A Study on the Automatic Control System for Quality Assurance of Clinical Pathology Examination Applying Machine Learning Technology

Abstract

The object of research is a study on a machine learning-based integrated quality control system to provide accurate test results quickly by improving the accuracy of test results and improving the utilization rate of test equipment by analyzing clinical pathology Examination data. As for the development method for research, IEC 62304, the international standard for medical S/W life cycle used in medical software development, was applied, and the S/W life cycle rule was applied to all processors by integrating agile methodology. Random Forest, XGboost (extreme Gradient Boosting), LightGBM (Light Gradient Boosting Machine), and DNN (Deep Neural Networks) to improve Rule Check accuracy to predict abnormality of inspection results, and to monitor abnormal results in real time by rule check calculation method technique was applied. In addition, to improve user convenience, automatic control and user interface technology using a dashboard function applied with machine learning technology, and automatic interlocking interface technology for various heterogeneous inspection devices were applied. Validation and verification were conducted through a qualified testing body (TTA) to ensure reliability. The following results were found through this study. As a result of implementing a module that applied big data-based machine learning technology to the algorithm used for quality control judgment of the first knowledge-based expert system, it was possible to implement a module with more than 95% accuracy. there was. Second, in order to determine whether a real-time alarm function was provided, the development module was linked to the clinical pathology information system and as a result of the experiment, it was found that it was operating normally. In addition, reliability was secured through certification by an accredited certification body. Third, as a communication support method for the interface of the inspection equipment, stability and various technologies were secured through a number of communication tests and certification tests such as RS232C, TCP/IP, and Serial HL7. Fourth, through multiple database tests (Oracle, MSSQL, MySQL, MS Access, etc.), cost savings were secured by resolving duplicate investment by providing database neutrality and interface with other systems. Fifth, utility and user satisfaction were enhanced by providing program functions for outputting the result report in various formats and configuring the UI settings, and the UI settings were modularized to reduce the program development costs and allow the modules to be reused. Through the results of research, small and medium hospitals can improve the reliability of inspection results through the machine learning-based quality control module, and through real-time monitoring of the inspection equipment, it is possible to quickly determine whether there is a failure and improve the operation rate of the inspection equipment. In addition, by providing a module that can be linked with the existing information system, it was made easy to link, and the convenience of the user was improved by providing various UI environments. As a result, it can be expected that the hospital's competitiveness and medical service will be improved by resolving the difficulties of quality control that small and medium-sized hospitals had and providing prompt and accurate test results.