MISSION OF THE CHALLENGE¶
The challenge addresses three major aspects of the inspection of angiographic cranial images, namely
- Detection of pathological changes of the vessel tree in the form of aneurysms
- The quantitative assessment of these aneurysms through segmentation as the basis for diagnosis and monitoring
- Image-based estimation of the stroke risk
These solutions could improve clinical workflows towards aneurysm screening in cranial angiographies, which might be acquired for a wide range of diagnostic questions.
Furthermore, they could improve the analysis of datasets acquired for patients with suspected subarachnoid hemorrhage (SAH) or known aneurysms, which are monitored.
The shared dataset represents a cohort of 115 patients who underwent rotational X-ray angiography because of suspected SAH.
Datasets have been acquired with a fixed C-Arm, which allows for the acquisition of CT-like 3D image volumes during diagnostic or interventional catheterization. A contrast agent was applied in the supplying artery of the examined vascular region (e.g. Vertebral artery). Image volumes show the arterial intracranial vasculature of the corresponding part of the brain.
In the training dataset, we store the positions of the aneurysm centers together with the image data. Furthermore, the segmented aneurysm heads are provided as stl- and mask-files. For each aneurysm, the rupture state is provided.
In correspondence to the aspects introduced above, we will have three analysis goals:
- Task 1: Automatically find the position and approximate bounding box for all pathological vessel dilations (aneurysms) in a dataset to ensure that these structures, which are associated with stroke risk, are not missed during the inspection of a dataset
- Task 2: Provide accurate segmentation masks for aneurysms (automatically or semi-automatically) in order to support the quantitative assessment for diagnosis, monitoring and therapy planning
- Task 3: Classify aneurysms according to their rupture risk (using machine learning, computational geometry, CFD, …) to support decision making for treatment planning
