Evaluating nnU-Net for Type B Aortic Dissection segmentation on CTA images

Accurate segmentation of Type B Aortic Dissection (TBAD) is crucial for clinical diagnosis and treatment planning. In this study, we trained nnU-Net on the ImageTBAD dataset for TBAD segmentation, achieving Dice scores of 0.94, 0.90, and 0.42 for true lumen (TL), false lumen (FL), and false lumen thrombus (FLT), respectively, surpassing the baseline methods by 0.08, 0.12, and 0.13. We identified challenges in segmenting small-volume FLT and proposed potential improvements using residual skip connections. The generalization capability of nnU-Net was validated on the external AVT dataset, where the Dice scores for TL and FL exceeded 0.9, and FLT achieved a Dice score above 0.86. nnU-Net demonstrated its efficacy in TBAD segmentation and holds promise for advancing segmentation techniques in this field.