Evaluation of the Automatic Segmentation of Aorta considering Aortic Diameter Measurements from Native MRI Angiographic Images

Motivation: The Dice coefficient used in evaluating automatic aorta segmentation doesn't directly reflect practical application accuracy. Goal(s): This study aims to validate the accuracy and robustness of automatic aorta segmentation by measuring maximum diameters at six specific locations in native MRI angiography images. Approach: The nnUNet segmented the aorta on native angiography images. We evaluated our method by the comparing maximum diameters at six aortic locations to those recorded in medical reports. Results: Among the six planes selected, the average absolute error for five planes is within 2.5 mm (about 2.5 voxels), and the r value for four planes exceeds 0.94. Impact: We validated an automatic method for the aortic segmentation from native angiography using the diameter measurements, considering the data from medical reports as the ground truth, instead of traditional metrics like the Dice coefficient or Hausdorff Distance.