Citation

Rahman, Kazi Ashikur and Ali, Nur Hasanah and Muda, Ahmad Sobri (2026) Deep Learning-Based Assessment of Collateral Circulation in Ischemic Stroke Using Intra-Procedural Cone-Beam CT. IEEE Access, 14. pp. 139208-139231. ISSN 2169-3536

Text 12.pdf - Published Version Restricted to Repository staff only Download (4MB)

Abstract

Ischemic stroke caused by large-vessel occlusion can result in rapid and irreversible brain injury if cerebral reperfusion is delayed. During this critical period, collateral circulation plays a decisive role in sustaining residual blood flow, limiting infarct expansion, and guiding treatment selection for endovascular thrombectomy (EVT). Despite its clinical importance, collateral circulation is still commonly assessed through manual visual grading, which is time-consuming and subject to considerable inter-observer variability. Consequently, an accurate, objective, and real-time assessment method remains a major unmet need in hyperacute stroke care. In this study, a deep learning-based framework was proposed for tri-class collateral circulation classification (Good, Moderate, and Poor) using intra-procedural cone-beam computed tomography (CBCT). Unlike previous approaches that primarily rely on computed tomography angiography (CTA) or binary grading schemes, the proposed method leverages CBCT acquired directly in the angiography suite, enabling workflow-integrated collateral assessment during EVT. Two residual convolutional neural network architectures, ResNet-50 and ResNet-18, were trained using a patient-wise two-phase strategy on a CBCT dataset comprising 45 ischemic stroke patients and 22,861 DICOM slices. Experimental results demonstrate that ResNet-50 outperforms ResNet-18, achieving an overall accuracy of 92% and a macroF1 score of 0.91, compared with 88.8% accuracy and a macro-F1 score of 0.87 for ResNet-18. The deeper architecture enables more effective learning of complex vascular morphology and distal perfusion patterns, while attention-based visualization confirms a consistent focus on clinically relevant collateral territories. Overall, the proposed framework provides a robust and clinically meaningful solution for real-time collateral circulation assessment in hyperacute ischemic stroke.

Item Type:	Article
Uncontrolled Keywords:	Deep learning, medical image analysis
Subjects:	R Medicine > R Medicine (General) > R856-857 Biomedical engineering. Electronics. Instrumentation
Divisions:	Faculty of Engineering and Technology (FET)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	02 Oct 2026 00:42
Last Modified:	02 Oct 2026 00:42
URII:	http://shdl.mmu.edu.my/id/eprint/16810

Downloads

Edit (login required)