Reduction of failure rates in automated analysis of difficult images: improved simultaneous detection of left and right coronary borders

A new model-based simultaneous coronary border detection method is reported that uses 3-D graph searching principles and detects borders that are optimal when taken as a pair. The method significantly reduces the failure rate for difficult images while substantially enhancing computational efficiency. It was compared to the conventional border detection method, which has been shown to be accurate for uncomplicated images. The robustness of each method was assessed for 43 difficult images in which conventional analysis was likely to fail. Minimum lumen diameters from the conventional and simultaneous detection methods were highly correlated for uncomplicated images. For difficult images, simultaneous border detection reduced the analysis failure rate from 32/43 to 11/43