深度学习重建在MRCP中对IPMN评估及胆总管结石检出的影响
本单中心回顾性研究纳入91例接受MRCP的患者,比较深度学习与常规重建在IPMN评估及胆总管结石检出中的价值。结果显示深度学习重建显著提高信噪比、对比度及对比噪声比,并改善图像质量主观评分。深度学习重建可提升对IPMN分型、恶性征象及胆总管结石的诊断信心。
要点
- 深度学习重建在2D MRCP序列中对胆总管信噪比、对比度及对比噪声比的改善最显著。
- 深度学习重建提升对IPMN分型、恶性征象及胆总管结石的诊断信心。
- 深度学习重建对胰管的图像质量改善较为有限且不稳定。
核心结论
- 深度学习重建可改善MRCP图像质量,并提升对IPMN及胆总管结石的诊断信心。
- 结果支持将深度学习技术整合到常规MRCP临床流程中。
研究设计与样本
- 单中心回顾性研究,纳入2024年3月至2025年1月期间91例接受MRCP的患者。
- 51例因IPMN评估入组,40例因可疑胆总管结石入组。
- 分析4种序列类型,包括2D与3D、有和无深度学习重建。
- 客观指标包括信噪比、对比度及对比噪声比;由两名放射科医师独立评估主观图像质量及诊断信心评分。
关键影像表现/征象
- 深度学习重建可明显减少噪声和伪影,改善胆总管的主观对比度表现。
- 深度学习重建对胆总管的信噪比、对比度及对比噪声比的改善在2D序列上更为显著。
- 深度学习重建在胰管区域的图像质量改善有限且各序列间表现不一致。
诊断标准与关键数值
- 主观诊断信心评分涉及IPMN分型、恶性征象及胆总管结石检出。
- 图像质量主观评分涵盖整体图像质量、伪影、噪声和对比度。
临床应用与阅片要点
- 2D序列可优先考虑应用深度学习重建以获得胆总管更优的信噪比与对比度。
- IPMN随访和胆总管结石可疑病例可借助深度学习重建提高报告诊断信心。
- 胰管细节评估时需注意深度学习重建的改善幅度有限。
局限性
- 单中心回顾性设计。
- 需要前瞻性研究及标准操作流程来评价深度学习重建对诊断准确性和患者结局的直接影响。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
BACKGROUND: Magnetic Resonance Cholangiopancreatography (MRCP) is the gold standard for evaluating ductal pathologies, particularly Intraductal Papillary Mucinous Neoplasms (IPMN) and common bile duct (CBD) stones. However, image quality can be limited by noise, artifacts, and acquisition constraints. Deep learning (DL) reconstruction algorithms may improve image quality and diagnostic confidence. This study aims to compare DL and non-DL reconstructions in MRCP regarding diagnostic confidence for IPMN evaluation and CBD stone detection.
METHODS: This single-center retrospective study included 91 patients who underwent MRCP between March 2024 and January 2025 for IPMN assessment (51 patients) or suspected CBD stones (40 patients). Four sequence types were analyzed (2D and 3D, with and without DL reconstruction). Image quality was objectively assessed (SNR, CR, CNR). Subjective criteria (overall image quality, artifacts, noise, contrast) and diagnostic confidence scores for specific IPMN and CBD stone features were independently evaluated by two radiologists.
RESULTS: DL reconstruction significantly improved SNR, contrast, and CNR for the CBD, particularly in 2D sequences. For the pancreatic duct, improvement was more limited and variable. Subjectively, DL images received higher ratings, notably for contrast, with significant noise reduction and better overall quality. Diagnostic confidence was enhanced for IPMN typing, malignancy criteria, and CBD stone detection.
CONCLUSIONS: Deep learning reconstruction improves MRCP image quality and reader diagnostic confidence for evaluating IPMNs and CBD stones. While prospective studies with procedural standards are needed to evaluate direct impacts on diagnostic accuracy and patient outcomes, these findings support integrating DL technology into routine clinical protocols.
引用
- Raynal, Pereira, Hordonneau, et al. Deep learning reconstruction in MRCP: Impact on IPMN characterization and common bile duct stone detection. Eur J Radiol. 2026. PMID: 42561893. doi:10.1016/j.ejrad.2026.113125
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。