-Diagnosis, prediction, machine learning, and decision making are all areas of application of artificial intelligence. Particularly, intelligent (medical) diagnosis systems are now becoming pervasive providing support to healthcare delivery. However, there is a lack of precision and approximation of the algorithms driving such diagnostics systems. Though there is a number of reasoning algorithms for carrying out this diagnostic task, the precision of these diagnostic algorithms are being impaired by their reasoning structures. This paper reviews and provides an enhancement to select and test (ST) reasoning algorithm. This algorithm, adjured to be the most precise among the existing diagnostic algorithms, will be enhanced by employing the use of semantic web reasoning structures. Reasoning at the abduction, deduction, and induction levels are oriented towards rule base reasoning pattern in the semantic web. Also, a series of modularized ontology knowledge bases are stacked together in building a complex but distributed knowledge base for the entire system. The implementation of this enhanced algorithm will be used as a test-bed for diagnosing and monitoring related ailments.
INTRODUCTION
Intelligent medical diagnosis is an area of research that leverages on artificial intelligence, and intelligent systems targeted at diagnosis of other systems are becoming pervasive. These intelligent medical systems are also referred to as medical expert systems (MES) and the driving force of these expert systems are reasoning algorithms. Some of such reasoning algorithms are fashioned after mathematical, statistical, fuzzy and rule-based models. To achieve the reasoning of this enhanced algorithm, the semantic web structures are employed. The semantic web is the power of inference making and reasoning over ontologically modeled knowledgebase. Therefore, we use semantic web rule languages to encode our rule systems for effective interoperability with the knowledgebase. As a result, taxonomies, metadata, classifications, context and ontology have been the basic building blocks of the Semantic Web [1]. A combination of a formal ontology for the medical domain with a fine-grained contextual inference making and reasoning algorithm is an exceptional tool in incorporating autonomous (health) systems.
This research in progress argues that employing the use of some semantic web technologies in ST algorithm will yield a higher precision, and an inference making medical diagnostic reasoning system.
II.
Expert systems are a program intended to make reasoned judgments or give assistance in a complex area in which human skills are fallible or scarce. Considering the work of [2] which stated that they are computer system that operates by applying an inference mechanism to a body of specialist expertise represented in the form of 'knowledge'. They are employed as decision support systems, and have many implementation approaches which includes; rule-base (MYCIN and PROSPECTOR), data-base approach, descriptive method (INTERNIST and CADUCEUS), and Causal Network method. Also, [3] developed the expert system that carries out its diagnoses by organizing symptoms into three groups namely Key group(Kg), Sub group(Sg) and Unexpected(Ue). Again, [4] designed ASTHMA, an expert system for the diagnoses of asthma. They combined some machine learning algorithms such as Context sensitive autoassociative memory neural network model (CSAMM), Backpropogation model, C4.5 algorithm, Bayesian Network, Particle Swarm Optimization to realize their design. Ex-Dr Verdis is an integrated expert system [5] Heart Disease Program (HDP) is a medical expert system that enables physicians to enter patient’s symptoms, laboratory tests, and physical examination. It then generates clinical data that support the diagnoses of heart disease [6].
Meanwhile, a review of medical reasoning algorithms is
discussed here. Scheme inductive reasoning algorithm works
based on forward thinking [7]. Pattern recognition is
employed in machine learning for assigning some outputs to
some inputs base on the coordination of a given algorithm
[8]. Hypothetico-deductive reasoning involves generating
and testing hypotheses in association with the patient’s
presenting symptoms and signs [9] Forward chaining system,
includes writing rules to manage sub goals. Whereas,
backward chaining systems automatically manage sub goals
[
In this section, we present and anatomize our enhanced
ST model. The modified model consists of the Abstraction
module and the three logical inference modules namely
Abduction, Deduction and Induction. More so, the existing
ST algorithm data is a-temporal (unable to monitor and store
relevant events, that could support diagnostic reasoning, with
respect to time of their occurrence), hence, this made us to
add a monitoring module to the ST model so as to make it
data-gathering procedure consistent and as well temporal.
Contrary to the ST model by [
The following subsections give a breakdown of each of the consisting components in figure 2.
Our abduction stage improves on the existing abduction module. Except that we enabled a semantic web based reasoning operation in the module. We propose the use of rule engine for this reasoning task. Both the abduction and the deduction stages here uses this rule engines. And compose a rule system for aiding diagnostic reasoning task. A new parameter, acceptanceThreshold is added to the existing likelihoodThreshood parameter. This to check if every deduction task passes a given acceptance value before we can conclude that it is correct. The abduction modules gets all diagnoses related to symptoms found, and reasons by hypothesis, studying facts and devising theory to explain it. The process of abduction: The whole process of abduction includes generation, criticism and acceptance of explanatory hypotheses.
Deductive Reasoning is a process in which general premises are used to obtain a specific inference. A form of logic that identifies a particular item by its resemblance to a set of accepted facts. Deductive reasoning moves from a general principle to a specific conclusion. It is inference by reasoning from generals to particulars. Deductions support their conclusions with TRUE result. They compute their results using heuristics. We modify the existing deduction module to be a rule-base deductive reasoning task. Hence, a coordinated rule system and a reasoned are added to semantically realize the deductive reasoning.
The process of mapping descriptive terms understood by patient onto a well-defined symptom entities modeled in the knowledgebase is known as abstraction. In this proposal, we seek to provide patients with a textbox for inputting their entering descriptive terms of how they feel. Our natural language NL-query to Semantic Web SW-query model, then semantically matches their inputs against an ontology of vocabularies in the knowledgebase. The modified abstraction module allows input to be in speech or textual. Patients may voice in their symptoms and this data will be processed by the voice processor.
It entails reasoning from the particular to the general. This may or may not be true. But it provides a useful generalization. At the induction state, we check if likely diagnosis meets diagnostic criteria. While Abduction & Deduction are termed clinical reasoning, Induction is termed clinical decision making. The induction stage in this modified ST model builds on the existing features of the existing ST model. Except that we develop a mathematical model for computing the criticalThreshold parameters, which is now added to calibrate and alert patient on the status of the ailment.
The monitoring agent works continuously in the system. It is more like a daemon which logs events into a SpatialTemporal-Thematic (STT) ontological database. The essence of agent is to be able to monitor development of the ailment in the patient’s body, and then adequately signal the needed alert or logs necessary information that the diagnostic algorithm will mine data from it. Temporal information gathered is a clinical data that helps in tracking the progression of a disease in a patient with respect to time. Spatial information models data that relates with patient and its environment. Thematic data models concepts and terms used in clinical operations.
Figure 3 shows the model that the monitoring module works on. On the left hand side, there are four components: the event monitoring, event selector, data gathering, and data modeling in spatial-temporal-thematic (STT) format. The event monitor receives information from the intelligent personal agent and then sends it to the event selector which appropriately sieve out the right information to store. Then the data gathering and reasoning faculty generates the requisite data base on the event. The last component then models the data in a STT format.
On the right hand side of Figure 3 is the knowledge modeling in OWL2. This will be fully discussed in chapter four. Now, the algorithm for this model will be discussed below.
Algorithm 1 lists out the steps required for performing the monitoring task of the monitoring module. Line 2of the algorithm rightly points out the fact that this monitoring module does its tasks as long as the application is running. The module sources its event data by raising some important questions at random, and then from the intelligent personal agent which shall be discussed in the next section. Once these data are gathered as indicated by lines 4-5, lines6-12 cleans the data, formats it into the required style and then models it in the STT pattern. Afterward, the patient or user is alerted in the case of any information that must be passed across.
IV.
THE KNOWLEDGE REPRESENTATION STACK OF THE
ENHANCED ST ALGORITHM
The proposed enhanced Select and Test (ST) algorithm discussed above is a rule base expert system. Figure 4 is a structural representation of the facts and coordinated rules system. The structure consists of four layers, and each layer comprises of facts (model with ontological language OWL) and rules (modeled with semantic web based rule languages, SWRL). Appended to these four strata is the monitoring agent knowledge model. The first layer models the knowledge of the abstraction layer. Basically, there are three modularized data representation. These are a thesaurus modeled with OWL, patient’s personal profile modeled with XML, and a rule for mapping descriptive terms of patients into a well-defined symptoms entities, modeled with SWRL. The second layer is a knowledge representation for the abduction phase. Knowledge representation at this phase comprises of the facts, modeled with OWL, and rule set for carrying out abduction (modeled with semantic web rule language SWRL). The deduction module is the next phase for knowledge modeling representation. This phase has a rule set modeled in Jess rule language (JessRL), and the facts modeled in OWL also. Algorithm 1: Monitoring module algorithm the responsibility of the induction layer to reason out the correct diagnosis from this three Di. This is achieved by reasoning out which one meet the diagnostic criteria.
Lastly, the induction phase also comprises of an ontological knowledge base and a rule set for induction, with the rule modeled with the Jena rules. The last component in this structured knowledge model is spatial-temporal-thematic ontological representation of the data generated during the monitoring process.
V.
RESULTS AND DISCUSSION
As proof of concept, we listed some common symptoms associated with skin disorders. These symptoms are identifiable with some specific skin disorder disease or ailment. The proposed enhanced ST algorithm proposed in this paper help patient to diagnose the particular skin disorder he/she is suffering from. We note that all Di and Sj are modeled ontologically alongside the necessary rule sets able to help reason out the set of Sj that could cause a particular Di
For example, given disease Di, we will need to cyclically reason through abstraction, abduction, deduction, until a refined result is reasoned out, then we move on to the induction where the real ailment is inductively picked out of the few left after the cyclic refinement.
We need to established Di manifesting a set of S, once our reasoning structure is able to accurately establish this set of S, then considering users/patient’s input; a correct diagnoses process is completed.
At the Abstraction layer, users input are collected and stored as symptoms. Furthermore, at the abduction layer, associated skin diseases or disorders of the collected symptoms are reasoned out of the modularized knowledgebase of the abduction layer, alongside its rule sets. These skin disorders or disease retrieved at the abduction layer is then sent to the deduction layer. The deduction layer must then intelligent map out the Di manifesting a set of S, for every disorder or disease. This cyclic pattern is continued until it comes to the induction layer. Say at the induction layer, three Di are sent into this module, it is Symptom ID
S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 S11 S12 S13 S14 S15 S16 S17 S18 S19 S20 S21
CONCLUSION
In this paper, we have introduced an enhanced Select and Test algorithm which may be employed in diagnosing disease or ailments that have related symptoms. Though it is a research in progress, however, this paper has introduced the model of the proposed enhanced algorithm. It shows four levels of reasoning: Abstraction, Abduction, Deduction and Induction. Furthermore, the knowledge stack of the modeled was as presented. And finally, a proof of concept was shown so as to explain the implementation of the algorithm.