MODEL-BASED JUNCTION DETECTION ALGORITHM WITH APPLICATIONS TO LUNG NODULE DETECTION

Among the many features used for classification in computer-aided detection (CAD) systems targeting pulmonary nodules, those based on differences between the shapes of nodules and vessels are most common. However, the explicit modeling of vessel junctions, often reported as the main source of false positive detections in CAD algorithms, has been largely neglected in the literature. We introduce a parametric junction model that captures the shape aspects of vessel junctions. Model parameters are expressed through probability distributions that encode medical knowledge on pulmonary vasculature. The usefulness of the model is demonstrated through its impact on the performance of a lung CAD algorithm