Scholarlink

Background <p>Balanced steady-state free-precession (bSSFP) cine imaging is the clinical standard for ventricular function assessment but requires multiple breath-holds, which can be challenging for pediatric and young adult patients. Deep learning (DL)–accelerated cine imaging offers the potential to reduce breath-hold burden, shorten scan time, and enable free-breathing acquisitions.</p> Objective <p>To clinically evaluate a commercially available DL-accelerated cine bSSFP sequence (SonicDL) across multiple acceleration levels and breath-hold/free-breathing configurations, and to assess diagnostic image quality, ventricular volumetric accuracy, and scan time reductions.</p> Materials and methods <p>This retrospective study included 25 patients with pectus excavatum and 15 with cardiomyopathy who underwent conventional cine imaging with breath-hold duration of nine cardiac-cycle interval (9-RR) and SonicDL acquisitions at 4-RR breath-hold, 1-RR breath-hold, and 1-RR free breathing. Diagnostic image quality was independently scored by three expert readers using a 5-point scale. Automated DL-based segmentation provided biventricular volumetric indices, with phase-contrast flow serving as the physiological reference for stroke volume. Statistical analysis included repeated-measures ANOVA, paired tests, ICCs, and Bland–Altman analysis.</p> Results <p>SonicDL significantly reduced scan time 57% (4-RR breath-hold), 79% (1-RR breath-hold), and 87% (1-RR free breathing) compared with 9-RR breath-hold imaging. Diagnostic image quality was highest for 4-RR breath-hold (median 4.50), significantly exceeding other protocols (<i>P</i>&lt;0.0001). Across all SonicDL protocols, volumetric indices showed small biases (&lt;2&#xa0;mL/m<sup>2</sup> for most parameters) with limits of agreement typically within ±6&#xa0;mL/m<sup>2</sup>. Stroke volumes showed close agreement with phase-contrast flow (biases 1-5&#xa0;mL/m<sup>2</sup>) across all protocols. In cardiomyopathy patients, diagnostic image quality was lower overall, but volumetric indices agreement remained within clinically acceptable ranges.</p> Conclusion <p>SonicDL cine bSSFP imaging substantially reduces breath-hold burden and scan time while maintaining diagnostic image quality and close agreement with conventional cine and physiologic flow measurements. The 4-RR breath-hold protocol provided the most favorable balance of acceleration and fidelity, while free-breathing acquisitions offered a practical alternative for patients with limited breath-hold capacity.</p> Graphical abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Clinical evaluation of accelerated breath-held and free-breathing cine cardiac MRI using model-based deep learning reconstruction (SonicDL) in children and young adults

  • 期刊论文

  • 发表日期：

    2026年5月28日

查看全文
  • Murat Kocaoglu,
  • Hieu Ta,
  • Sean M. Lang,
  • Cara E. Morin,
  • Amol Pednekar

摘要

Background

Balanced steady-state free-precession (bSSFP) cine imaging is the clinical standard for ventricular function assessment but requires multiple breath-holds, which can be challenging for pediatric and young adult patients. Deep learning (DL)–accelerated cine imaging offers the potential to reduce breath-hold burden, shorten scan time, and enable free-breathing acquisitions.

Objective

To clinically evaluate a commercially available DL-accelerated cine bSSFP sequence (SonicDL) across multiple acceleration levels and breath-hold/free-breathing configurations, and to assess diagnostic image quality, ventricular volumetric accuracy, and scan time reductions.

Materials and methods

This retrospective study included 25 patients with pectus excavatum and 15 with cardiomyopathy who underwent conventional cine imaging with breath-hold duration of nine cardiac-cycle interval (9-RR) and SonicDL acquisitions at 4-RR breath-hold, 1-RR breath-hold, and 1-RR free breathing. Diagnostic image quality was independently scored by three expert readers using a 5-point scale. Automated DL-based segmentation provided biventricular volumetric indices, with phase-contrast flow serving as the physiological reference for stroke volume. Statistical analysis included repeated-measures ANOVA, paired tests, ICCs, and Bland–Altman analysis.

Results

SonicDL significantly reduced scan time 57% (4-RR breath-hold), 79% (1-RR breath-hold), and 87% (1-RR free breathing) compared with 9-RR breath-hold imaging. Diagnostic image quality was highest for 4-RR breath-hold (median 4.50), significantly exceeding other protocols (P<0.0001). Across all SonicDL protocols, volumetric indices showed small biases (<2 mL/m2 for most parameters) with limits of agreement typically within ±6 mL/m2. Stroke volumes showed close agreement with phase-contrast flow (biases 1-5 mL/m2) across all protocols. In cardiomyopathy patients, diagnostic image quality was lower overall, but volumetric indices agreement remained within clinically acceptable ranges.

Conclusion

SonicDL cine bSSFP imaging substantially reduces breath-hold burden and scan time while maintaining diagnostic image quality and close agreement with conventional cine and physiologic flow measurements. The 4-RR breath-hold protocol provided the most favorable balance of acceleration and fidelity, while free-breathing acquisitions offered a practical alternative for patients with limited breath-hold capacity.

Graphical abstract