Pulmonary Vascular Tree Segmentation from Contrast-Enhanced CT Images
This work addresses pulmonary hypertension detection in patients, but it is incremental as it builds on existing segmentation methods and shows midfield performance on a benchmark dataset.
The authors tackled pulmonary vessel segmentation from CT images to derive clinical parameters for detecting pulmonary hypertension, achieving results on a 24-patient dataset that distinguish patients with and without PH.
We present a pulmonary vessel segmentation algorithm, which is fast, fully automatic and robust. It uses a coarse segmentation of the airway tree and a left and right lung labeled volume to restrict a vessel enhancement filter, based on an offset medialness function, to the lungs. We show the application of our algorithm on contrast-enhanced CT images, where we derive a clinical parameter to detect pulmonary hypertension (PH) in patients. Results on a dataset of 24 patients show that quantitative indices derived from the segmentation are applicable to distinguish patients with and without PH. Further work-in-progress results are shown on the VESSEL12 challenge dataset, which is composed of non-contrast-enhanced scans, where we range in the midfield of participating contestants.