3D whole-brain vessel wall cardiovascular magnetic resonance imaging: a study on the reliability in the quantification of intracranial vessel dimensions

The prospective study was approved by the local institutional review board. Thirty-four healthy subjects (24 males; 14 aged 31–49 years and 20 aged 50–66 years) without known cerebrovascular diseases and 10 patients (7 males; 42–69 years, mean 51.2 years) with known ICAD were recruited. Written informed consent was obtained from all subjects.

All CMR examinations were performed on a 3-Tesla whole-body system (MAGNETOM Verio, Siemens Healthineers, Erlangen, Germany) with a 32-channel head coil. Subjects were scanned in a supine position with a foam padding to minimize head movement. Two repeated 3D IVW CMR scans were performed with an off-table break for healthy subjects and 7 to 11-day intervals for patients [18, 20]. When any of the two scans exhibited motion-related image blurring at the discretion of the CMR technologist, reacquisition was attempted only once to simulate real clinical settings. Relevant imaging parameters were as follows: sagittal imaging orientation, repetition time (TR) /echo time (TE) = 900/15 ms, receiver bandwidth = 488 Hz/pixel, field of view = 170 × 170 × (110–127) mm³, matrix size = 320 × 320 × (208–240) with 7.7–6.7% partition oversampling, spatial resolution = 0.53 × 0.53 × 0.53 mm³ (without zero-filled interpolation), turbo factor = 52, echo train duration = 271 ms, 6/8 partial Fourier in the partition-encoding direction, parallel imaging (GRAPPA) acceleration rate = 2 in the phase-encoding direction, scan time = 7 min 10 s – 8 min 10 s depending on the head size.

In 19 out of the 34 healthy subjects, T1-weighted 2D TSE was also performed during the rescan session to provide an CMR imaging reference for assessing the inter-method agreement. Due to its relatively poor scan efficiency, the acquisition was prescribed only for three arterial segments that often present with ICAD in patients, including the distal basilar artery (BA), distal internal carotid artery (ICA) supraclinoid segment (C4), and proximal middle cerebral artery (MCA) M1 segment. Immediately after the 3D IVW CMR rescan, 3D images were reconstructed into three contiguous 2-mm-thick cross-sections at each of the three segments by an experienced CMR technologist using the multiplanar reconstruction (MPR) functionality available on the imaging console. A 2D TSE image was then acquired for each of these cross-sections with following imaging parameters: TR/TE = 800/12 ms, receiver bandwidth = 411 Hz/pixel, field of view = 170 × 170 mm³, matrix size = 320 × 320, spatial resolution = 0.53 × 0.53 mm², slice thickness = 2 mm, turbo factor = 9, signal averages = 4, scan time per slice = 1 min 37 s.

All images were transferred to a workstation (Syngo MultiModality Workplace, Siemens Healthineers). The scan and rescan 3D IVW CMR image sets were first co-registered using an image fusion functionality to account for head repositioning. At the same locations on both image sets, 5 following vessel segments were analyzed for each healthy subject: the distal BA, the distal vertebral artery (VA; V4), the distal ICA C4, the proximal MCA M1, and the proximal anterior cerebral artery (ACA) A1. Three contiguous cross-sections of 2-mm thickness were generated via MPR for each segment. For each patient, 3 contiguous cross-sections of 2-mm thickness centered at the thickest location of the most stenotic plaque were also generated via MPR.

All these reconstructed cross-sectional images and corresponding 2D TSE images underwent vessel wall and lumen dimension quantification using commercial software (VesselMass, Leiden University Medical Center, Leiden, the Netherlands). Each image was magnified 4–6 times with bilinear interpolation. Lumen and outer wall boundaries were traced manually along the interfaces between the lumen and wall and between the wall and surrounding tissue respectively, generating two contours (Fig. 1). When part of a boundary was invisible, the contour was completed to maintain the continuity of the vessel’s curvature [9]. The entire vessel wall region encased by the two contours were automatically divided into ten evenly spaced segments. The software generated the following measurements: the average and maximum wall thickness (i.e. the mean and maximum value of the ten distances between contours), the lumen area (i.e. the area inside the luminal contour), and the wall area (i.e. subtracting the inner contour area from the outer contour area). Additionally, normalized wall index was calculated as the ratio of the wall area to the outer contour area. Contouring and determination of the above wall and lumen dimensions for any 3 consecutive slices required a processing time of approximately 2.5 min per scan. For each vessel segment or plaque, the measured normalized wall index and mean/maximum wall thickness were, respectively, averaged over the three slices; lumen volume and wall volume were obtained by summing the area measurements of the three slices and multiplying by 2 mm.

Two readers (with 6-year and more than 10-year experience in vascular CMR imaging, respectively) independently performed above vessel wall and lumen measurements on the images from the first 3D IVW CMR scan. After two weeks, one of the readers performed a second-round measurement on the same data, and the other performed measurement on the images from the second scan followed by measurement another 2 weeks later for the 2D TSE scan when available.

All statistical analyses were performed using SPSS (version 19.0, International Business Machines, Armonk, New York, USA) and R (version 3.4.1). Intra-class correlation coefficient (ICC) was obtained from a two-way random model with two raters for inter-rater reproducibility and a two-way mixed model with two raters for intra-rater and scan-rescan reproducibility. Confidence intervals for the overall ICC were calculated by bootstrap taking in account the correlation between segments in the same patient. An ICC value of less than 0.4 was considered poor agreement, a value of 0.4–0.75 was considered good agreement, and a value of 0.75 or greater was considered excellent agreement [26]. Bland-Altman analysis was also used to determine the scan-rescan, intra-, and inter-observer reproducibility of 3D IVW CMR as well as inter-method agreement between 3D IVW CMR and 2D TSE in quantifying vessel dimensions for volunteers.

Based on the scan-rescan data analysis, the sample size required for each of dimension measurements to compare placebo and treatment group in a clinical trial with 80% of power at 5% significance level was calculated using a t-test with equal variances. It was assumed that the mean of the placebo group would be equal to the mean from the healthy subjects in our study and that the mean of the treatment group would be 5, 10, 15, and 20% different. The standard deviations for placebo and treatment groups were assumed to be equal and given by the subject variance estimated from a linear mixed model with subject as fixed effect and scan as random effect.

3D IVW CMR provided visually consistent delineation of the vessel wall (Fig. 2) and plaques (Fig. 3) in both scans. In some plaques, high signal-intensity features were observed (Fig. 3 case A). For healthy subjects, morphologic measurements and corresponding ICC values, when combining all assessed segments, are summarized in Table 1 (Segment-based results are summarized in Additional file 1: Table S1). Each of the assessed morphologic indices had all ICCs greater than 0.75, indicating excellent reproducibility. More specifically, for the intra-observer reproducibility, all ICCs were equal to or greater than 0.93. For the scan-rescan and inter-observer reproducibility, all ICCs except for that for the inter-observer reproducibility on normalized wall index were equal to or greater than 0.83. For patients, vessel wall and lumen measurements at the most stenotic plaque and corresponding ICC values are summarized in Table 2. All ICCs except for that for the inter-observer reproducibility on maximum wall thickness (ICC = 0.87) were equal to or greater than 0.91, indicating excellent reproducibility.

The Bland-Altman plots for all arterial segments of healthy subjects are shown in Fig. 4 for lumen volume, normalized wall index, and mean wall thickness, respectively. Random error scattering patterns and independence of the difference on the mean value were observed.

All ICCs of the healthy subgroup age ≥ 50 years were equal to or higher than that of the < 50 years subgroups, but in most cases, were lower than that of patients (Fig. 5). As shown in Table 3, there were no significant difference in lumen or wall volume between the two age groups. However, normalized wall index and mean and maximum wall thickness were significantly larger in the age group of ≥50 years (P ≤ 0.05).

Figure 6 shows representative 3D IVW images and reformatted vessel wall cross-sections as well as slice thickness- and location-matched 2D TSE images for three arterial segments. The two acquisition methods provided visually comparable vessel wall delineation. When all three segments were evaluated together, all paired morphologic measurements exhibited an excellent agreement as indicated by an ICC with 95% CI of 0.98 (0.95–0.99), 0.96 (0.84–0.98), 0.96 (0.92–0.97), 0.92 (0.82–0.96), and 0.88 (0.32–0.96) for lumen volume, wall volume, normalized wall index, mean wall thickness, and maximum wall thickness, respectively. Segment-based morphologic measurements and corresponding ICC values for 3D and 2D IVW CMR are summarized in Additional file 1: Table S2. The differences between 3D and 2D IVW CMR and the mean values with limits of agreement for all segments are illustrated in Bland-Altman plots (Fig. 7). The mean differences between 3D and 2D IVW CMR were 2.4 mm³ for lumen volume, 3.0 mm³ for vessel wall volume, 0.002 for normalized wall index, 0.02 mm for mean wall thickness, and 0.04 mm for maximum wall thickness. Bland-Altman analysis demonstrated good agreement with small bias between the two techniques.

Table 4 shows sample size required to compare placebo and treatment group for all measures means considering MCA segment with 80% of power at 5% significance level using a t-test for two independent samples with equal variance. The sample sizes required for the measures in other segments are also presented in Additional file 1: Table S3. Normalized wall index requires the smallest sample size while lumen volume requires the highest sample size to compare two groups. The large-sized segments (ICA, VA, and BA) requires the smaller sample size than small-sized segments (MCA and ACA).