Anthony Luo, SEAS '23, Computer Science, Columbia University
Supervising Faculty, Sponsor, and Location of Research
Professor Andrew F. Laine, Summer at SEAS, Heffner Biomedical Imaging Lab, Columbia University
An accurate and automatic system for artery and vein classification of pulmonary vessels has the potential to drastically increase efficiency of medical diagnoses and analyses that require artery and vein labeling of pulmonary vessels. In particular, labeled arteries and vessels facilitate site isolation for vascular pathology, and assist inferences of causes of pulmonary vascular dysfunction. We present a software pipeline to perform automatic artery and vein classification up to a user specified diameter without the need for contrast enhanced CT or manual seed point initialization. The performance of this software pipeline is significantly faster than manual labeling by an expert analyst and provides additional flexibility through vessel diameter and branch length filtering options. Our pipeline uses an expert system based approach to classification by leveraging the closeness and co-orientation of pulmonary arteries and bronchi. Compared to prior art, our system introduces a novel combination of straight-line and per-voxel co-orientation and co-distance calculations. We plan to compare our pipeline against a semi-automated region growing approach and manually annotated ground truth.
pulmonary vessels, artery vein classification, airways