net work model nodes, a unified multifunctional decision module is suggested.The problems solved using one unified multifunc tional decision module are associated with a technologi cal component of the general solution. This component can include preliminary diagnosis based on epicrisis and medical examination; retrieval of additional information from patient database; stage of diagnosis elucidation with regard to standard examination results; stage of diagnosis elucidation with regard to results of instrumental exami nation, etc.A diagram of the decision module (DM) of a net work information logical model is shown in Fig. 1. The decision module contains program input/output inter faces with given specifications. The input/output inter faces are for input/output of facts, data, decision rules, addresses, control information, and training information.Interface I 1 provides input/output of human health state parameters to the decision module. Interface I 1 also provides input/output ...
Medical expert systems are automated artificial intelligence systems that make prognostic and diagnostic conclusions and decisions on optimal therapy and pro phylaxis on the basis of available data (interview data; results of visual, laboratory, and instrumental examina tions). At its output, the system produces text messages that can be understood by the user.The efficiency of medical expert systems depends to a considerable extent on the suitability of its mathematical tools for the specific prognostic or diagnostic task [1,4].Multiple studies in image recognition have revealed that the class structures under consideration (in syner getic approach, attractors, and frames) can overlap con siderably in the informative feature space, especially in the case of prognosis and early (prenosological) diagnosis [4,5]. State classes (attractors) can vary their position in the feature space. Various measuring scales (ordinal, name, interval) can be used for feature measurement. The features can be of various natures and belong to different data sets (interview data; results of visual, laboratory, and instrumental examinations). Some of the required informative features might be unavailable. Both the initial features and the system made decisions can be fuzzy.Data available from the literature and the results of our studies show that two decision theory approaches can be effectively used to solve the problem: 1) use of fuzzy decision logic and network structures; 2) study of class structure and hypothesizing on the optimal classifiers in a computational experiment (dialog systems for image recognition). Combined use of these approaches provides additional capabilities for classifier development and allows the efficiency of medical expert systems to be increased. In this work, the two approaches are combined to solve the problem of generation of network based fuzzy decision structures in three stages.In the first stage, exploratory analysis of the geomet ric structure of the classes under study in the informative feature space is performed (structure is defined here as the relative position of objects belonging to different classes in the learning sample).In the second stage, the partial membership function supports and parameters are selected. These membership functions are used for classification over subspaces and regions of the initial feature space. The variables and parameters are selected so that each partial membership function provides maximum confidence of classification or prediction at each step of the decision making process.In the third stage, the partial membership functions are combined into network structured fuzzy decision rule sets. These rule sets should provide the required perform ance quality.A special application software package was developed in our laboratory to study the class structure using exploratory analysis methods. The software performs the following functions: identification of characteristic points of the learning sample (multidimensional centers of class es and separated objects, groups ...
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.