Identification of Diabetic Patients Using the Nonlinear Analysis of Short-Term Heart Rate Time Series

Abstract

The non-invasive method for identifying the volunteers suffering from the type 2 diabetes mellitus (T2DM) is suggested. The method is based on the symbolic analysis of short series (∼300 points) of RR-intervals of a single-lead electrocardiogram. To obtain the initial symbol sequences, 4 different methods of formation of symbols from the time series were used. Using the SVM classifier with a linear kernel, a selection of significant symbols was made. The most significant symbols became the input parameters for the SVM model with RB F kernel. The model has shown high efficiency: the sensitivity on the test sets was 70-82%, and the specificity was 73-77%. The proposed method uses the posterior probability, which has been accompanied by a class label for each new sample, being a criterion of the results reliability. For the obtained model, the threshold value of the posterior probability was 91%. It has been shown that the use of the posterior probability does not impair or improve the quality of the forecast. While using the posterior probability, the sensitivity of the model can increase up to 88% and specificity can increase up to 90%, being objective for up to 50% of all predicted values.