Comparison of Deep Learning-Based Free Breathing with Conventional Breath Holding CMR for Quantifying Degenerative Mitral Regurgitation.
Authors
Affiliations (2)
Affiliations (2)
- Department of Cardiovascular Medicine, Gagnon Cardiovascular Institute, Morristown Medical Center/Atlantic Health, Morristown, NJ.
- Department of Cardiovascular Medicine, Gagnon Cardiovascular Institute, Morristown Medical Center/Atlantic Health, Morristown, NJ. Electronic address: [email protected].
Abstract
Cardiovascular magnetic resonance (CMR) imaging acquisition frequently relies on breath-holding, which poses challenges in patients with certain underlying comorbidities. A new technique consisting of deep learning-based accelerated 2D steady-state free precession (SSFP) cine imaging (Sonic DL Cine, GE HealthCare) allows for 1 heart beat free-breathing SSFP acquisition. This study aimed to demonstrate the accuracy of free-breathing acquisition SSFP using Sonic DL Cine and free-breathing phase contrast imaging versus standard breath holding for mitral regurgitation (MR) quantification in patients with degenerative MR. Patients with degenerative MR were prospectively enrolled and underwent both breath-holding and free-breathing CMR. Short axis SSFP images and phase contrast images in the ascending aorta and proximal pulmonary artery were acquired using breath hold and free-breathing techniques to quantify ventricular volumes and blood flow in the ascending aorta and proximal pulmonary artery. MR volume and fraction were calculated for each patient using both techniques. Pearson's correlation and Bland-Altman plots were utilized to compare the two groups. A total of 27 patients (64 ± 9 years; 67% male) underwent 33 CMR studies. There was no statistically significant difference in left ventricular end-diastolic volume (LVEDV), LV stroke volume (SV), aortic phase contrast, pulmonary artery phase contrast, MR volume or MR fraction between breath-holding and free-breathing CMR. Pearson's correlation and Bland-Altman plots demonstrated excellent correlation and good agreement for LVEDV (r<sup>2</sup> = 0.96; mean difference 0.7mL; p<0.001), LVESV (r<sup>2</sup> = 0.93; mean difference 2mL; p<0.001), LVSV (r<sup>2</sup> = 0.96; mean difference -1 mL; p<0.001), MR volume (r<sup>2</sup> = 0.96; mean difference -1mL; p<0.001), and MR fraction (r<sup>2</sup> = 0.93; mean difference -0.4mL; p<0.001). Compared to conventional breath-holding, free-breathing CMR is feasible and comparable in the evaluation of ventricular parameters and MR quantification in patients with degenerative MR. This CMR technique would allow clinicians to accurately quantify MR in patients who are unable to perform conventional breath hold CMR.