SOLUTIONS SYNTHESIS METHOD TO THE CONDITION PREDICTION PROBLEM OF PATIENTS IN THE MEDICAL MONITORING SYSTEMS

Abstract

This article proposes a mathematical model and computational method for solutions synthesis to the condition prediction problem of patients in medical monitoring systems. Research describes construction method of diagnostic models for quality criteria assessment of a bio-medical system (BMS) elements condition using monitoring data. Authors show an informativeness (significance) estimation method of the systems diagnosing models variables, obtained based on artificial neural networks (ANN) theory. The article describes a forecasting method of multidimensional time series obtained based on the condition variables monitoring data of the dynamic systems. Authors present a method for solving the condition classification problem of the complex systems elements. The developed method for solutions synthesis to the prediction problem is implemented in the computer decision support system (CDSS) for diagnosing patients in medical monitoring systems. Application examples of the described method for solutions synthesis to the condition prediction problem of patients in medical institutions are presented.

The paper propose following developed innovations:

a)   Mathematical method for solutions synthesis to the condition prediction problem of patients based on trend-analysis concept. Unlike existing, this method takes into account the high dimensionality of the states space and informativeness changes of monitored variables, which depend on the patients’ condition.

b)   Method of informativeness (significance) variables estimation. It takes into account the measurement accuracy of the state variables and the presence of pair correlation between them, that allows to use this method for the analysis of the completeness of the models.

c)   Statistical method of the patients’ condition classification. It contains orthogonalization procedure and the dimension reduction procedure of the condition variables factor space. In contrast to the existing it uses as a closeness measure precedents in the principal components space and the ability to determine the disease stage, which is not recognized by modern bio-markers (for example, hormone-refractory state);

The classification problem of the patients condition was solved for the diagnosing result verification. The results of solving the classification problem of patients condition based on the monitoring data are obtained for the selected type of disease. It was found that with the using developed method and its implementation in CDSS the itself- recognizing probability of the class exceeds 75%.