SCMR/ISMRM Co-Provided Workshop
SCMR 22nd Annual Scientific Sessions
Background: In recent years, image reconstruction techniques that either correct for, or resolve, respiratory motion by binning imaging data into respiratory states using a self-gating signal have shown promise in Coronary Magnetic Resonance Angiography (CMRA) [1, 2]. However, breathing patterns are highly individual ; suggesting that a self-gating method that works perfectly in one subject might not work as well in another. Therefore, a comparison of techniques for extraction of such signals in a large and diverse cohort is warranted. Our aim is to test in CMRA XD-GRASP [2, 4], if respiratory signals extracted by applying Principal Component Analysis (PCA) to Superior Inferior (SI) projections [1, 5], a robust approach in our experience, provides better image quality than signals extracted from the center of k-space in SI readouts  or Independent Component Analysis thereof .
Methods: CMRA datasets from 69 cardiac patients and 20 healthy volunteers, who had provided written informed consent, were included in this retrospective IRB approved study. All data were acquired using a prototype ECG-triggered 3D radial bSFFP sequence [6, 7] on a 1.5T clinical MRI scanner (MAGNETOM Aera, Siemens Healthcare). To compare the three self-gating methods (Fig. 1), a respiratory signal was extracted and used to sort the data into four respiratory bins for each method and subject. Two reconstructions, motion-resolved gridding and XD-GRASP, were performed with identical parameters for the three strategies. Subsequently, the image quality of all respiratory phases was estimated with a convolutional neural network , previously trained to gauge CMRA image quality on a scale from 0 (poor) to 4 (excellent). For both reconstruction types, the number of subjects where the different self-gating methods provided the respiratory phase with the highest image quality was counted. Moreover, for each method, the average score of the best respiratory phase for each subject was computed across the cohort in the XD-GRASP reconstructions. The XD-GRASP image quality was compared using t-tests (p < 0.05 considered statistically significant) both among methods (paired, Bonferroni correction) and between healthy subjects and patients.
Results: The subjects with the smallest and largest differences in image quality between different self-gating methods are depicted in Figure 2. The average XD-GRASP image quality for the different methods across the full cohort, volunteers, and patients are summarized in Figure 3a. In general, respiratory signals from PCA applied to SI-projections provided the highest image quality scores, both in the gridding and the XD-GRASP reconstructions (Fig. 3b). Overall, the image quality was higher in the healthy subjects than in the patients (Fig. 3c).
Conclusion: Self-gating using PCA of SI-projections provided higher image quality scores than the other methods in our heterogeneous cohort suggesting better robustness. It seemed to be of particular benefit in the patient sub-group (Fig. 3a).