Deep Brain Vessel Profiler

In the project, methods were developed to automatically detect and analyze the most important brain vessels in 3D imaging such as CT and MR angiography. The aim was to visualize differences in vascular anatomy between patients and to investigate their significance for diseases such as strokes or aneurysms.

The size, shape and connection of the most important cerebral vessels – in particular the so-called Circle of Willis – differ from person to person. For example, the arterial ring is not fully developed in most people. Whether and how these differences influence disease progression and treatment has hardly been systematically investigated to date. Our method now enables an automated, scalable and comparable analysis of such anatomical variants for the first time – virtually at the touch of a button.

Technically, our approach is based on machine learning methods, in particular convolutional neural networks (CNNs). For this purpose, a high-quality image data set was created from CT and MR angiographies and annotated together with specialists from the University Hospital Zurich. A novel virtual reality approach was used to manually mark the vessels. We actively involved the international research community in two international data challenges (TopCoW 2023 and 2024). Particular attention was paid to the correct reconstruction of the vessel structure and its geometric description.

Overview of the results:

• Development of a VR-based method for vessel annotation
• Publication of a bimodal, clinically tested image data set (CTA+MRA, approx. 200 cases)
• Comparison and evaluation of various reconstruction methods (proposed by the community)
• Own methods for segmentation, reconstruction and shape description of the main arteries
• Initial analyses with stroke data – further studies are being planned

The tools developed can be used in a variety of ways: for clinical questions or as a basis for computer-aided decision support and biomechanical models.