Does Deep Learning Vascular Segmentation on CTA Improve Vertebral Artery Dissection Detection?

The study investigated whether a deep learning model could improve the detection of vertebral artery dissection on CT angiography (CTA). The research team trained an nnU-Net model on 101 CTA images and tested it on 40 CTA images. The model achieved high accuracy in vascular segmentation with a Dice coefficient of 0.96. However, in a clinical study with 10 readers, diagnostic accuracy with the model was lower than without it (72% versus 84%, p < 0.001). Interpretation time and reader confidence did not change significantly. Seven of ten readers would use the tool in acute settings. The authors concluded that although the model accurately segments vessels, further validation is needed to confirm its clinical utility.