Systemy ekspertowe w diagnostyce medycznej

• Autorzy: Radomski D. , Jakubiak A.

Warianty tytułu

EN

Expert systems in medical diagnosti

• Języki publikacji: PL

Abstrakty

PL

Dynamiczny rozwój metod laboratoryjnych oraz technik obrazowania radiologicznego przyczynia się do znaczącego wzrostu danych klinicznych, stanowiących merytoryczne przesłanki przy podejmowaniu decyzji diagnoz. Fakt ten utrudnia szybką analizę danych zgromadzonych podczas badania chorego i podjęcie poprawnej decyzji klinicznej, wpływając na wydłużenie procesu diagnostycznego. Aby skrócić czas diagnozy stosowane są systemy ekspertowe wspomagające proces decyzyjny. Mają one głównie zastosowanie w masowych badaniach przesiewowych. Celem artykułu jest omówienie różnych klas systemów ekspertowych, uwzględniających losowy charakter diagnostycznego procesu decyzyjnego. W opisie zawarto również ocenę klinicznej przydatności danej klasy systemu.

EN

Dynamical development of laboratory methods and medical imaging techniques contributes to significant increase of clinical data which circumstances for diagnostic decisions. Therefore, the expert systems are used for computer aided diagnosis, particularly during screening tests. In our consideration, we assume the binary diagnostic decision distinguished between normal and pathological state. In this manner we neglect the differential problem between diseases with similar symptom. We introduce our classification of expert systems depending on their mathematical background. The first class contains systems constructed based on statistical theory of decisions. We show the equivalence between some elements of this theory and the rules of diagnostic decisions making. The first group of these systems is composed of the threshold detectors whose optimal threshold value is calculated based on received -- operating curve (ROC). We present the seldom used method for comparing two or more ROC, by introducing term the curves equivalence. This method is less restrictive than traditional requirement of curves equality but sufficient from diagnostic point of view. We show the example of statistical test, named Schuirmann test which verifies hypothesis about curves equivalence. The other described statistical systems are constructed by logistic regression model a special kind of regression dedicated for binary predictors. We show the method of parameter identification of such model and Wald test for statistical inference about identified parameters. Moreover, we propose and adaptive version of logistic regression where the parameters are updated by EM algorithm. The next class of expert systems is neural systems which are based on neural networks. We show that neural networks can be treated as a generalization of GLM models with least square method of parameters estimation. Therefore, neural networks could be used as nonlinear regression model mapped decision variable set into diagnostic decision. The background of the third class is the fuzzy set theory. We present the main idea of fuzzy sets, showing the standard fuzzy operators and present the concept of fuzzy diagnostic detector. The alternative way which describes uncertainty is the rough set theory. It assumes that a given element belongs to a given rough set if this element will be similar (in sense of special relation) to other elements of the rough set. We present the basic methodology of the rough sets application to diagnostic expert system. The last introduced class is the hybrid systems which combine different previous methodology. We distinguish two structures of hybrid systems, i.e. hierarchical structure and mixture structure. The hybrid expert systems are the most frequent used in clinical practice. The examples of such application are presented too. The last section of this article contains the summarized notes of each class. Our theoretical consideration show that particular classes distinguish the following features: the amount of medical knowledge needed to system building, the size of training sample needed to parameter estimation, the clinical uses. Hybrid systems appear to be the most useful in medical practice.

Słowa kluczowe

PL

system ekspertowy; statystyczne podejmowanie decyzji; krzywe ROC; diagnostyka medyczna

EN

expert system; statistical decision systems; ROC curve; medical diagnostic

• Wydawca: Wydawnictwo Naukowe PWN SA
• Czasopismo: Kwartalnik Elektroniki i Telekomunikacji
• Rocznik: 2003
• Tom: Vol. 49, z. 4
• Strony: 481--502
• Opis fizyczny: Bibliogr. 12 poz.

Twórcy

autor

Radomski D.

• Instytut Radioelektroniki, Politechnika Warszawska, ul. Nowowiejska 15/19, 00-665 Warszawa

autor

Jakubiak A.

• Instytut Telekomunikacji, Politechnika Warszawska, ul. Nowowiejska 15/19, 00-665 Warszawa

Bibliografia

• 1. K. Szamotulska: Epidemiologia Kliniczna. PZWL, 1997.
• 2. J. Halley, B. Mc Neil: The meaning and use of the are under ROC curve. Radiology, 1982, 143, pp. 29-36.
• 3. X. Zhou, N. Obuchowski, D. Mc Clish: Statistical Methods in Diagnostic. Medicince, Wiley, 2002.
• 4. D. Radomski, P. I. Roszkowski: Possible risk factors associated with endometriosis cysts. Fertility & Sterility, 2002, 77, p. 35.
• 5. J. Lachin: Biostatistical Methods: The Assessment of Relative Risks, Wiley, 2000.
• 6. M. Woodward: Epidemiology. Study design and data analysis. CRC, 2000.
• 7. H. Kosko: Fuzzy systems as universal aproximators. IEEE Tran. Camp., 1994, 43, pp. 12324-1333.
• 8. Y. Yao: A comparmive study of fuzzy sets and rough sets. J. Inform Sc., 1998, 109, pp. 227-242.
• 9. M. Panedo, M. Carreira, A. Mosquera, D. Cabello: Computer-aided diagnosis - a neural networks based approach to lung nodule detection. IEEE Trans. Med. Im., 1988, 16, pp. 874-881.
• 10. L. Hadjiiski, B. Sahiner, H. P. Chan, N. Petrick, M. Helvie: Classification of malignant and benign masses based 0n hybrid ART2LDA approach. IEEE Trans Med Im., 1999, 18, pp. 1178-87.
• 11. H. Cheng, R. Freimains: A novel Approach to Microcalcification detection using Fuzzy Lagic Technique. IEEE Trans. Med. Im., 1998, 17, pp. 442-450.
• 12. F. Chabat, D. Hasell, G. Yang: Computerized Decision Supporr in Medical Imaging. IEEE Eng. Med. Biol., 2000, 1O, pp. 89-100.