Clinical aortic diameter measurement is currently based on 2D imaging techniques such as echo or MRI. These are inaccurate, given the difficulty in prospectively prescribing a plane that is truly transverse to the target vessel (for cross-sectional diameter or area measurement) and it is even more challenging to achieve consistent plane prescription on follow up serial studies. 3D techniques are highly favorable. 3D MRI enables full 3D visualization of the vessel, offers precise plane definition (perpendicular to the vessel of interest in a reproducible location) in postprocessing, and permits robust co-registration of serial studies. 3D imaging also provides measurements at all levels of the aorta instead of a limited set of specific locations acquired in separate 2D acquisitions. In this study, we propose to achieve aortic functional measurements with an accelerated non-contrast-enhanced 3D aortic CINE MRI technique and a Level-Set Methods based automatic segmentation method on aorta.

An accelerated non-contrast-enhanced 3D cardiac MRI sequence has previously been developed¹ and applied to acquire aortic imaging during the whole cardiac cycle in this study. The sequence has been applied on 5 healthy volunteers (31.0±3.1 years, two female) on a 3.0T MR scanner (GE Medical Systems, Milwaukee, WI) with an 8-channel cardiac coil during free breathing. A 3D balanced Steady-State Free Procession (bSSFP) sequence with a golden-ratio based pseudo-random sampling strategy CIrcular Cartesian UnderSampling (CIRCUS)² was applied, with FOV=280mm, TR/TE=4.0/1.6ms, FA=45°, BW=±125kHz, slice thickness of 2mm, image matrix=256×160×32 (75% partial acquisition in ky&kz), and scan time of 3 minutes. Retrospective cardiac gating and respiratory self-gating (with 50% gating efficiency) were applied¹. Cardiac phases were reconstructed with a temporal resolution of 50 ms with a combined compressed sensing and parallel imaging method, k-t SPARSE-SENSE^3,4. The overall scan time acceleration was around R=6. Segmentation of the aorta is crucial for achieving accurate functional measurements. Automatic segmentation in 3D or 4D (3D+time) MRI data is a difficult task due to the inherent noise associated with MRI data caused by different reasons and the inhomogeneous image gradient. Previous studies have exploited the Level-Set Methods by using a class of deformable models where the shape to be recovered is captured by propagating an interface represented by the zero level-set of the level-set equation⁵. The evolution of the interface can be regarded as a derivation of a variational formulation and then the segmentation problem can be expressed as the minimization of an energy functional, which will represent the properties of the target objects. Considering that the aorta is normally circular, our approach firstly uses the Hough transform⁶ to find the circular structures, which will be used as the initial contours. In order to take the different characteristics of the aorta into consideration, a level-set equation with local Gaussian distribution fitting energy⁷ is then incorporated into the algorithm to segment the aorta. The evolution equation can read:

$$\frac{\partial\phi}{\partial{t}}=-\delta(\phi)(e_1-e_2)+\upsilon\delta(\phi)div(\frac{\nabla\phi}{\left|\nabla\phi\right|})+\mu(\nabla^2\phi-div(\frac{\nabla\phi}{\left|\nabla \phi \right|}))$$

where the image-based term $$$(e_1-e_2)$$$ is independent of scale of local intensities caused by intensity inhomogeneity. $$$\mu$$$ that controls the smoothness of zero level set is a fixed parameter. Besides, $$$\upsilon$$$ increases the propagation speed and $$$\delta(\phi)$$$ is the Dirac function.

In this study, we explored the Level-Set Methods for automatically segmenting aorta lumen with the acquired non-contrast-enhanced 3D CINE (4D) aorta images.

We have successfully acquired data from all five subjects. Figure 1 shows the reformatted images of aorta in three orthogonal plans (columns), acquired with the conventional 2D imaging and our proposed 3D imaging respectively (top and bottom rows) from a representative subject. The reformatted images with 2D imaging show obvious slice misregistration errors due to the variations of the breath-holding position, while 3D imaging provides continuous coverage of the aorta for more precise post-processing. Figure 2 shows the results using our proposed automatic segmentation algorithm on the aorta, along a slice covering both ascending (highlighted contour on the left) and descending (highlighted contour on the right) aorta at end-systolic (top row) and end-diastolic (bottom row) phases respectively. Entire aorta segmentation at a representative cardiac phase is displayed in Figure 3a, and its straightened visualization is shown in Figure 3b by reconstructing the aorta with a straightened centerline. Cross-sectional area was calculated at each location along the aorta and throughout the cardiac cycle, shown in Figure 3c), which provides a practical and precise way for aortic functional measurements, such as aortic aneurysm evaluation and aortic stiffness measurement. In summary, we have developed an accelerated free-breathing self-gated non-contrast-enhanced 3D CINE MRI method and automatic aorta segmentation algorithm for improving aortic functional measurements.