Automatic stenosis detection and quantification in renal arteriography.

Abstract

Visual assessment of the degree of renal artery stenosis on renal arteriography has a large inter- and intraobserver variability. This degree is usually estimated by the ratio between the most narrowed portion of the artery and the reference diameter. The latter is a priori unknown information and thus operator dependent. The objective of the present work was to test the performances of a computer system that was designed to analyze and quantify lesions on 2D renal arteriograms. The main hypothesis was to consider that the most frequent diameter computed along the artery was a good candidate to approximate the reference diameter. Forty nine patient images were collected from the EMMA randomized trial, a multicenter study comparing two treatment strategies in unilateral atheromatous renal artery stenosis of at least 60%. For each image, the degree of stenosis was evaluated by five independent experts and the mean value was used to represent the gold standard for the computer system. The system is based on a fuzzy automaton and performs a syntactic analysis of the arterial segment providing automatic and reproducible quantification of lesions. Both the radiologist caring for the patient and the system were compared to the gold standard. Compared to individual radiologists, the computer system gave a more precise estimation of percent stenosis and did not over or under estimate the severity of the lesion.