Automated Identification of Thrombectomy Amenable Vessel Occlusion on Computed Tomography Angiography using Deep Learning

Abstract Objectives Recent advancements have extended the treatment window for large vessel occlusion in acute ischemic stroke, prompting a shift in the standard of care for patients presenting within 6 to 24 hours. We developed and externally validated an automated deep learning algorithm for detecting thrombectomy amenable vessel occlusion (TAVO) in computed tomography angiography (CTA). Methods The algorithm was trained on 2,045 acute ischemic stroke patients who underwent CTA, and validation was conducted using two external datasets comprising 64 (external 1) and 313 (external 2) patients with ischemic stroke. TAVO was defined as occlusion in the intracranial internal carotid artery (ICA), or M1/M2 segment of the middle cerebral artery (MCA). Utilizing U-Net for vessel segmentation and EfficientNetV2 for TAVO prediction, the algorithm’s diagnostic performance was assessed using the area under the receiver operating characteristics curve (AUC), sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Results The mean age in the training and validation dataset was 68.7±12.6; 56.3% were men, and 18.0% had TAVO. The algorithm achieved AUC of 0.950 (95% CI, 0.915–0.971) in the internal test. For the external datasets 1 and 2, the AUCs were 0.970 (0.897–0.997) and 0.971 (0.924–0.990), respectively. Notably, the algorithm demonstrated robust sensitivity and specificity (approximately 0.95) for intracranial ICA or M1-MCA occlusion, but a slight reduction in performance for isolated M2-MCA occlusion. Conclusion This validated algorithm has potential applications in identifying TAVO and could aid less-experienced clinicians, potentially expediting the treatment process for eligible patients.