Semi-automated computer assessment of the degree of carotid artery stenosis compares favorably to visual evaluation

Abstract

To validate a semi-automated computer approach for the assessment of the degree of carotid artery luminal narrowing by comparing it to the visual evaluation by a neuroradiologist. In a retrospective cross-sectional study, consecutive emergency department patients who underwent computed tomography angiography (CTA) of the carotid arteries were identified. CTA studies were reviewed by a neuroradiologist, and also independently processed with a computer algorithm that automatically measures the degree of luminal narrowing at the level of the internal carotid artery bulb. The findings of the neuroradiologist and computer assessment were compared using Chi2 tests/kappa calculations and linear regression for categorical and continuous measurements of carotid stenosis, respectively. The study population consisted of 125 patients (74 no stroke/TIA, 18TIA, and 33 stroke). 201 carotid arteries showed no significant stenosis; 33 showed > or =70% stenosis, 5 showed 95-99% stenosis, and 11 showed complete occlusion. There was excellent agreement between the neuroradiologist's visual assessment and the automated computer evaluation of the category of carotid stenosis (kappa=0.918, p<0.001). The automated computer algorithm for quantifying the degree of carotid stenosis is reliable and shows high concordance with the interpretation of an experienced neuroradiologist.