Misclassification Loss for Segmentation of the Aortic Vessel Tree

Our Contributions

• A novel loss function designed to improve segmentation accuracy by reducing false positives and rescuing false negatives in complex structures like the aortic vessel tree.
• The paper introduces a differentiable XOR operation to identify misclassifications, enhancing the precision of segmenting fine anatomical structures.
• Top Performance in SEG.A Challenge 2023: The method ranked among the top-six in the validation phase of the SEG.A 2023 challenge, achieving a Dice score of 0.93 and a Hausdorff Distance (HD) of 3.50 mm.

  Paper Available Here