Feasibility of deep learning-based single-heartbeat cardiac cine MRI in children with challenging and non-challenging conditions

Background: Segmented cardiac cine MRI is the gold standard for ventricular volumetric assessment, but in patients with arrhythmia or impaired breath-holding it may yield suboptimal image quality. To overcome these limitations, a deep learning (DL)-based accelerated cine acquisition method, Sonic DL, has been developed. However, most prior studies focused on healthy volunteers or general populations, without specifically addressing cohorts prone to arrhythmia or limited breath-hold capacity, particularly in children. This study aimed to compare acquisition time, image quality, and volumetric measurements between Sonic DL and standard segmented cine MRI in pediatric patients with and without these challenging conditions.

Methods: In this prospective single-center study, consecutive pediatric patients with various cardiac diseases and healthy volunteers underwent MRI from January through June 2025. Both standard cine and Sonic DL cine were acquired in short-axis view at 3.0 Tesla. Subjective image quality was rated on a 5-point Likert scale in three categories: myocardial edge definition, temporal fidelity, and artifact level, with an overall score calculated as the mean of all three. Objective edge sharpness was assessed on a mid-ventricular slice at end-diastole. Left ventricular (LV) and right ventricular (RV) volumetry and LV mass were compared between cine methods using Bland-Altman plots and Wilcoxon signed-rank tests.

Results: Seventy-seven participants (mean age, 9.3 ± 4.7 years; 50 male) were included. DL cine significantly reduced acquisition time compared with standard cine (24.4 ± 1.8 s vs. 138.7 ± 26.8 s; P < 0.001). In 30 patients without arrhythmia or breath-holding difficulty (mean age, 11.7 ± 2.4 years; 19 male), no significant differences were observed in subjective scores or edge sharpness (all P > 0.05), and all overall image quality scores were above the clinically acceptable cutoff (≥3.0). LV volume measurements also showed no significant differences between the two methods. In 47 patients with arrhythmia or unreliable breath-holding (mean age, 7.7 ± 5.1 years; 31 male), DL cine demonstrated superior image quality (artifact score: 4.5 ± 0.5 vs. 3.4 ± 1.1, P < 0.001; edge sharpness: 0.25 ± 0.07 vs. 0.19 ± 0.08, P < 0.001). Notably, 17% (8/47) of these patients had overall image quality scores < 3 with standard cine, while none were observed with Sonic DL. Interobserver agreement was strong for both methods (intraclass correlation coefficients, 0.82–0.97).

Conclusion: The DL cine technique substantially reduced acquisition times while maintaining noninferior diagnostic quality compared with standard cine. Moreover, in children with arrhythmia or limited breath-hold capacity, it provided superior diagnostic quality and enabled comprehensive functional analysis.

Table 1. Characteristics of Study Population

	Total participants (n=77)	Children with non-challenging conditions (n=30)	Children with challenging conditions (n=47)
Age, y	9.3±4.7	11.7±2.5	9.1±4.9
0-6	20	0	20
6-12	31	17	14
12-18	26	13	13
Sex
Male	50	19	31
Female	27	11	16
Height, cm	130.6±33.5	147.6±16.4	128.5±36.4
Weight, kg	34.3±18.4	42.0±13.7	34.0±19.8
BMI, kg/m2	18.2±3.3	18.8±3.4	18.3±3.1
BSA, m2	1.10±0.44	1.31±0.28	1.08±0.48
HR, beats/min	90.2±20.0	81.8±17.7	91.7±16.9
SDP, mmHg	106.4±15.8	108.7±15.3	107.2±16.1
DBP, mmHg	65.2±13.7	63.4±13.3	67.7±14.5
Sedation
None	61	30	31
General anesthesia	16	0	16
Study indications
Cardiomyopathy	19	11	8
Congenital heart disease or/and valve disease	18	5	13
Coronary artery disease	2	1	1
Chest pain	3	2	1
Heart failure	4	1	3
Arrhythmias	36	0	36
Healthy volunteers	11	10	1

BMI, body mass index; BSA, body surface area; HR, heart rat; DBP, diastolic blood pressure; SDP, systolic blood pressureTable 2. Results from comparison of objective and subjective image quality of standard and sonic DL Cine imaging in children with challenging and non-challenging conditions.

	Children with non-challenging conditions				Children with challenging conditions
	Standard	Sonic DL	P		Standard	Sonic DL	P
Visual scores
Myocardial edge definition	4.6±0.5	4.5±0.5	0.161		3.2±1.1	4.5±0.6	< 0.001
Temporal fidelity	4.8±0.4	4.7±0.5	0.255		3.4±1.1	4.3±0.6	< 0.001
Artifact level	4.6±0.5	4.5±0.6	0.787		3.4±1.1	4.5±0.5	< 0.001
Overall image quality	4.7±0.4	4.6±0.4	0.163		3.3±1.0	4.4±0.5	< 0.001
Edge sharpness	0.29±0.08	0.27±0.06	0.097		0.19±0.08	0.25±0.07	< 0.001

Figure 1: Imaging examples of sonic DL and standard cine in an 11-year-old cooperative child with sinus rhythm (A), a 7-month-old child with sinus rhythm and poor breath-hold capacity (B), an 8-year-old child with arrhythmia and adequate breath-hold capacity(C), a 2-year-old child with arrhythmia and poor breath-hold capacity(D). Representative image of different phase and pixel profile of line through both ventricles plotted as function of time were shown from top to bottom.