A collaborative paper will be presented a part of the Electronic Poster session. Yueh Lee, of the University of North Carolina at Chapel Hill Dept of Radiology, will present the paper ‘Automated Vessel Extraction for Assessment of Leptomeningeal Collateral Status.’ The paper was authored by Dr. Lee, Stephen Aylward (Kitware), David Y. Huang (UNC Dept of Neurology), James E. Faber (UNC Dept of Cell & Molecular Phisiolgy), Gisele Matheus (Medical University of Source Carolina), and J. Keith Smith (UNC Dept of Radiology).
The decision to initiate therapy in acute ischemic stroke remains a balance between potentially worsening disability versus promoting significant salvage of brain tissue. Although it is well known that collateral status is an important determinant of penumbral volume, final infarct size and clinical outcome, rapid and systematic non- or minimally invasive assessment of status remains challenging. While the gold standard for evaluating leptomeningeal collaterals remains digital subtraction angiography (DSA), more recent work has focused on manual evaluation based on computed tomographic angiography (CTA) and MRI-based approaches. The goal of this study was to explore automated vessel extraction techniques for the evaluation of leptomeningeal collaterals on CTA.
Materials & Methods
Five patients with acute stroke and an M1/A1 occlusion were identified retrospectively. Technically adequate data sets were identified that did not have evidence of significant venous contamination or patient motion. CTA data were anonymized and transferred offline for analysis. Custom software was written using the ITK/VTK framework and TubeTK libraries for the identification and extraction of vessels within the CTA. Arterial vessels were extracted in a semiautomatic fashion. No specific effort was made to exclude venous vessels. The extracted vessels then were rendered in 3D for evaluation by a neuroradiologist. The presence of vessels identified as leptomeningeal collaterals was graded based on comparison to the contralateral side. Leptomeningeal collaterals also were evaluated independently after blinding to the 3D data, based on maximum intensity projections (MIPs) of the data set in the axial plane.
Good agreement was obtained between leptomeningeal collateral grade based on the semiautomated vessel extraction relative to the MIP data sets. Vessel geometries could be identified readily on the 3D images.
The rapid evaluation of leptomeningeal collaterals can provide important diagnostic information for acute stroke patients. Our preliminary work demonstrates that this evaluation may be performed in a semiautomated manner, which is an important component in a systematic, quantitative evaluation of collateral flow. Furthermore, this will provide the framework for more advanced analysis and research in the future. Work is ongoing to further improve collateral detection, extraction and flow for correlation with clinical outcomes.