Medical Decisions Based on Fuzziness and Uncertainty
 |
.jpg)
A World of Imprecision
Precise and definite information about real world objects is difficult to obtain, and, in the realm of medicine, such information is not always accessible to doctors for diagnosis and treatment.
This is not so surprising as the relationship between symptom and disease is sometimes vague, and the state of a patient can be very hard to define. The explosion of medical knowledge further complicates the problem. Doctors are faced with a large amount of fuzzy and uncertain information, from which medical conclusions have to be derived and on which therapeutic actions have to be based. Who can they turn to for help?
Prof. K. S. Leung of the Department of Computer Science and Engineering has come up with a fuzzy expert shell called Z-III. With this development tool he and other researchers at the University have built several medical expert systems that can be applied to obstetrics and gynaecological cases. The project, entitled 'An Expert Computer System on Medical Consultation and Management', was financed by an earmarked grant of HK$310,000 from the Research Grants Council in 1989.
A Fuzzy Expert Shell -- Z-III
Figure A: The Structure of Z-III |
Z-III is a structured and modular rule-based shell (see figure A) that is menu-driven and easy to learn. It incorporates both fuzzy logic as propounded by L.A. Zadeh and the certainty factor model used in MYCIN -- a medical expert system. Fuzzy concepts such as 'very dry' or 'rather heavy' are represented by fuzzy sets in the form of 11 numbers, which denote different degrees on an imaginary conceptual continuum. Fuzzy numbers are used to store fuzzy certainty factors ranging between -1.0 and 1.0, with negative factors reflecting disconfirmation. Built in a PC environment, Z-III can handle both exact and inexact reasoning.
Example:
Rule: If height is tall and sex is male, then weight is heavy. Certainty factor is 0.7.
(Both 'tall' and 'heavy' are fuzzy concepts.)
Data Input: Height is 'very tall'. Sex is 'male'. Certainty factor is 0.9.
Inference: Weight is 'very heavy'. Certainty factor is 0.63.
Three Medical Expert Systems
From Z-III, three medical expert systems called ABVAB, INDUCE36, and ESROM have been developed.
ABVAB can diagnose the cause of abnormal vaginal bleeding from the past history and physical examination results of a patient. While medical knowledge is represented as expert rules, imprecise medical concepts are represented by fuzzy types such as 'freshness of blood'. Certainty factors indicate the degree of confirmation for the individual diagnoses. The expert system gives more than one possible diagnosis, in order of preference, and the first preference has the highest certainty factor.
A consultation with ABVAB involves making numerous inferences using the expert rules stored in the knowledge base and the information stored in the medical data base, which can be accessed by the users during a consultation. A sample diagnostic conclusion of the possible causes of abnormal bleeding in a patient, in preference order, is shown in Table 1.
| Table 1 |
Diagnoses Made by ABVAB
It is extremely certain (0.97) that the diagnosis should be dysfunctional uterine bleeding.
It is very certain (0.95) that the diagnosis should be pelvic infection.
It is very certain (0.94) that the diagnosis should be malignant genital tumour.
It is indeed certain (very close to 0.92) that the diagnosis should be benign genital tumour.
It is pretty certain (0.8) that the diagnosis should be complications of pregnancy.
It is quite certain (0.72) that the diagnosis should be hormone induced.
It is almost certain (0.6) that the diagnosis should be a foreign body in the vagina.
It is somewhat certain (0.58) that the diagnosis should be endometriosis.
It is somewhat certain (0.53) that the diagnosis should be genital injury.
It is little certain (0.32) that the diagnosis should be an intrauterine contraceptive device.
|
Forty-four real patient cases were used to verify the accuracy of ABVAB, and the results were satisfactory. (Table 2)
| Table 2 |
Correct Diagnoses Made by ABVAB
As the first choice
As the first three choices
Listed | Percentage
68
90
100 |
INDUCE36 concerns the decision to induce labour in pregnant patients after 36 weeks of gestation. This artificial initiation of the delivery process is sometimes required when the continuation of the pregnancy could prove more risky, either for the baby or the mother, than delivering the baby. But, the delivery of an immature foetus means significant risk for the baby, and induction under unfavourable circumstances can also result in a caesarean section which is a relatively high-risk procedure for the mother. Induction of labour thus involves careful appraisal of the duration of the pregnancy and its risk factors, and all pointers of maternal or foetal compromise. INDUCE36 can be an important consultative system here.
The system was evaluated by the researchers using 30 hypothetical cases of varying difficulty. Its performance was compared with that of six doctors of different ranks (two interns, two registrars, and two senior registrars). Three consultants examined each of the seven sets of recommendations, i.e. of the six doctors and of INDUCE36, without knowing which set belonged to whom. INDUCE36 received the highest average score, and proved its usefulness.
ESROM (Expert System on Rupture Of Membranes) can be used for the diagnosis and management of ruptured membranes in obstetrics. When foetal membranes break and the amniotic fluid inside the uterus leaks from the cervix, it can be a sign of labour and delivery. But when it occurs before labour, it can be associated with infection of the foetus and be dangerous. If the foetus is very premature, immediate delivery may increase the possibility of mortality and morbidity. It is thus very important to maintain the delicate balance between how long the pregnancy should be prolonged and when the baby should be delivered.
ESROM has three goals:
Diagnosis -- to decide whether the membranes are ruptured;
Detection of infection -- to indicate whether infection of the foetus is present; and
Management -- to decide whether the foetus should be delivered.
ESROM has 43 objects, 61 rules, and 4 fuzzy types (size, well-being, amount, and moisture). The researchers tested the system and verified the knowledge base on 30 hypothetical cases, and the results agreed well with the domain experts.
Looking Ahead
The results obtained by ABVAB are encouraging for the development of other medical expert systems based on Z-III, and the verification method for INDUCE36 introduces a new approach to the evaluation of medical expert systems. The multi-layer structure of ESROM demonstrates a way to handle complicated medical diagnoses and treatments.
In collaboration with Prof. Jack Cheng of the Department of Orthopaedics and Traumatology, Prof. Leung is working on another medical expert system for use in spinal orthopaedics.
Prof. Leung hopes that, with the rapid increase in the physical memory of the PC, more complex multi-layer expert systems can be developed by Z-III, and expert system applications can be more widely accepted in fields other than medicine.
Prof. K. S. Leung graduated from the University of London with a B.Sc. (Eng) degree in 1977 and obtained his Ph.D. degree in 1980. He then worked for five years in England as a senior engineer/systems analyst, first at ERA Technology, then at the headquarters computer centre of the Central Electricity Generating Board. He is a member of IEE, BCS, and ACM, a senior member of IEEE, and a chartered engineer.
Prof. Leung's research interests cover knowledge engineering, expert systems, genetic algorithms and programming, automatic knowledge acquisition, Chinese processing, fuzzy logic applications, and AI architecture. Since joining The Chinese University in 1985, he has developed several novel expert systems and shells in various areas of application.
Prof. Leung is now professor in the University's Department of Computer Science and Engineering, and head of the graduate division of computer science.