Towards the Automatic Segmentation, Modeling and Meshing of the Aortic Vessel Tree from Multicenter Acquisitions

The aorta is the main artery of the human body and, with its branch arteries, it forms the aortic vessel tree (AVT) [1] and supplies the whole body with blood. Aortic diseases, like aneurysms and dissections, can lead to severe consequences if left untreated. Their treatment with open surgery is of high risk and therefore delayed with constant monitoring and drug treatments whenever possible. Yet, this requires a regular screening of the vessels for disease development [2]. The standard image modality for clinical assessment is computed tomography angiography (CTA), which provides a detailed view of the AVT. Optimally, the whole AVT geometry is reconstructed and compared with the geometry of the subsequent CTA scans. Not only to detect changes related to the pathology, but also to detect peripheral changes, either resulting from the primary pathology or new comorbidities.

However, manual execution of this task requires a slice-by-slice contouring, which can require up to a whole day for the aortic vessel tree of one scan, making this task unfeasible in clinical practice. Furthermore, an accurate reconstruction can be used to analyze the blood flow and the outcome of endovascular surgeries by means of numerical simulations. For this, AI-supported automatic segmentation methods have shown to be a possible solution, which can potentially run in real time or in the background of the clinical routine. An open problem is the translation of these algorithms to 1) work in several clinical institutions, because of different scanning protocols, especially with regards to scanning device, radiation dose and contrast agent, which lead to varying Hounsfield values in the AVT, and 2) rely on a limited amount of labelled data given the long annotation time. This challenge comprehends one main task and two optional subtasks. In the main task, we target the problem of vessel tree segmentation before the diagnosis of an aortic pathology. We provide the challenge participants with a training set of AVTs and corresponding manual segmentations from three institutions. Participants are expected to design algorithms for an automatic AVT segmentation based on this training set. All the proposed methods will be evaluated based on a hidden test set from a fourth institution using Dice Similarity Score (DSC) and Hausdorff Distance (HD). The evaluation will consider the the variance and the sensitivity of the evaluation metrics (DSC and HD) generated by different CTA variabilities, such as intensities, rotations, translations, noise and artificial motion artifacts. For this, the ranking will also consider the quantitative sensitivity indices [7] to define how the proposed method copes with large input variation. B) The reconstruction of the AVTs needs to be ideally artifact free for visualization and blood flow simulation. In the first optional subtask, the reconstructed AVT surface geometries will be qualitatively evaluated by clinical specialists. The qualitative evaluation will focus on the number of branching arteries and on the visual quality of the produced results. In the second subtask, we expect the AVT reconstruction to be in the form of a volume mesh. This will be quantitatively evaluated in terms of mesh validity for applications of computational fluid dynamics and ranked using the scaled Jacobian and the number of mesh elements.

[1] Jin, Yuan, et al. "Ai-based aortic vessel tree segmentation for cardiovascular diseases treatment: status quo." arXiv preprint arXiv:2108.02998 (2021).
[2] Pepe, Antonio, et al. "Detection, segmentation, simulation and visualization of aortic dissections: A review." Medical image analysis 65 (2020): 101773.
[3] Radl, Lukas, et al. "AVT: Multicenter aortic vessel tree CTA dataset collection with ground truth segmentation masks." Data in Brief 40 (2022): 107801.
[4] Maier-Hein, Lena, et al. "Why rankings of biomedical image analysis competitions should be interpreted with care." Nat. Commun. 9(1), (2018): 5217.
[5] Reinke, Annika, et al. "How to exploit weaknesses in biomedical challenge design and organization." Int. Conf. Med. Image Comput. Comput. Assis. Interv., Springer, Cham (2018).
[6] Maier-Hein, Lena, et al. "BIAS: Transparent reporting of biomedical image analysis challenges." Medical Image Analysis, (2020): 101796.
[7] Iooss, B., Lemaître, P.. A Review on Global Sensitivity Analysis Methods. In: Dellino, G., Meloni, C. (eds) Uncertainty Management in Simulation-Optimization of Complex Systems. Operations Research/Computer Science Interfaces Series, vol 59. Springer, Boston, MA (2015). https://doi.org/10.1007/978-1-4899-7547-8_5
[8] Saltelli, Andrea, et al. "Global sensitivity analysis: the primer." John Wiley & Sons, 2008.
[9] Oeltze, Steffen, and Bernhard Preim. "Visualization of vasculature with convolution surfaces: method, validation and evaluation." IEEE Transactions on Medical Imaging 24.4 (2005): 540-548.
[10] Tarjuelo-Gutierrez, Jaime, et al. "High-quality conforming hexahedral meshes of patient-specific abdominal aortic aneurysms including their intraluminal thrombi." Medical & biological engineering & computing 52.2 (2014): 159-168.