Deep Learning-Based Breath-Hold and Free-Breathing Cine MRI for Comprehensive Cardiac Evaluation.

Abstract

Objective: To evaluate and compare scan times, measurement accuracy, and image quality (IQ) of free-breathing (FB) and breath-hold (BH) deep learning (DL) cine MRI sequences versus standard cine MRI, with a specific focus on patients with arrhythmia and dyspnea.

Materials and methods: Seventy participants were prospectively enrolled, including 24 with arrhythmia, 17 with dyspnea, and 29 with normal sinus rhythm and eupnea (mean age, 49 ± 17 years). Each patient underwent three cine MRI acquisitions (standard cine, BHDL, and FBDL) on a 3T scanner. Quantitative assessments of biventricular function, left ventricular mass, and myocardial strain were independently performed by three radiologists, blinded to image acquisition techniques. IQ was evaluated by the same readers using both a five-point Likert scale and objective metrics.

Results: Both BHDL and FBDL significantly reduced total examination times compared to standard cine (BHDL: 58 ± 5 s; FBDL: 88 ± 12 s; standard cine: 208 ± 12 s; adjusted P < 0.001). Quantitative measurements from BHDL and FBDL showed no statistically significant differences compared to standard cine and showed strong correlations (correlation coefficients > 0.85) with standard cine. BHDL consistently demonstrated narrower 95% limits of agreement (LOA) than FBDL across all parameters. For BHDL, the 95% LOA for left and right ventricular ejection fractions were -3.5% to 3.9% and -3.4% to 4.0%, respectively; for FBDL, they were -4.6% to 5.8% and -7.8% to 9.3%, respectively. In patients with arrhythmia, BHDL achieved significantly higher IQ Likert scores (4.44 ± 0.56) than both standard cine (4.00 ± 0.99; adjusted P = 0.043) and FBDL (3.94 ± 0.56; adjusted P = 0.030). In patients with dyspnea, FBDL received the highest IQ scores (4.24 ± 0.47), outperforming standard cine (3.41 ± 0.97; adjusted P = 0.028) and BHDL (3.68 ± 0.56; adjusted P = 0.028).

Conclusion: Both FBDL and BHDL significantly reduced scan times compared to standard cine without compromising quantitative measurement accuracy. BHDL offered superior measurement accuracy and shorter scan time than FBDL. Furthermore, BHDL demonstrated robust suitability for patients with arrhythmia by minimizing arrhythmia-related artifacts, whereas FBDL was more effective in patients with dyspnea by mitigating respiratory motion artifacts.