Prediction of Arterial Stiffness Risk in Diabetes Patients through Pulse Wave Velocity and Deep Learning Techniques

Abstract

Diabetes and arterial stiffness are the primary health concerns related to each other. The understanding of both factors provides efficient disease prevention and avoidance. For the development of cardiovascular disease, arterial stiffness and Diabetes are pathological process considerations. The existing researchers reported the association of these two factors and the complications of arterial stiffness with Diabetes are still in research. Arterial stiffness is measured through pulse wave velocity (PWV), which influences cardiovascular disease in diabetic patients. Moreover, this study developed a medical prediction model for arterial stiffness through the machine and deep learning models to predict the patients who are high-risk factors. Brachial–ankle pulse wave velocity (baPWV) and fasting blood glucose (FBG) are the consideration of baseline. Gaussian-Least absolute shrinkage and selection operator (LASSO) with whale optimization is proposed for feature selection. Initially, key features are extracted from the wave measurement using LASSO, and Principal component analysis (PCA) has been used to remove the outliers. Second, Gaussian regression chooses the PWV-based relevant features from the LASSO identified features. The parts are the critical points to increasing the accuracy of the prediction model. Hence, the selected features are further improved with an evolutionary algorithm called the cat optimization approach. Third, the prediction model is constructed using three machine and deep learning algorithms such as a Support vector machine (SVM), a convolution neural network (CNN), and Gated Recurrent Unit (GRU). The performance of these methods is compared through the area under the receiver operating characteristic curve metric in the dataset. The model with the best performance was selected and validated in an independent discovery dataset (n = 912) from the Dryad Digital Repository (https://doi.org/10.5061/dryad.m484p). From the experimental evaluation, LSTM performs better than other algorithms in classifying arterial stiffness with the AUROC of 0.985 and AUPRC of 0.976.