Comparison of Different Acceleration Factors of Deep Learning Reconstruction in Cine Steady State Free Precession Sequences in Patients with Congenital Heart Disease

Background: Cardiovascular magnetic resonance (CMR) is the goal standard for assessment of ventricular volumes and function accomplished by a cine balanced steady state free precession (bSSFP) multislice short axis acquisition. Deep learning reconstruction (Sonic DL) of undersampled MRI k-space data can maintain clinically diagnostic image quality while reducing scan time.^1,2 This study evaluated the clinical utility of an accelerated bSSFP cine sequence with varying amounts of Sonic DL acceleration in patients with congenital heart disease (CHD).

Methods: A retrospective cohort study was performed on a 1.5T SIGNA Artist scanner (GE HealthCare, Waukesha, WI) using Sonic DL at two different acceleration factors (6x and 12x) bSSFP short axis stack cines in consecutive patients with CHD. Patient characteristics, CHD type, and previous CHD surgery was collected. Quantitative bi-ventricular volumetry, function, and left ventricular mass measurements were performed by a cardiologist with >12 years of experience and compared between the two acceleration factors using HeartIT Precession (Intelerad, Montreal, Canada) (Figure 1). Descriptive statistics reported as mean (SD) or median (IQR) depending upon data distribution and comparisons between acceleration factors made with paired T-test or Wilcoxon signed rank where appropriate.

Results: Thirty-eight consecutive patients with CHD were studied with patient characteristics in Table 1. The most common CHD diagnoses were tetralogy of Fallot, bicuspid aortic valve, and single ventricle (Table 1). Most of the patients had a history of at least one surgery for CHD (n=28, 74%). Only seven patients (18%) required anesthesia. All the measured CMR volumes, mass, and functions were similar between the two acceleration factors except the right ventricular end-diastolic and end-systolic volumes indexed were lower in the 12x acceleration compared to the 6x acceleration (Table 2), though only to a modest degree. The heart rate at the time of the sequence was higher in the 12x acceleration (Table 2). The sequence time to complete the entire short axis stack was 4-times less in the 12x acceleration compared to the 6x acceleration (Table 2).

Conclusion: The high acceleration factor Sonic DL bSSFP cine has clinically similar right and left ventricular CMR measurements to low acceleration factor Sonic DL bSSFP cine with significantly decreased acquisition time. High acceleration Sonic DL bSSFP cine can be used reliably in patients with CHD.

Table 1. Patient characteristics at time of cardiovascular magnetic resonance.

Median (IQR), Mean (SD), or N (%)	Nf38</strong>
Male	20 (53%)
Race
White	21 (55%)
Black	15 (39%)
Asian	2 (5%)
Age at CMR (y)	16.5 (14.3 – 22.9)
Weight at CMR (kg)	66.1 (23.6)
Height at CMR (cm)	164.8 (156.7 – 172.4)
BSA at CMR (m2)	1.68 (0.37)
CHD Diagnosis
Atrial Septal Defect	5 (13%)
Bicuspid Aortic Valve	9 (24%)
Hypoplastic Left Heart Syndrome	4 (11%)
Pulmonary Valve Stenosis	3 (8%)
Single Ventricle	8 (21%)
Tetralogy of Fallot	10 (26%)
Transposition of the Great Arteries	2 (5%)
CHD Surgery, Yes	28 (74%)
Anesthesia, Yes	7 (18%)

CHD = congenital heart disease; CMR = cardiovascular magnetic resonance; IQR = interquartile range; SD = standard deviation.Table 2. Cardiovascular magnetic resonance characteristics.

CMR parameters Mean (SD) or Median (IQR)	12x Acceleration	6x Acceleration	p-value
Heart rate (bpm)	76.4 (12.1)	74.4 (12.9)	0.003
Sequence Time (min)	1.3 (1.2 – 1.5)	4.6 (4.0 – 6.8)	< 0.001
Right Ventricle, n = 35
RVEDVi (ml/m2)	89.6 (68.6 – 108.5)	93.0 (74.3 – 113.1)	< 0.001
RVESVi (ml/m2)	44.0 (30.6 – 56.1)	43.4 (33.5 – 60.9)	0.018
RVSV (ml)	77.3 (25.5)	78.9 (25.4)	0.066
RVEF (%)	51.2 (7.5)	50.8 (7.8)	0.483
Left Ventricle, n = 34
LVEDVi (ml/m2)	74.9 (59.9 – 84.2)	75.5 (61.7 – 85.0)	0.214
LVESVi (ml/m2)	30.9 (23.6 – 36.8)	32.7 (24.7 – 36.7)	0.270
LVSV (ml)	75.0 (25.5)	76.4 (26.6)	0.069
LVEF (%)	57.4 (6.6)	57.2 (6.4)	0.742
LVED Mass Index (g/m2)	53.4 (44.9 – 62.5)	51.9 (46.4 – 57.5)	0.550

CMR = cardiovascular magnetic resonance; LVED = left ventricular end-diastolic; LVEDVi = left ventricular end-diastolic volume indexed; LVEF = left ventricular ejection fraction; LVESVi = left ventricular end-systolic volume indexed; LVSV = left ventricular stroke volume; RVEDVi = right ventricular end-diastolic volume indexed; RVEF = right ventricular ejection fraction; RVESVi = right ventricular end-systolic volume indexed; RVSV = right ventricular stroke volume. Figure 1. Balanced steady state free precession cine stack at high (12x) and low (6x) deep learning acceleration factors. Right ventricular endocardial borders (orange circles) and left ventricular endocardial (fuchsia circles) and epicardial (teal circles) are drawn at end-diastole and end-systole.