Quantification of cerebral veins in patients with acute migraine with aura: A fully automated quantification algorithm using susceptibility-weighted imaging

• Published in: PloS one Vol. 15; no. 6; p. e0233992
• Main Authors: Breiding, Philipe Sebastian, Kellner-Weldon, Frauke, Grunder, Lorenz, Scutelnic, Adrian, Fischer, Urs, Meinel, Thomas Raphael, Slavova, Nedelina, Gralla, Jan, El-Koussy, Marwan, Denier, Niklaus
• Format: Journal Article
• Language: English
• Published: San Francisco Public Library of Science 03.06.2020; Public Library of Science (PLoS)
• Subjects: Algorithms; Analysis; Asymmetry; Automation; Biology and Life Sciences; Blood vessels; Brain; Cerebral hemispheres; Diagnosis; Diseases; Ethics; Headache; Hemispheric laterality; Image processing; Image processing software; Image segmentation; Magnetic resonance imaging; Mathematical analysis; Medical imaging; Medicine and Health Sciences; Methods; Migraine; Morphology; Neuroimaging; Neurology; Oxygenation; Patients; Physical Sciences; Radiology; Research and Analysis Methods; Stroke; Studies; Time; Veins & arteries; Switzerland
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0233992

Introduction Susceptibility weighted imaging (SWI) is a very sensitive technique that often depicts prominent focal veins (PFV) in patients with acute migraine with aura (MwA). Interpretation of visual venous asymmetry (VVA) between brain hemispheres on SWI may help support the clinical diagnosis of MwA. Our goal was to develop an automated algorithm for segmentation and quantification of cerebral veins using SWI. Materials and methods Expert readers visually evaluated SWI of patients with acute MwA for VVA. Subsequently a fully automated algorithm based on 3D normalization and 2D imaging processing using SPM and MATLAB image processing software including top-hat transform was used to quantify cerebral veins and to calculate volumetric differences between hemispheres. Results Fifty patients with MwA were examined with SWI. VVA was present in 20 of 50 patients (40%). In 95% of patients with VVA, the fully automated calculation agreed with the side that visually harboured more PFV. Our algorithm showed a sensitivity of 95%, specificity of 90% and accuracy of 92% for detecting VVA. Patients with VVA had significantly larger vein volume on the hemisphere with more PFV compared to patients without (15.90 ± 5.38 ml vs 11.93 ± 5.31 ml; p = 0.013). The mean difference in venous volume between hemispheres in patients with VVA was larger compared to patients without VVA (16.34 ± 7.76% vs 4.31 ± 3.26% p < 1E-10). The average time between aura onset and SWI correlated negatively with venous volume of the dominant brain hemisphere (r = -0.348; p = 0.038). Conclusion A fully automated algorithm can accurately identify and quantify cerebral venous distribution on SWI. Absolute quantification may be useful for the future assessment of patients with suspected diseases, which may be associated with a unilateral abnormal degree of venous oxygenation.