Free-running 3D whole-heart angiography and cine in less than 2 minutes: A feasibility study (RF_TH_247)

Two-dimensional (2D) breath-hold cardiac MR cine is the golden standard to determine volumetric measurements of the heart[1]. But the data acquisition for 2D cine can be challenging for ill or pediatric patients, as they have significantly reduced breath-holding capabilities [2].

In this study, we investigate the feasibility of a fast free-breathing 3D whole-heart angiography and cine technique based on the Heart-NAV technique [1] together with a deep-learning-based image reconstruction for highly undersampled data. In this work, the data is 8-fold undersampled to significantly reduce the acquisition time to one and a half minutes. To address the low SNR and CNR due to high undersampling factor, a deep-learning reconstruction was used to improve the image quality.

Methods:

A 3D steady-state free precession (SSFP) sequence together with the Heart-NAV technique and Ferumoxytol contrast was used for data acquisition. Heart-NAV technique is used to measure the displacement of the heart to get the respiratory signal and hence minimizing the respiratory motion artifacts. The data were then 8-fold undersampled, enabling an average acquisition time of one and half a minute in 17 pediatric patients with congenital heart diseases. The acquired voxel size is 1.9 x 1.9 x 1.9 mm³ and 20 cardiac phases were obtained. Image reconstruction was performed using two methods for image quality comparison. The in-line compressed-sensing (CS) reconstruction, which was provided by the MR vendor (Philips, Best, Netherlands) and a deep-learning based reconstruction algorithm termed Adaptive CS-Net [3].

To analyze the quality of the reconstructed images, MATLAB was used to select regions of interest within the images to determine the SNR and CNR values (in decibel) for both the CS and Adaptive CS-Net reconstructed images based on the heart, lungs, and liver.

Results:

Comparing traditional 2D cine images acquired with breath holds to 3D cine with the free-breathing technique, it was determined that the images were similar visually, while the 3D cine images showed greater contrast as shown in Figure 1. Furthermore, in the first part of Figure 2 it can be noted that there was not a significant difference (p > 0.05) in volumetric measurements for EDV, ESV, SV, and EF except for the right ventricle ejection fraction (p = 0.035).

The SNR and CNR values using Adaptive CS-Net were greater in comparison to the images reconstructed with CS and showed a significant difference (p < 0.05) for the heart, lungs, liver, and pulmonary veins, as shown in the second part of Figure 2.

Visually comparing 3D cine images reconstructed with Adaptive CS-Net and the clinical contrast-enhanced 3D MR angiography acquired using the bSSFP sequence in Figure 3 showed similar image quality while reducing the scan time from 7 minutes per cardiac phase for the clinical MR angiography sequence to less than 2 minutes to acquire 20 phases of the heart cycle.

Conclusion: In this work, we proposed a free-breathing 3D whole-heart angiography and cine technique using the Heart-Nav technique and Ferumoxytol contrast. The data were highly undersampled and enabled the acquisition time of less than 2 minutes. A deep-learning reconstruction algorithm was then used to overcome the undersampling artifacts. Qualitative and quantitative results showed that the proposed framework enabled similar clinical diagnosis as the techniques used in clinical practice, but the proposed framework significantly reduced the required time and eliminated the need for breath-holds.