Exploring Feature-Based and Data-Driven Techniques for Quality Control in Brain Vessel Segmentation

Brain vessel segmentation plays a central role in many clinical and research applications, including the diagnosis and monitoring of cerebrovascular diseases, the planning of surgical or endovascular interventions, and the quantitative study of vascular anatomy and pathology. Accurate segmentation of the vascular network is therefore of great importance, yet it remains a particularly challenging task. The thin, tortuous, and highly variable morphology of vessels makes their delineation difficult, and even small errors can have significant clinical consequences. While deep learning and advanced image processing techniques have achieved promising results, the automated evaluation of segmentation quality is still underdeveloped. Reliable quality control (QC) methods are essential to ensure the trustworthiness and applicability of vessel segmentation in both research and clinical practice. This work addresses this need from two complementary perspectives. First, it investigates whether morphological and topological descriptors of brain vessel segmentations can serve as informative features for QC. To this end, it introduces a flexible feature extraction tool capable of generating structured descriptors of vascular geometry and connectivity. Beyond its use in regression-based QC experiments, this tool is designed as a standalone resource that can be readily integrated into diverse medical and research scenarios, serving both quality control and segmentation studies. Second, it provides a broader review of learning-based strategies for QC that could be adapted or developed specifically for vascular imaging. In particular, it discusses uncertainty-based and reconstruction-based methods, highlighting their potential and limitations for brain vessel segmentation. By combining methodological development with a structured overview of QC research directions, this work aims to lay the foundation for practical and robust quality control in brain vessel segmentation, ultimately supporting more reliable downstream analyses and clinical decision-making.