A Study on the Expert System Automation Module for Comprehensive Evaluation of Clinical Pathology Results

Main Article Content

Jai-Woo Oh

Abstract

Background/Objectives: The purpose of this study is to implement a rule-based automated diagnostic test result interpretation module to automate the process of presenting the findings based on the diagnostic test results for accurate diagnosis and treatment of health screening examinees.


Methods/Statistical analysis: This study involved a comparative analysis of the findings presented by laboratory medicine specialists and the findings presented by a rule-based intelligent system based on health screening results. Test results were classified by case, and the findings were classified accordingly. Rules were predicted for a total of 204 cases, which were classified into 22 types under 5 different groups in order to implement a rule-based automation module and to validate the module. The inference engine used for this purpose was Neuron Data’s Intelligent Rules Element, which is capable of dynamic object creation and simultaneous forward and backward inferences, while Visual C++ and Neuron Data’s Open Interface Element were used as the interface between the system that was developed and the user.


Findings: The predictability of the rule-based automation module and the feasibility of applying the module in a clinical setting were examined by using the diagnostic test results of health screening examinees. First, with respect to the inference result on quality control, the findings regarding the test results of patients detected at the ∆ check value or panic value check showed 100% predictive power. Second, with respect to the results of individual diagnostic tests, the findings on glucose, virus, morphology, and bacteria matched 100%, but the inference results related to the suspicion of specific diseases matched in 25 of 28 cases, indicating a match rate of 84%. Third, the analysis of the association between the test results showed a match rate of 94.4%, on average. Fourth, in the case of ordering additional tests, there was an 81.3% match rate, the lowest predictive power among all the items examined, and it was deemed that the reliability of the prediction was still low. This study revealed the need to set a standard reference value for each test in the rule-based automation module, taking into account several variables, such as when results related to other test items are displayed and when processing a single test item.


Improvements/Applications: The results of this study showed that the rule-based automation module developed can improve the quality of medical services by utilizing the accumulated knowledge from ordering tests to drafting a written opinion on the findings through comprehensive and efficient management of test results and patient information.

Article Details

Section
Articles