Automated extraction, labelling and analysis of the coronary vasculature from arteriograms

Abstract

For clinical decision-making and documentation purposes we have developed techniques to extract, label and analyze the coronary vasculature from arteriograms in an automated, quantitative manner. Advanced image processing techniques were applied to extract and analyze the vasculatures from non-subtracted arteriograms while artificial intelligence techniques were employed to assign anatomical labels. Lumen diameters of 11 phantom vessels were assessed with an accuracy of 0.27±0.19 mm (d _true = 0.45 + 0.92d _measured ; r> 0.99) and 0.21±0.15 mm (d _true =0.42+0.91d _measured ; r> 0.99), from cine and digital images, respectively. We collected a total of 15 routinely acquired cine-arteriograms showing 74 vessel segments with 18 stenoses (severity larger than 30% assessed quantitatively), and 53 digital arteriograms showing 236 vessel segments with 69 stenoses. From the cine arteriograms we extracted 64 (86%) of the vessel segments without manual correction and 196 (83%) from the digital arteriograms. Repeated analysis (3 times) of the arteriograms by the same operator resulted in a standard deviation of the mean segment diameters (precision) of 0.064 mm for the cine-images and 0.020 mm for the digital images, while the standard deviations in the measurement of the minimal luminal diameter of the observed stenoses were 0.020 mm and 0.019 mm, respectively. The LAD artery, the septal and diagonal branches were correctly identified automatically in 86% of the segments. From these evaluations we conclude that our automated approach provides reliable tools for the assessment of multi-vessel disease, both in an offand on-line environment.