Automatic segmentation of true lumen and false lumen of aortic dissection based on a two-stage learning framework

Accurate segmentation for true lumen (TL) and false lumen (FL) is essential for the treatment of aortic dissection (AD) disease. However, the complex morphology of AD raised challenges for AD segmentation, especially for precise TL and FL separation. In this study, a novel two-stage algorithm based on the mesh model was developed to achieve the segmentation of TL and FL. In detail, a MeshSegNet was firstly applied to complete the coarse segmentation of AD and then fast geodesic curvature flow (FGCF) was used to optimize the segmentation result. A total of 150 type B AD cases were included to train and evaluate the effect of the proposed segmentation algorithm. The results showed that FGCF could significantly enhance the segmentation precision with the maximum increasing dice score of 0.18 compared to that with MeshSegNet method only. The average accuracy of all included testing cases was more than 93%. This two-stage segmentation algorithm could achieve precise and efficient segmentation of TL and FL, thus might assist in the treatment of AD in the future.