其他名称：
25-21-00165
项目负责人：
Geltser Boris
发表日期：
2025
主持机构：
FEDERAL STATE BUDGETARY EDUCATIONAL INSTITUTION OF HIGHER EDUCATION "VLADIVOSTOK STATE UNIVERSITY",
国家：
俄罗斯
开始日期：
2025
结束日期：
2026
简介：
Medical information systems store large volumes of medical data, including information about patient complaints, anamnesis of their diseases, objective, laboratory and instrumental studies, etc. indicators. This information contains hidden knowledge about the relationships between parameters characterizing the clinical and functional status of the patient and disease outcomes. Key elements of prognostic research in clinical medicine are the identification of risk factors, adverse events and the development of prognostic models. Machine learning (ML) methods are increasingly being used to solve this problem, providing acceptable predictive accuracy, but we have limited ability to explain and clinically evaluate its results. The lack of transparency of ML models is an obstacle to their widespread development in clinical practice. At the same time, the most accessible form of explanation adopted in the predictive decision model is a set of conditions - rules that are understandable to doctors and generally accepted principles of clinical practice. To generate such rules, use decision tree algorithms. At the same time, this method has the disadvantage of low accuracy of forecasts and, therefore, an incorrect set of rules. Thus, the scientific problem that the project aims to solve is to develop methods for generating correct rules extracted from trees or ensembles of trees and explanatory conclusions that generate predictive ML models.
Scientifically innovative research leads to the development of new methods:
- a new method for generating decisions based on calculated optimal threshold values (OptimumDT), which allows you to develop a predictive model with quality metrics;
-a new method for generating a random forest based on the method for generating tree-like solutions OptimumDT (OptimumRF);
-a new method of bringing a pool of decision rules for observation, providing transparency in assessing the cause-and-effect relationships of factor analysis with endpoints, allows practitioners to make effective decisions (SimpleRules).
The practical innovation of the work lies in the fact that development methods make it possible to provide formal decision-making procedures and, as part of the implementation of this project, to implement a software component (microservice) in a medical decision support system.
专业领域：
信息技术
语种：
英语
