This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.
This paper describes the research and approach to the analysis of a medical knowledge management system that involves data acquisition and its representational formalisms based on processed information and causal knowledge mechanisms. In doing this, a use-case scenario is described with the central idea of integrating medical knowledge with available resources based on the Internet. This will satisfy the need of making medical knowledge and advice available to users based on services provided by the embedded expert system.
The methodology presented for modelling the Medical Advisory Systems (MAS) follows a prototypical method used for hybrid expert systems that forms an integral part of an information system. A basic contribution in this paper is the upgrading of the medical information system based on accumulated medical records format to a knowledge-based format that is supported by the system manager interface. In addition a real-time information exchange interface that augments the user information with direct support from operational physicians is accounted for.
Expert systems are the most visible products of Artificial Intelligence (AI) that have proved their efficiency and usefulness in several areas of knowledge management. This is in spite of the varieties of implementation orientations adopted. Normally, semantic modelling based on symbolic manipulation or a fusion of knowledge acquisition algorithms including statistical based clustering methods and fuzzy-system based soft-computing methods such as adaptive fuzzy logic, neural networks and genetic algorithms are used.
For expert systems to perform they require knowledge and this knowledge is not easy to come by because of the mode of acquisition and representation. Many researchers have identified the availability of data as the backbone of knowledge acquisition.
This is true to the extent of understanding that we put into the data. Presently, data can be viewed as abundant, vital and a necessary resource. However, a collection of data is not information. We can tap into the reservoir of data that pervade the world of computers and the Internet, by utilizing new ways to channel raw data into meaningful information. That information can in turn become the knowledge that leads to wisdom. But, a collection of information again is not knowledge. Information resource exploitation at present mainly concern how to gather, code, digitize, store, classify, control, process, transmit, retrieve, measure data and information. It is the user of information who must judge whether or not a piece of information has any usefulness. That a collection of information is not knowledge means that a collection of data for which there is no relation between the pieces is not information. Now, we can utilize new information and communication technologies to extract knowledge from data. From the knowledge engineering perspective, we can simply state that information is an understanding of the relationship between pieces of data or between pieces of data and other information. By knowledge engineering we mean the process through which knowledge is transformed and transferred from the forms in which it is available in the world into forms that can be used by knowledge systems. Knowledge engineering activity, therefore involves knowledge acquisition, representation and use based on understanding provided by knowledge as known in the source domain.
However, while information entails an understanding of the relations between data items, it does not provide a foundation for why a particular piece of data is what it is nor an indication as to how the data is likely to change over time. Information has a tendency to be relatively static in time and linear in nature. Therefore, it is pertinent to say that information is what it is based on its great dependency on context for its meaning and implication for the future. Semantically speaking, in the context of knowledge engineering, understanding comes from people and knowledge in the system is implemented using formal symbol structures - knowledge representation. Knowledge engineering problems bothers on basic decisions about what knowledge to look for, how to use the knowledge, how to represent it, and how to extract it from people who have it and encode it as knowledge base.
The extraction of knowledge from a domain expert is one of the stages of knowledge acquisition and represents a Herculean task. So, seeking for meaning in data or information becomes a daunting task asking for clarification. According to Bateson (1988), beyond relation there is what is called pattern. Pattern is more than simply a collection of relations. It embodies both a consistency and completeness of relations, which to an extent creates its own context. Pattern serves as an arch-type with both an implied repeatability and predictability. When a pattern relation exists amidst data and information, the pattern has the potential to represent knowledge. It only becomes knowledge when the patterns and their implications are realized and understood to be more self-contextualizing. This meaning that the pattern tends to a great extent to create its own context rather than being context-dependent to the same extent that information is. A pattern which represents knowledge also provides a high level of reliability or predictability as to how the pattern will evolve over time. This can happen only when the pattern is understood or learned. . Patterns again are seldom static, so, patterns which represent knowledge possess completeness to them, which information simply does not have.
It is knowledge that governs how future information are processed and interpreted for man's welfare and the services they derive. Along with the rising tide of the Internet, a medical knowledge can persist in a knowledge based environment where resources abound to support services on offer.
The medical expert (i.e. the physician) deals daily with patients; generating data and making decisions that concern these patients. The physician can be supported with a lot of knowledge to ease out this function through knowledge extraction and inferences based on them and a special system modelled to emulate this knowledge.
The structure and function of a medical knowledge-based system anchored on the Internet linkage is underscored and presented in this paper. The methodology that generated this system is also explained.
Knowledge-Based Service Centre
In the context of an organization that is engaged in providing medical services to clients (patient and medical personnel), data represents facts or values of result and relations between data or other relations that have the capacity to represent information. Patterns of relations of data and information and other patterns, in general terms have the capacity to represent knowledge. For any representation to be useful, it must be understood, and when understood, the representation is information or knowledge to someone that understands. In knowledge engineering, semantic rule patterns are used to ensure control in expert systems in order to provide usability to the developed system.
Medical knowledge can be described as either implicit or explicit. The explicit knowledge is that which is expressed in the knowledge base of an expert system. Such knowledge is often extracted from domain experts. The doctor for example treats patients based on the knowledge acquired over the years in training and in practice. It is this type of knowledge which is lodged in the mind (or brain) of the domain expert that is referred to as implicit knowledge. This type of knowledge is described as tacit and difficult to express and so constitutes a problem situation for knowledge engineering to respond to.
The very idea of knowledge acquisition and representation in order to provide service results in the development of a knowledge base which consists of a data base of facts and a rule base. This is a kind of knowledge transfer from one object to another. In this case, the explicit knowledge that is stored in an expert system and the implicit knowledge belonging to the doctor are now transferred to the users who are not knowledgeable in the subject domain. Such users may include patients who seek medical advice or medical personnel who seek knowledge in particular subject domain. There are also the expert users who augment practice with expert advice provided by the expert systems; such medical advice aids diagnosis and saves time.
Explicit knowledge transfer can be realized by an expert system using direct methods to extract knowledge from the domain experts, or indirectly by learning systems that use historical data to induce rule. Explicit medical knowledge acquisition involves eliciting facts and procedures from the doctor who understands issues concerning patients' diseases, symptoms and the way diagnosis is conducted in order to arrive at certain conclusions (i.e. treatment, prescription of drugs, etc). This knowledge is made explicit through representation and transformations in form of rules and facts such that the computer can process it to produce valid information.
The Medical Advisory System (MAS) is a kind of information system that is based on specialized knowledge-based service system. A knowledge-based system (KBS) by definition is an expert system that transfers the domain experts' knowledge which it stores in the knowledge base for the benefit of its users. In order to function and provide services the KBS uses an inference mechanism to convey information from the knowledge base to the users of the system. The user will normally send a query to the KBS which it dutifully replies to by sending a message to the inference engine which uses a forward-chaining or backward chaining mechanism to ferret out results. By implication, we say that the expert system (the KBS) reasons or thinks with the inference engine by using the expert's (i.e. doctor's) knowledge stored in the knowledge based and outputs the results of such thinking to the user any time the user issues out a query.
Transferring implicit knowledge from the expert remains the most difficult aspect of knowledge engineering. Extracting medical knowledge, especially diagnostic knowledge requires deep understanding of the medical field especially as known by doctors. One major reason is that experts find it difficult to explain what they know and how they know things. It is left to the knowledge engineer to fathom how to get this knowledge. The rule-of-thumb or heuristics used by experts are studied and expressed as knowledge. Sometimes, they fail to get the things right. However, such knowledge shift or transfer from the expert to the machine (the computer) will require getting existing facts from medical books, discussions with medical experts, use of surveys, interviews and exchange of ideas between domain experts and the knowledge engineer and of course the patients. Experimentation with subjects also generates data (facts) by which these expert systems are validated and verified and proved satisfactory for real life usage. Thus, the link between those who have knowledge and those who need knowledge must be established. The implicit knowledge that is connected on the Internet is another kind of transfer of knowledge via the Internet. The Internet bridges the gap between the people who have knowledge and those who need it. The basic contribution of this paper hinges on the satisfiability of the needs of users of computer applications who need expert advice. This user satisfaction is ensured by expanding the medial advisory system to offer services in three dimensions - data, information, knowledge - over the Internet.
Knowledge management entails therefore, the ability to store and enforce a collection of rules that are part of the semantics of an application. Such rules describe integrity constraints about the system in use and as well allow the derivation of data that is not directly stored in the system. The main idea is that if the knowledge engineer has the need to obtain some basic rules, and a set of examples from the domain expert, it is possible to define the basic structure of the intended system with its facts. The structure of the medical advisory system (MAS) is presented and described in the next section.
The Basic Structure of the Medical Advisory System
The MAS structure is shown in Figure 1 and consists of three main blocks: System Management (SM), the Real-Time Information Exchange System (RTIES) and the Medical Expert System (MES) with their accompanying interfaces.
The first block in MAS is the systems management component, which manages all activities and services provided by the system.
The second block contains other real-time information exchange system (RTIES) which among other things provides information on the medical experts operating the system, the question recognition modules, the expert selection module and the real time information exchange module between the expert and the user. This exchange could be between the doctor and his patients or the doctor and his colleagues.
Real-time Information Exchange System
Figure 1: MAS Structure
The third block is the medical expert-system (MES) which includes the knowledge base (KB), inference engine (IE) and the user interface (UI).
It is pertinent for us to further explore the structure of MAS by taking a closer look at the knowledge-based expert system component, the MES, which processes and advises the users based on their queries. First we shall consider the methodology employed in modelling the system.
Obtaining adequate rules from domain experts is a real difficulty. However, from the research point of view, knowledge based systems have the potential to help organize and synthesize knowledge information of different types. It is therefore imperative for knowledge engineers to focus and apply diverse avenues of research to solve difficult problems and link together quantitative data, simulation models, basic research results into knowledge-based models of how difficult medical decisions can be made. The basic idea of expert systems development shifts the focus of research in knowledge management to knowledge dissemination in contrast to knowledge accumulation. Thus, the experts system, in combination with powerful ICT resources and new technological innovations has the potential to open up a whole warehouse of accumulated knowledge to medical practice. This can be achieved based on knowledge extraction functions, knowledge storage with minimum logic representation and transformations from the traditional data model to a special way of knowledge representation and putting medical knowledge into the knowledge-base of the MAS expert system.
The knowledge engineering process as used in this research is shown in Figure 2 as a knowledge extraction cycle that depicts how knowledge is extracted from experts. The physician is extremely important in our development, as he is primarily a potential user.
thru training or adaptation
Figure 2: Knowledge extraction cycle
Initially, the task of knowledge acquisition consists in extracting knowledge of the domain expert, who in this case is the doctor, by defining rules in a short time such that a series of examples can also be supplied for system implementation. Figure 3 shows the relationship as acknowledged by knowledge engineering experts.
Sets of Basic Rules (IFâ€¦Then)
Figure 3. Relationship between the knowledge engineer and the domain expert.
The basic rule sets could be improved to capture uncertainties associated with human cognitive processes, such as reasoning, and/or thinking. The versatile and forward looking models for current endeavours are those proposed through fuzzy logic. These basic rules take the generalized form of fuzzy-rules (Zadeh, 1988, Zimmermann, 1991) However, the generalized rule structure or pattern is a series of the IF---THEN type of rules of the form:
The basic rules are translated by AND/OR graphs, such that they define the basic structure of a knowledge-based system. In other words, the semantic graph represents layers of evaluation in the rule's network architecture.
The basic rules consist of the IF-part (antecedent) and the THEN-part (consequent), such that the IF-part indicates the premise of argument and expresses as the present medical situation the symptoms of a disease, while the THEN-part deals with the action and expresses the possible diagnoses (Fu, 1992, Gupta & Rao1994).
Each of the rules presents a membership degree which represents the connection weights among items on the network. The membership degrees can be as in IF-part as in the THEN-part of the rules or only in the then part.
Validation will commence after the basic rules have been extracted from the set of examples. In other words, a correspondence is established between the obtained solutions with those given by the domain experts. Where a negative case is encountered, it is referred back to the expert to solve and adapt for future use.
The cases that succeed are saved in the data base of the expert system, where they are used with a learning algorithm to refine the MES. A learning algorithm is a form of adaptation of the facts to generate new rules or patterns that constitute knowledge, as explained earlier. The domain expert is now hired to validate the knowledge based on the modification in the basic structure of the rule network. Hence, a new set of facts (sample cases) are generated to test the system again. Where the system performance is good, it is supposed to represent the goal as proposed by the designers of the expert system.
3. A DIAGNOSTIC MEDICAL EXPERT SYSTEM
This section presents an application in which decision support is needed to advice on possible medical advice for patients with multiple symptoms. Depending on the information given, a possible GP (General Practitioner) patterned expert system can be evoked. The system called "medical expert system (MES)" describes a web-based application embedded as the diagnostic component of the Medical Advisory System (MAS). The MES consists of two sub-systems (the inference engine and the knowledge base) and a user interface. The MES provide access to medical advice to patients, especially those who live in the rural communities where access to medical personnel for medical advice is scarce. The system is designed as web-pages on the Internet based on the expert system methodology explained earlier. The knowledge-base part contains static pages in the html format which provides information on common diseases, symptoms, investigation, drugs, services and preventive measures. These knowledge artefacts are collected from domain experts from the relevant domains of medicine that are subscribed to by the system. The rule-base of the MES as contained in the knowledge-base pertains to the diagnostic problems it solves and provides dynamic pages, which contain expert advice simulated on the domain to the end users who are interfaced online with the system. The user interacts with the system by providing answers to questions posed by the system as reflected by the control strategies employed by the inference engine, after making first contact by logging himself into the MAS application.
To support our earlier discussion in section 1 on the contextual nature of information and knowledge based on data interpretation, we relate it to the knowledge used by the medical advisory system by theorising as follows that:
On data, we speculate that items of data such as the numbers 190 and 140 on their face value are completely out of context and can be considered as just pieces of data. But when semantically analysed, they can be interpreted in a variety of ways. Their meanings therefore can only be known only when they are viewed in contextual forms, that is, when the system binds them to some meaningful concept the user may understand.
On information, if we can establish hypertension as the basis of context, then, blood pressure, systolic blood pressure, and diastolic blood pressure become meaningful in the context in which specific interpretations are made. This then, is what we call information and we can have:
The systolic measure of blood pressure is 190
The diastolic measure of blood pressure is 140.
On knowledge, therefore we recognise what it is when a patient is sick and the systolic blood pressure is 190 and the diastolic blood pressure is 140, then based on the interpretation of the semantic rule pattern that both information make on the system, the system will be able to say that the pattern is hypertensive and requires immediate attention of the physician. This semantic pattern as represented by rules of inference represents what we call knowledge which if understood, allows the understanding of how the pattern will evolve over time and the results it will produce. If the blood pressure readings get higher, the patient would die abruptly or develop stroke. If we gather many examples of these data with their consequences, the knowledge pattern will be established as to what happens as a result of particular combination of the blood pressure readings in a patient which further enhances the basic knowledge gathered from the domain expert.
Prototype Specification of the MES
The basic architecture of an expert system will normally include two principal components which are the knowledge base and the inference engine. However, the components of the MES as explained earlier will include the knowledge base block, inference engine block and the user interface which connects the user with the system.
The Knowledge base consists of facts and rules stored in the database and the rule base and represents the problem-dependent set of data declarations.
The User Interface establishes the interaction between the user and the inference engineer. Here the graphical user interface is mostly built with controls such as menu bars, iconic buttons, labels, and combo and list boxes among others. Most times, the user interface is built to be user friendly and flexible for browsing purposes as provided by the web-based interfaces.
The Inference Engine is the heart of the expert system and has the main effect to infer results from the facts and knowledge stored in the knowledge base. It is the problem-independent program that relates knowledge to data in order to produce information The systems inference is controlled by different rules of symbolic logic like modus ponens, resolution, unification, etc that seek out relevant information while ignoring irrelevant facts to the case at hand. It is the set of rules in the inference engine that forms the main driver program of the expert system. Inference or reasoning in medical expert systems is normally by deduction or by abduction. Deduction uses antecedents of rules to infer conclusion of a rule from implications of the rules premise and the rules consequents. This takes the form of modus ponens of reasoning of the form
Abduction is commonly practiced in medical diagnosis by physicians who use antecedents of rules based on the implications of the rule premises to infer a conclusion from the premise and its consequent. This computationally abduction has the form
According to Charniak and McDermott (1985), relationships between causes and effect can always be represented by implication. Such causal relationships have been described as abduction, where inferences are made by inferring causes from observations. In abduction, causal relationships would match P in (3) with physiological states or symptoms and Q would be diseases. All depends on the heuristics applied by the domain experts in diagnosing human health problem situations.
A Use-Case pattern analysis diagram of the MES based on a UML representational formalism is shown in figure 4. The use-case pattern analysis is generally used to represent the generic aspect of a system and can be in the narrative form or be diagrammatically represented.
Domain Expert (the physician)
Figure 4. Use-case diagram of a generic medical expert system.
A class structure is a structure diagram that describes the static structure of a system. It describes how the system is structured instead of how it behaves. A class on the other hand is a group or category of things that have similar attributes and common behaviours (i.e. attributes, services, rules and relationships).
Figure 5 shows the generic class pattern for the inference engine of the MES block of the advisory system. This represents the run time view of the requirements of the system.
The core objects is the knowledge base which has all the major attributes requested for by the various methods in the user and system manager classes. These classes are considered as views in some literature. The user and systems manager views execute the messages that are exclusive competencies, and delegate the rest; in fact, they add states and behaviours. Many more classes can be built in order to integrate new behaviour to the existing views or add more methods to existing classes.
The knowledge base, which includes rules and facts, is a problem-specific module that contains information that controls the inference process. It traditionally employs the IF...THEN rules to represent information. Goals, including sub-goals are defined to be achieved by the MES. For example, given the goals below, the rules to achieve them are defined in the rule base. With facts supplied by the user (i.e. the patient), the goals are achieved by few rules defined for them.
The users may have the goals: hypertension, malaria, cholera, typhoid, dysentery, and diarrhoea. Some could be achieved with a single or multiple rules as given by the production system below:
RI: IF the patient is running temperature
AND if the patient is having strong headache,
AND if the patient is restless,
THEN the patient may have Malaria.
R2: IF the patient is having yellow eye,
AND if the patient is running temperature,
AND if the patient is having running stomach,
AND if the patient is having yellow urine,
THEN the patient may have Typhoid.
+void: diagnosis ()
+void: advise ()
+void: diagnosis ()
+void: advise ()
Figure 5. Class diagram of the MES advisory system.
R3: IF the patient is running temperature,
AND IF the patient is having severe headache
AND IF the patient is having running stomach
THEN the patient may have Cholera.
R4: IF the patient is having systolic blood pressure greater than 120
AND if the patient is having diastolic blood pressure greater than 90
AND IF the patient is having severe headache
THEN the patient may be having high blood pressure
R5: IF the patient is having high blood pressure
THEN the patient may have hypertension
A complete process is normally represented using the activity chart to depict the flow of control or transition from one activity to another. Figure 6 shows the activity chart for diagnosis in the MES advisory system.
4. RESULTS AND DISCUSSION
We have kept faith with knowledge distinction that helps explain or interpret information we get using applications that require data or raw facts in order to execute. When the system is operational and the sample goals are tested against the rules defined in the rule base, the patient's goal will succeed or fail depending on the supporting facts available to the system. Since the system lists the diseases as goals to be achieved, picking any of the options that concerns a patient's health condition will constrain the search to the targeted sub-goals that constitute the premise in order to match the conclusions reached as actions or consequents.
The premise side of the rule is used by the system to extract information from the user who answers "yes" or "no" to question posed to him.
Figure 6: Activity Chart of MAS
Based on the answers given the consequent rule can be linked to the goal situation or some other rules are tried when the facts fails to match the given premise and therefore did not satisfy the conclusion with the goal.
It is the systems management system (SMS) that provides the basic facts that describe the static information requirement of the MAS, such as the functions of the body, the patient's bio-data, the available diseases, symptoms, doctors, and others.
On the other hand, the dynamic part allows users to query the system and based on these queries, provides the result of diagnosis. A valid conclusion will normally return to the user information like disease name, possible tests to be carried out, treatment and some other pertinent information. Generally, all these constitute the medical advice on the health condition of the patient.
In effect, the medical advisory system as portrayed in this paper is referred to as an expert system. This is because the systems' interface emulates the functionality of the domain expert (the physician) who is very knowledgeable in the subject area. This it does at real time using knowledge artefacts embedded in its component parts.
The paper sets out to show how knowledge can be transferred from one object to another, especially from a human agent, the domain expert to the machine through artificial means. Artificial intelligence had given us the ontological basis and tools to deal with such knowledge transfer issues, especially in narrow domains. It is not an easy task to build expertise artificially in large domains due to the bottleneck connected with knowledge acquisition and representational formalisms.
We have mainly demonstrated the capability of AI to deal with issues that concern knowledge acquisition, knowledge representation, reasoning, knowledge exchange and information dissemination through the services provided by expert system based on the Internet protocol.
The advantage of an information system based on the expert system formalism as we have shown is the capability of an ES, to represent knowledge with minimum logic units. Another advantage is the basic transformation from the traditional data model to the special knowledge engineering model that uses semantic modelling to represent knowledge as semantic rule patterns. This novel way of representation makes medical knowledge available for decision-making, especially for non-medical users and rural dwellers who can always log into the Internet for medical advice. Instead of the traditional data and programs, medical expert systems now deal mainly with facts and rules which are stored in systems with special formats as knowledge bases.
To construct a full-blown medical advisory system requires the involvement of several medical experts with specialized knowledge in different aspects of medicine. From time the medical profession had provided the knowledge engineering professionals with the initial tool that were adapted for diagnostic systems. MYCIN (Shortliffe, 1976) as the pioneer medical expert system has remained the most educative real-life expert system in existence today. The physicians have always had to specialise their knowledge by training in narrow domains, which has become the basis of analogy in many expert systems that are daily spawned in the field of medicine through knowledge adaptation.
This paper has revealed the potential of an AI-based medical expert system which if deployed on the Internet will fast track the reach of medical advice to the hapless poor in distant and rural environments where medical care is near absent or totally traditional. We strongly recommend the adoption of medical expert systems by the Federal and State ministries of health and their agencies to provide appropriate health management services and advice particularly to rural dwellers and the public at large.
We envisage that in due course expert system interfaces must be developed in local languages for medical advisory systems, even where their contents are generated using specialized languages. However, on the Government side, we hope that MES advisory centres will be built country-wise to add/replace the TV viewing centres in order to promote perceptible national psyche and mentality based on good life and healthy living.