Towards adequate prediction of prediabetes using spatiotemporal ECG and EEG feature analysis and weight-based multi-model approach

Abstract

Prediabetes is a metabolic condition before the occurrence of diabetes. The diagnosis of prediabetes can slow down or eliminate the growing cases of diabetes around the world. This paper presents a novel approach to identifying some vital physiological features for prediabetes prediction to stem the growing trend of type-2 diabetes. A standard OGTT experiment was conducted using BIOPAC 150MP, g-SAHARA and Mindray physiological device to capture continuous electrocardiogram (ECG) rhythm and electroencephalogram (EEG) of 40 human subjects while measuring blood glucose value at a regular interval. Features from the captured physiological signals were analyzed using an integrated space–time principal component analysis, independent component analysis, least absolute shrinkage and selector operator, and piecewise aggregate approximation techniques. The results from feature analysis show that certain features, namely HRV, QT, and ST from ECG; alpha, beta, and theta from the right parental hemisphere, along with alpha and delta from the left occipital hemisphere from EEG show significant correlation with change in the blood glucose. Furthermore, a weight-based multi-model was proposed by combining five (5) classification methods. The selected ECG and EEG features were applied for training the proposed multi-model classification, which is used to predict prediabetes. The evaluation of the multi-model performance produced accuracy, precision, and F1-measure of 92.0%, 88.8%, and 82.7% respectively, which is higher than the individual methods. The experimental results show that the coupling of multi-model electrophysiological data acquired with wearable multi-sensor devices can be utilized to diagnose diabetes early.