An absolute magnitude deviation of HRV for the prediction of prediabetes with combined artificial neural network and regression tree methods

摘要

Early diagnosis of prediabetes is an effective solution to the rising cases of diabetes around the world. The heterogeneous physiological characteristics of the ECG signal recorded from the heart make it challenging to implement an efficient diagnostic system. Therefore, this paper proposes a new approach to handling the heterogeneous characteristics of heart rate variability (HRV) with an absolute magnitude deviation analysis and an integrated machine learning technique for prediabetes prediction. We conducted an oral glucose tolerance test to acquire a resting-state ECG signal and the corresponding blood glucose value. We analyzed the HRV pattern from the ECG signal with a block-sliding window technique. We proposed a hybrid model to classify normal and prediabetes based on the extent of the absolute deviation of HRV values and avoiding a single point of failure. We adopted the model from the classification and regression tree (CART) and neural network (NN) algorithms. The experimental results reveal that when the blood glucose level increases, the maximum and range values of CARTHRV decreases while the minimum value increases. The proposed hybrid model had a better performance than the two methods with 100% sensitivity, specificity, and F1-score measures against CART and NN that recorded < 100% for the same number of prediabetes in the training and test sets. The outcome from the analysis shows that the changes in blood glucose can be observed in ECG signals. The fast approximation of the proposed method to 100% accuracy suggests that it is possible to achieve the diagnosis of prediabetes and overcome the discrepancies in physiological signals among individuals.