Artificial intelligence for cerebrovascular digital subtraction angiography: methods, evidence, and challenges.
Authors
Affiliations (3)
Affiliations (3)
- University of Science and Technology Beijing, Haidian Street, Beijing, Beijing, 100083, China.
- University of Science and Technology Beijing, 30 Xueyuan Rd, Haidian District, Beijing, Beijing, 100083, China.
- University of Science and Technology Beijing, HaiDian, Beijing, 100083, China.
Abstract
Cerebrovascular diseases remain a major cause of mortality and long-term disability worldwide. Digital subtraction angiography (DSA) provides high-spatiotemporal-resolution visualization of cerebrovascular anatomy and contrast-agent dynamics and plays an important role in diagnosis, interventional treatment planning, and treatment assessment. Artificial intelligence (AI) has increasingly been applied to cerebrovascular DSA, yet the available evidence is distributed across diverse imaging settings, technical tasks, datasets, and validation strategies. This structured topical review provides a DSA-centered overview of AI-assisted cerebrovascular image analysis across 2D DSA, three-dimensional rotational angiography, and four-dimensional DSA. Literature from Web of Science Core Collection and major engineering databases was screened at title, abstract, and full-text levels using predefined eligibility criteria. The literature is organized around six interconnected tasks: image preprocessing, vessel and lesion segmentation, image registration, reconstruction and generation, disease diagnosis and detection, and hemodynamic analysis. Representative conventional and learning-based methods are reviewed together with their datasets, evaluation metrics, validation designs, and available data or code resources, with particular attention to the strength and limitations of the supporting evidence. Recent studies demonstrate increasing use of spatiotemporal modeling, deep learning, and image-generation techniques, but clinical translation remains constrained by small or single-center datasets, heterogeneous acquisition protocols and evaluation practices, limited external and prospective validation, model interpretability, and deployment requirements. Future work should emphasize acquisition-aware and physics-informed modeling, multicenter and privacy-preserving collaboration, clinically meaningful validation, and closer integration with neurointerventional workflows. By linking methodological developments with their supporting evidence and translational challenges, this review aims to support the development of more robust and clinically applicable AI methods for cerebrovascular DSA.