Cross-View Detection of Large Vessel Occlusion in Computed Tomography Angiography

Why It Matters

By leveraging complementary anatomical perspectives, the approach markedly improves automated LVO detection, accelerating treatment decisions and potentially reducing stroke‑related disability.

Key Takeaways

• Cross‑view method analyzes axial, coronal, sagittal CTA simultaneously
• Achieves 79.2% sensitivity at one false positive per case
• Uses KD‑tree to enforce spatial consistency across views
• Improves FROC AUC to 87% versus single‑view baselines
• Enhances rapid stroke diagnosis, supporting timely intervention

Pulse Analysis

Large vessel occlusion accounts for a substantial share of acute ischemic strokes, and every minute of delay in reperfusion therapy translates into lost brain tissue. Computed tomography angiography (CTA) is the frontline imaging modality because it rapidly visualizes intracranial arteries, yet interpreting the volumetric data demands expert radiologists and can be time‑consuming. Traditional automated tools have focused on single‑slice or single‑view analyses, leaving untapped the rich spatial relationships that span axial, coronal, and sagittal planes.

The cross‑view detection framework addresses this gap by running a slice‑based detector on each anatomical view independently, then mapping all candidate lesions into a unified coordinate system. A KD‑tree efficiently identifies spatial neighborhoods, allowing the algorithm to retain only those detections that receive corroboration from multiple views. This multi‑angle validation, followed by three‑dimensional non‑maximum suppression, yields a sensitivity of 79.2% at a single false positive per case and lifts the FROC‑AUC to 87%, outperforming conventional single‑view models.

Clinically, the technology promises to streamline stroke workflows by flagging probable LVOs before a radiologist’s final read, thereby shortening door‑to‑needle times for endovascular therapy. Integration into PACS or AI‑assist platforms could provide real‑time alerts, reduce diagnostic variability, and support hospitals lacking round‑the‑clock neuroradiology expertise. As AI adoption expands in neuroimaging, cross‑view consistency may become a standard design principle for robust, high‑stakes detection tasks, ultimately improving patient outcomes and lowering healthcare costs.