РАЗРАБОТКА И АПРОБАЦИЯ ПРОТОТИПА ЭКСПЕРТНОЙ СИСТЕМЫ ДЛЯ ДИАГНОСТИКИ ПОДАГРЫ

Abstract

Significance. Gout is a rheumatological disease that affects up to 2% of the population of the Russian Federation according to various sources. The pathogenesis of the disease is associated with a disturbance in uric acid metabolism and manifests itself in the deposition of monosodium urate crystals in various tissues of the body. Timely diagnosis and treatment of the disease can prevent the development of gout complications and disability of the patient, however, according to various sources, it takes from 44 to 8 years to diagnose gout. One possible way to improve the quality of gout diagnosis is to develop and implement an expert system for preliminary diagnosis, diagnostic techniques and consultations, and final diagnosis. The purpose of the study is to develop and test a prototype expert system for gout diagnosis. Material and methods. The study used system analysis methods, including decomposition, analysis and synthesis methods, knowledge engineering methods, including knowledge representation methods (frames, semantic networks), ontological approach, fuzzy logic, hierarchical structuring methods and effective evaluation methods, including ROC analysis. Results. A prototype expert system for gout diagnosis has been developed. The prototype consists of a knowledge base, a logical solver, and an explanation subsystem. A specialized knowledge base editor was created for the development and subsequent maintenance of the knowledge base. The knowledge base was filled using 2 information objects jointly developed with a group of experts: the nomenclature of medical concepts (495 basic and 679 synonymous concepts) and logical schemes (40 schemes) related to gout diagnosis. The structure of the knowledge base is a combination of frames and semantic networks and includes 18 objects (7 types of concepts and 11 types of relationships). A graph database is used to store the knowledge base that determines the logic of the logical solver and the explanation subsystem. The logical solver analyzes the patient's clinical data and searches the optimal path, the end points of which are the recommendations of the expert system prototype. The explanation subsystem uses this path to argue the selected solution. The prototype expert system was tested on two depersonalized samples containing 1725 patients provided by the Tyumen Healthcare Department. Conclusion. A prototype expert system has been developed to determine the preliminary diagnosis of gout, prescribe diagnostic techniques and consultations for its verification, and make a final diagnosis. The prototype includes 1174 concepts and 5,640,522 relationships. Testing of the prototype expert system with ROC-analysis showed that its sensitivity was 91.3% CI [89.1%; 93.1%], specificity was 85.4% CI [82.9%; 87.6%], and AUROC was 0.954 CI [0.944; 0.963]. Scope of application. The research results can be used to integrate the prototype expert system into the doctor’s automated workplace of a medical information system in a medical organization.