Fuzzy pulmonary vessel segmentation using optimized vessel enhancement filtering

Vessel segmentation within pulmonary images serves as a basis for a variety of applications, including PE detection and visualization, lung nodule detection, assistance in bronchoscopic navigation, lobe segmentation, and surgical planning. Although applications have different segmentation requirements, speed and accuracy is a clear benefit. A new approach combining a single parameter vessel enhancement filter and fuzzy connectedness is presented. The advantages of vessel filtering are brought to bear with a minimal impact on time by limiting the scales. Vesselness and intensity features are combined within a fuzzy segmentation framework, reducing the number of required scales and avoiding some of the drawbacks of each feature alone. Validation was performed on five datasets and Dice Similarity Coefficients (DSC) demonstrate an improvement of 9% (from 81% to 90%) on average for small vessels without influencing the accuracy for large vessels (95%) compared to an intensity-based method alone.