Automatic collateral circulation scoring in ischemic stroke using 4D CT angiography with low-rank and sparse matrix decomposition

• Published in: International journal for computer assisted radiology and surgery Vol. 15; no. 9; pp. 1501 - 1511
• Main Authors: Aktar, Mumu, Tampieri, Donatella, Rivaz, Hassan, Kersten-Oertel, Marta, Xiao, Yiming
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.09.2020; Springer Nature B.V
• Subjects: Aged; Algorithms; Angiography; Blood vessels; Brain Ischemia - diagnostic imaging; Cerebral Angiography; Clotting; Collateral Circulation; Computed Tomography Angiography; Computer Imaging; Computer Science; Contrast agents; Decision Making; Diagnosis, Computer-Assisted - methods; Female; Four-Dimensional Computed Tomography; Health Informatics; Healthy Volunteers; Humans; Image Processing, Computer-Assisted - methods; Imaging; Inspection; Ischemic Stroke - diagnostic imaging; Machine Learning; Male; Medical imaging; Medicine; Medicine & Public Health; Middle Aged; Optimization; Original Article; Pattern Recognition and Graphics; Radiology; Robustness; ROC Curve; Sparse matrices; Stroke; Surgery; Vision; Collateral supply; CT angiography; Ischemic stroke; Low-rank and sparse
• ISSN: 1861-6410; 1861-6429; 1861-6429
• DOI: 10.1007/s11548-020-02216-w

Purpose Sufficient collateral blood supply is crucial for favorable outcomes with endovascular treatment. The current practice of collateral scoring relies on visual inspection and thus can suffer from inter and intra-rater inconsistency. We present a robust and automatic method to score cerebral collateral blood supply to aid ischemic stroke treatment decision making. The developed method is based on 4D dynamic CT angiography (CTA) and the ASPECTS scoring protocol. Methods The proposed method, ACCESS ( A utomatic Collateral C irculation E valuation in i S chemic S troke), estimates a target patient’s unfilled cerebrovasculature in contrast-enhanced CTA using the lack of contrast agent due to clotting. To do so, the fast robust matrix completion algorithm with in-face extended Frank–Wolfe optimization is applied on a cohort of healthy subjects and a target patient, to model the patient’s unfilled vessels and the estimated full vasculature as sparse and low-rank components, respectively. The collateral score is computed as the ratio of the unfilled vessels to the full vasculature, mimicking existing clinical protocols. Results ACCESS was tested with 46 stroke patients and obtained an overall accuracy of 84.78%. The optimal threshold selection was evaluated using a receiver operating characteristics curve with the leave-one-out approach, and a mean area under the curve of 85.39% was obtained. Conclusion ACCESS automates collateral scoring to mitigate the shortcomings of the standard clinical practice. It is a robust approach, which resembles how radiologists score clinical scans, and can be used to help radiologists in clinical decisions of stroke treatment.