Nodule-aware deep learning model improves pulmonary vessel segmentation on chest CT

A multi-center study found that the NHPVS model for pulmonary vessel segmentation on chest CT achieved a Dice score of 89.2%, improving vessel length by 31.1% and branch count by 60.5% vs nnUNet-v2, with a nodule misclassification rate of 1.8%.

• NHPVS was externally validated and compared with state-of-the-art nnUNet-v2.
• The volume of segmented small pulmonary vessels (diameter < 3 mm) was significantly greater with NHPVS (p < 0.001).

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.