Comparison of Inter‐Observer Bias between Low Resolution and High Resolution Scans using 3T and 7T Scanners

Abstract

IntroductionMRI can be used to assess atherosclerotic disease severity and to identify plaque components noninvasively. Vessel wall thickening can be measured with MRI and is associated with atherosclerosis. However, there are observer biases when measuring the vessel wall. We investigate whether there is a difference in inter‐observer bias between low resolution and high resolution imaging in both 3T and 7T field strengths.MethodsAtherosclerosis was induced in 5 male New Zealand White rabbits. The rabbits were placed on a high fat diet for 4 months and then imaged. Animals were each imaged twice, once on a 7T clinical scanner (Magnetom, Siemens Healthineers), and once on a 3T clinical scanner (mMR, Siemens Healthineers). The 7T scanner utilized a custom rabbit coil consisting of a transmit‐only quadrature‐driven high‐pass birdcage and 12‐channel receive‐only loop elements. The 3T scanner used a product Tx/Rx15‐ channel knee coil. On both scanners, localizer scans and time‐of‐flight (TOF) angiography were used to identify the rabbit aorta. A 3D T2 weighted SPACE (Sampling Perfection with Application optimized Contrasts using different flip angle Evolution) sequence was acquired using 0.6 mm3 (Low Resolution) and 0.4mm3 (High Resolution). The sequence was acquired in the sagittal plane, covering the rabbit abdominal aorta. Imaging data was analyzed with Osirix software. All images were reformatted in the axial plane, and axial slices from the left renal artery to the aortic bifurcation were analyzed. Vessel wall area was measured by tracing and subtracting the inner and outer vessel wall contours. Two independent observers traced each data set. Inter‐observer reliability was evaluated using Bland‐Altman analysis.ResultsFor the 3T low resolution scans, the percent bias was 37.41, the percent limit of agreement was from −5.868 to 80.69, the absolute bias was 0.05566, and the absolute limit of agreement was from −0.01227 to 0.1236. For the 3T high resolution scans, the percent bias was 34.31, the percent limit of agreement was from −5.987 to 74.62, the absolute bias was 0.03818, and the absolute limit of agreement was from −0.01884 to 0.0952. For the 7T low resolution scans, the percent bias was 42.85, the percent limit of agreement was from −1.001 to 86.70, the absolute bias was 0.06657, and the absolute limit of agreement was from −0.002535 to 0.1375. For the 7T high resolution scans, the percent bias was 41.35, the percent limit of agreement was from −6.707 to 89.4, the absolute bias was 0.04268, and the absolute limit of agreement was from −0.01173 to 0.0970.ConclusionBland‐Altman analysis revealed a bias for all imaging sequences analyzed between two observers. For both 3T and 7T field strengths, percent and absolute bias were smaller for the high resolution scans. Furthermore, the absolute and percent biases were lower for images acquired at 3T. Overall, these data indicate that a significant percent and absolute bias in vessel wall area measurements can be observed among different analysts. These findings underscore the importance of analysis standardization among observers when calculating absolute measures of plaque burden in animal studies or clinical trials. Furthermore, increasing imaging spatial resolution may help to reduce bias between observers and allow for more robust quantitative analysis.This abstract is from the Experimental Biology 2018 Meeting. There is no full text article associated with this abstract published in The FASEB Journal.