To discriminate normal from abnormal coronary endothelial function, recent studies have used MRI with isometric handgrip exercise^1-4 as the endothelial dependent stressor and reported excellent and reproducible results.⁴ However, the sensitivity of MRI to measure small changes in cross sectional area of the coronary arteries in response to stress remains to be quantitatively examined. Since the spatial resolution of MRI is limited relative to these expected area changes, it is of utmost importance to address this question. In this study, we have therefore measured the sensitivity of radial MRI for detecting small changes in coronary cross-sectional areas.

Phantom setup: A phantom was designed to simulate various cross-sectional areas of human coronary arteries⁵ by drilling holes of different diameters in a block of Polyacetal copolymer (POM-C) (Figure 1a). Twenty-two different diameters, ranging from 3.00mm to 3.42mm, in steps of 0.02mm, were each assigned to 5 random locations on the phantom, so as to avoid introducing potential bias due to magnetic field inhomogeneities (Figure 1b). The phantom was placed in a container filled with tap water and doped with gadolinium (concentration of 5.9mM) to simulate the time-of-flight effect observed in cine imaging. Finally, the phantom was placed on a moving tray to simulate a sinusoidal cardiac motion with a frequency of 40bpm and a maximal displacement of 2cm (Figure 2).

Data acquisition: Data were acquired on a 3T clinical scanner (MAGNETOM Prisma, Siemens Healthcare) using a conventional 2D radial retrospectively ECG-gated cine sequence with an 18-channel chest coil and a 32-channel spine coil. The imaging plane was placed perpendicular to the drilled holes and images were acquired using five different isotropic in-plane resolutions (0.5, 0.6, 0.7, 0.8 and 0.9mm). The acquisition was repeated 10 times for each resolution with the following parameters: FOV=260×260mm², matrix=288-512, slice thickness=6.5mm, TE/TR=2.5-2.9/4.9-5.1ms, radiofrequency excitation angle=22˚, and temporal resolution=40ms.

Cross-sectional area measurements: Two cine frames with minimum motion were visually selected to measure the cross-sectional areas of the drilled holes with a fully-automated custom-written software package developed in MATLAB. The automatic segmentation followed a similar procedure as described previously⁶ and uses the full-width half maximum criterion (FWHM) for area measurements. Figure 3 illustrates the various stages of the segmentation.

Statistical analysis: The areas measured for each nominal diameter $$$d$$$ of the drilled holes were grouped together for statistical analysis (Figure 4a) and are denoted by $$$X_d$$$. The normality of the measurements was tested using both the Lilliefors and Jarque–Bera tests. The mean $$$\mu_{X_d}=E(X_d)$$$ and variance $$$\sigma_{X_d}^2=E\big((X_d-\mu_{X_d})^2\big)$$$ for each diameter were computed to derive a probability distribution function that describes the probability of the possible measured areas, $$$X_d\sim N(\mu_{X_d},\sigma_{X_d}^2)$$$. Two normally distributed measures $$$X_i\sim N(\mu_{X_i},\sigma_{X_i}^2)$$$ and $$$X_j\sim N(\mu_{X_j},\sigma_{X_j}^2)$$$ where $$$\mu_{X_i} > \mu_{X_j}$$$ were considered statistically different if the probability of $$$X_i - X_j \sim N(\mu_{X_i}-\mu_{X_j},\sigma_{X_i}^2+\sigma_{X_j}^2)$$$ being positive (i.e., $$$\ge0$$$) is $$$\ge0.95$$$ (Figure 4). Each pair of $$$X_i$$$ and $$$X_j$$$ was compared using this technique. The differences of diameters ($$$i-j$$$) and the results of the tests were stored for each pair. The sensitivity of MRI or smallest detectable change in cross-sectional area was defined as being greater than the highest difference of diameters that has not passed the statistical test. Bland-Altman plots and linear regression analyses were also used for statistical comparisons of the distributions.

A total of 110 distributions of area measurements (22 diameters and 5 resolutions) were analyzed and tested for normality. Each diameter was measured 100 times (5 holes per image x 2 images per acquisition x 10 acquisitions). The Lilliefors and Jarque–Bera normality tests confirmed that 81.8% and 90.9% of the distributions, respectively, could be well-modeled by a normal distribution. The Bland-Altman plots in Figure 5 show a statistically significant bias in the measured areas that is inversely proportional to the image resolution; however the 95% confidence interval characterizing the spread of the data remains approximately constant at 0.4mm² for all resolutions. The smallest detectable area change ranged between 0.21-0.31mm² for the different resolutions (Figure 5f) and did not significantly correlate with the image resolution (R=0.13).We presented a moving phantom experiment to quantify the sensitivity of MRI for detecting small changes in coronary cross-sectional areas. Our results suggest that the above MRI approach is capable of distinguishing area changes in the order of 0.2-0.3mm², which correspond to a percentage area change of 3-4% for a nominal diameter of 3mm. To put this in perspective, and for healthy subjects, the values of coronary endothelial responses reported in the literature using both MRI^1-4,7 and invasive techniques^8-10 range from 13.2% to 23.1%.