Estimation of common carotid intima-media thickening using neural networks in adults with and without ischemic stroke

Abstract

Introduction: carotid intima-media thickening indicates potential atherosclerosis and risk of ischemic stroke. The objective was to predict the presence of carotid intima media thickening using neural networks in adults with and without stroke. Materials and methods: analytical and cross-sectional study of a secondary database of 600 patients with and without a history of ischemic stroke. The dependent variable was the intimamedia thickness of the right and left common carotid artery (RCC and LCC). Biochemical markers frequently used in primary care, systolic and diastolic blood pressure, were used. Multilayer perceptron-type neural networks with area under the curve (AUC) were used. Results: without a history of stroke, the perceptron predictive model for RCC was good (AUC=0.852). For LCC, it was acceptable (AUC=0.799). In patients with a history of ischemic stroke, the predictive model for RCC was good (AUC=0.826). The model for LCC was acceptable (AUC=0.789). In the absence of stroke, the neural network test had a percentage of correct predictions for right and left common carotid intima-media thickening of 81.30% and 79.20%, respectively. With a history of ischemic stroke, it was 82.80% and 91.50%, respectively. Conclusions: the multilayer perceptron-type neural network model, based on tests performed in primary care, had a high capacity to correctly predict intima media thickening of the common carotid artery in patients without a history of ischemic stroke.