A point-event detection algorithm for the analysis of contrast bolus in fluoroscopic images of the coronary arteries

The paradigm of event detection is general enough to be used in a multitude of applications from various domains. In this contribution we describe an improved method to mark the appearance of blood mixed with contrast agent in a sequence of fluoroscopic images of the coronary arteries. This is needed for various computer-vision based technologies aimed at supporting the physician during a Percutaneous Transluminal Coronary Angioplasty (PTCA). PTCA is a surgical intervention conducted for the purpose of reopening blocked coronary arteries. We show how to extract a feature describing the amount of contrast agent present in each fluoroscopic image and how to establish a threshold over this feature, to separate the event of contrast-agent appearing from the normal case, when no contrast agent is present. For this purpose we estimate the likelihood of feature-values given the normal case and decide to mark the event for images whose feature has a very small likelihood. We test our algorithm on a number of sequences acquired in clinical routine.