Training a high accuracy model to visualize blood clots during mechanical thrombectomy for the treatment of Acute Ischemic Stroke

Background: Mechanical thrombectomy is the standard of care for Acute Ischemic Stroke caused by proximal large-vessel occlusion in the anterior circulation. In the stent retriever approach, a nitinol stent engages the clot via outward radial force to enable removal. However, current procedures lack direct clot visualization under fluoroscopy, which can reduce retrieval efficacy and often require multiple passes. Improving first-pass success is critical given the time-sensitive nature of stroke intervention. Methods: This study presents a clot visualization method using the spatial arrangement of radio-opaque markers on the Medtronic Solitaire™ stent. A deep learning model, Clot[U]-Net, based on the U-Net architecture, was trained on 800 anteroposterior and lateral in-vitro images and evaluated on a separate test set. Results: The Clot[U]-Net model achieved strong performance in clot boundary prediction, with a mean Intersection over Union (IOU) of 87.9% and an AUROC of 89.9%, and standard deviations of 2.2 and 3.16, respectively. Conclusion: The proposed method enables clot visualization during stent retriever thrombectomy without altering existing clinical workflows. With further pre-clinical and clinical validation, this approach may support real-time decision-making and improve procedural outcomes.