![]() |
|||
![]()
|
![]() |
![]() |
![]() |
3.1. BAYESIAN STATISTICAL APPROACHESThe Bayesian approach is a mechanism to calculate the probability of a disease, in light of specified evidence, from the a priori probability of the disease and the conditional probabilities relating the observations to the diseases in which they may occur. A variety of additional assumptions must be made in this approach, such as: (1) the diseases under consideration are assumed to be mutually exclusive and exhaustive (i.e., the patient is assumed to have exactly one of n diseases); (2) the clinical observations are assumed to be conditionally independent over a given disease; and (3) the incidence of the symptoms of a disease are assumed to be stationary (i.e., the model generally does not allow for changes in disease patterns over time). In several domains, the technique has been shown to be exceedingly accurate, but there are also several limitations to this approach that make a particular application unrealistic and hence unworkable (Shortliffe et al., 1979):
While much work has been done on Bayesian approaches to medical expert systems in the earlier studies, few applications were recently found, perhaps due to the above limitations. 3.2. RULE-BASED REASONINGRule-based reasoning is the most general structure -- it is also called a production system. It uses knowledge encoded in generation rule (IF ... THEN). Rules usually have a condition part (antecedent) and an action part (conclusion). Each rule represents one of the knowledge units related to an expert field. Many related rules may correspond to an inference chain, which deduces a useful conclusion from several facts known. Rule-based reasoning has been the most popular choice of knowledge engineers for building an expert system in medicine. Through previous experience in developing rule-based systems (including MYCIN) came a better understanding for their development, bringing about an explosion of applications. Of 233 medical expert systems developed since 1992, as seen in the Appendix, 110 systems (47.2%) were developed by the rule-based reasoning. The survey conducted in Germany (Kraut and Mann, 1996) showed an even higher percentage (69%) for the rule-based reasoning approach. There are several advantages of the rule-based reasoning approach (Durkin, 1994). First, it has a natural expression. For many problems, humans naturally express their problem-solving knowledge in IF ... THEN type statements. Second, since rules are independent pieces of knowledge, they can easily be reviewed and verified. Third, rules are transparent, and are certainly far more transparent than the modes of knowledge representation employed by neural networks, as discussed later. However, there are also several disadvantages. First, it requires exact matching. If we do not have an exact match, the rule will not work. Second, rules do not efficiently or naturally capture the representation of complex domain knowledge. Third, systems with a large set of rules can be slow. 3.3. NEURAL NETWORKTo solve some of these problems, there have been several attempts to acquire a knowledge base directly from the medical database. A neural network is one such attempt. A neural network is essentially a type of information processing technology inspired by studies of the designs in the brain and nervous system. Consequently, these systems operate in a fundamentally different manner from traditional computing systems. They are made up of many simple, highly interconnected processing elements that dynamically interact with each other to "learn" or "respond to" information rather than simply carry out algorithmic steps or programmed instructions. Information is represented in a neural network in that pattern of interconnection strengths among the processing elements. Information is processed by a changing pattern of activity distributed across many units. Learning occurs through an interactive adjustment of interconnection strengths based upon information within a learning sample. The neural network is considered to be a kind of nonlinear discriminant function. A self-learning mechanism of the neural network has been applied to medical expert systems as an automatic knowledge acquisition and representation method. As seen in the Appendix, the neural network has been applied to the following medical domains for the last 5 years: analysis of protein structure, male infertility, trauma, neuroscience, nuclear medicine, lung scan, psychiatry, cardiac surgery, cervix cancer, stomach cancer, molecular biology, genetics, asymptomatic liver disease, hepatobiliary disease, epilepsy, pediatric, rheumatic disease, allergic rhinitis, asthma, orthodontia, acute abdominal pain, brain auditory evoked potentials, glaucoma, dentistry, and nurse scheduling. Of 233 expert systems listed in the Appendix, 54 systems (23.2%) were developed by neural network. The survey of expert systems developed in Germany (Kraut and Mann, 1996) also showed a similar percentage for neural networks. Neural networks have various advantages and disadvantages compared with the rule-based system (Hillman, 1990). Rule-based systems tend to be domain specific and function extremely well when problems are well defined, but neural networks have a broad response capability based on their ability to provide a general classification of a set of inputs. Furthermore, the rule-based system implementation can be a lengthy process, depending on the size of the domain and the range of cases that must be realized; neural networks can analyze a large number of cases quickly to provide accurate responses. However, validation of the content of neural network (i.e., the determination of the completeness and consistency of the representation) is relatively more difficult than rule-based reasoning. Another problem in the use of neural networks in medical decision-making has been the evaluation of the reliability of the output of the networks. Moreover, the output of the neural network systems is difficult to interpret for most of clinical practitioners. There should be more studies on these problems to help the user of the system evaluate the confidence on the suggested diagnosis by neural network.
|
![]() |
|
Use of this site is subject certain Terms & Conditions. Copyright (c) 1996-1999 EarthWeb, Inc.. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of EarthWeb is prohibited. Please read our privacy policy for details. |