Prof. Shanjun Zhang of Kanagawa University, Japan

Cerebral artery diseases have always been an important topic in medical research due to their direct association with stroke and other life-threatening conditions. In recent years, significant attention has been devoted to vessel segmentation and completion, particularly for the arteries surrounding the circle of Willis (CoW), a critical structure for cerebral blood circulation. Accurate reconstruction of these vessels is essential for understanding cerebrovascular anatomy, supporting surgical planning, and improving the diagnosis of vascular diseases. However, most existing computational approaches rely on computed tomography angiography (CTA) data, which requires the use of contrast agents. These agents can pose risks such as allergic reactions, kidney burden, or contraindications for vulnerable patients, limiting the clinical applicability of CTA-based methods. Furthermore, current vessel completion techniques often suffer from topological inconsistencies, such as incomplete reconstructions or erroneous connections, which can compromise downstream analysis and clinical reliability. To address these limitations, we propose a method that leverages non-contrast magnetic resonance angiography (MRA), which is both safer and non-invasive, to achieve structure-aware completion of missing vessel segments. We curated a dataset consisting of MRA scans from 125 patients, with manual annotations of 13 vessel segments that constitute the circle of Willis. This detailed annotation enables supervised learning with fine-grained vessel-level supervision. Our approach integrates PointNet++ for local geometric feature extraction with Graph Convolutional Networks (GCNs) for modeling the global connectivity and topological relationships among vessels. By jointly capturing both geometry and topology, the model is designed to learn not only the appearance of vessels but also the structural rules governing their connections. Extensive experiments demonstrate that the proposed framework can effectively reconstruct missing vessel structures, achieving higher completeness and reducing topological errors compared to existing baselines. In particular, the method preserves the integrity of the circle of Willis, ensuring more faithful modeling of cerebral vasculature. These results suggest that our approach provides a promising alternative to contrast dependent CTA-based techniques, with the potential to enhance cerebrovascular analysis in both research and clinical practice.

Biography:

He received his B.S. degree in Computer Software Engineering from Huazhong University of Science and Technology in 1986, and his M.S. and Ph.D. degrees in Information Engineering from Hokkaido University in 1991 and 1994, respectively. In 1994, he became an Assistant Professor in the Department of Information Engineering at Muroran Institute of Technology. He later joined Kanagawa University, where he was appointed Associate Professor in the Department of Information Science, Faculty of Science, in 2002, and promoted to Full Professor in 2010. From 2005 to 2006, he was a visiting scholar at the Center for Neural Systems, Boston University. He is currently a Professor and the Director of the Department of Computer Science, Faculty of Informatics, Kanagawa University. His research interests include computer vision, medical image processing, and virtual reality.

Address:China