Completely free-breathing cardiac MRI using deep learning reconstruction reduces sedation and scan time in children.
Authors
Affiliations (5)
Affiliations (5)
- UPMC Children's Hospital of Pittsburgh, 4401 Penn Ave, Pittsburgh, PA, 15224-1334, USA. [email protected].
- UPMC Children's Hospital of Pittsburgh, 4401 Penn Ave, Pittsburgh, PA, 15224-1334, USA. [email protected].
- UPMC Children's Hospital of Pittsburgh, 4401 Penn Ave, Pittsburgh, PA, 15224-1334, USA.
- Children's National Hospital, Washington, DC, USA.
- GE HealthCare, Chicago, IL, USA.
Abstract
Traditional cardiac magnetic resonance (CMR) imaging scan times are long and require breath-holds, often necessitating intubation and mechanical ventilation in young or sick children. To compare CMR duration and anesthesia requirement between highly accelerated single cardiac cycle (1RR) free breathing cine acquisition with deep learning reconstruction and conventional segmented breath-hold cine imaging in pediatric patients. Free breathing with deep learning imaging was used in 150 patients and compared to 150 sequential patients imaged with segmented breath-hold imaging. Specifically, CMR duration, use of anesthesia, anesthesia duration, and emergence time were evaluated. A subgroup analysis was performed on patients less than 35 kg. CMR duration decreased significantly in the 1RR group (29.2 ± 9.8 min vs 43.0 ± 13.2 min, P<0.01). The use of endotracheal intubation with muscle relaxation during the CMR decreased significantly with free breathing with deep learning sequences for those that required sedation (14% vs 90%, P<0.01), as did anesthesia duration (70.8 ± 24.2 min vs 98.3 ± 26.7 min, P<0.01) and emergence time from anesthesia (15 ± 3.7 min vs 27.1 ± 14.9 min, P<0.01). There was no degradation in contrast-to-noise ratio using blood pool/myocardium contrast difference (534.2 ± 165.1 vs 491.8 ± 165.8, P=0.03). Subanalysis performed on 125 patients (age 12.3 ± 4.1 years) with no significant valve regurgitation confirmed reliable volumetric measurements with free breathing with deep learning given high correlation between both right ventricular cardiac index and main pulmonary artery flow (R=0.92, P<0.001) and left ventricular cardiac index and aorta flow (R=0.84, P<0.001). Free breathing with deep learning single RR acquisition significantly reduces study duration, rates of endotracheal intubation, anesthesia duration, and anesthesia emergence time while maintaining diagnostic image quality and clinical quantitative assessments.