Magnetic resonance imaging (MRI) is uniquely suited to study the pathophysiology of arteriosclerosis. So far, magnetic resonance (MR) measurements of vessel dimensions have mainly been done by manual tracing of vessel wall contours. However, such data postprocessing is very time-consuming and has limited accuracy due to difficulties in precise tracing of the thin vessel wall.
Purpose: To assess the accuracy and reproducibility of quantitative vascular MR imaging applying a data analysis method based on (1) vessel wall unwrapping, followed by (2) a gradient detection algorithm for MR data postprocessing. Vascular MR imaging studies were done both in vessel phantoms and in healthy volunteers (n=29) on a clinical 1.5 T MR scanner. A dark blood double-inversion turbo spin echo sequence with fat suppression was applied, with proton-density-weighted and breath-hold acquisition for aortic imaging and T2-weighted acquisition for carotid imaging. Intraobserver and interobserver variability were systematically evaluated by two independent observers. A repeat study within 10 days of the first MRI was performed in 10 of these subjects for assessment of interstudy reproducibility.
Results: The semiautomated edge detection software revealed a clear view of the inner and outer vessel wall boundaries both in the phantoms and in the volunteers studied. There was close agreement between MR-derived measurements and phantom dimensions (mean difference of 1.1+/-16.9 mm2, 8.0+/-19.9 mm2, 9.0+/-12.1 mm2 for vessel wall cross-sectional area, inner vessel area, and total vessel area, respectively). Quantification of vessel dimensions was feasible in all 29 healthy volunteers studied. Semiautomated quantification of cross-sectional vessel wall area (mean+/-SD, 253.6+/-208.4 mm2) revealed close correlation for repeated measurements by one or two observers (r=0.99 each). Both intraobserver and interobserver variability of vessel wall area MR measurements were low (mean difference 7.5+/-16.7 mm2 and 14.4+/-24.6 mm2 , respectively). In the repeat study of 10 volunteers, MRI with semiautomated postprocessing quantitation revealed a high correlation and agreement of vessel dimensions between the two scans (r=0.994, mean difference 2.6+/-25.1 mm2).
Conclusion: Semiautomated analysis methods can provide approaches that benefit from the human understanding of the image and the computer's ability to measure precisely and rapidly. Thus, by combining the latest MRI methods and semiautomated image analysis methods, we are now able to reproducibly determine the geometric parameters of blood vessels.