摘要目的:探讨基于深度学习（DL）结合压缩感知（CS）技术在屏气三维磁共振胰胆管成像（3D MRCP）图像质量改善的可行性。方法:前瞻性纳入2023年3至8月在四川大学华西医院行上腹部3D MRCP检查的患者68例，3D MRCP序列包括CS组，加速因子为24（CS-24）；DL-CS组，加速因子分别为24（DL-CS-24）、33（DL-CS-33）。测量3组图像的信噪比（SNR）、对比度（CR）和对比噪声比（CNR），并对图像的整体图像质量、背景抑制、伪影、各级胆管及胰管可见性进行主观评价。采用卡方检验和Friedman检验分别对3组序列图像中不满足诊断图像数量与主客观指标进行统计学分析。结果:DL-CS-33组的扫描时间为9 s，较CS-24组和DL-CS-24组的扫描时间13 s缩短约31%。DL-CS-33组68例患者的图像均满足临床诊断要求，CS-24组7例患者的图像不满足诊断要求，DL-CS-24组4例患者的图像不满足诊断要求，3组间差异均有统计学意义（ P<0.05）。DL-CS-33组和CS-24组的SNR、CR、CNR，整体图像质量、伪影、各级胆管和胰管的可见性评分差异均无统计学意义（ P均>0.05）。DL-CS-24组的SNR、CR、CNR，肝内胆管、主胰管以及整体图像质量均好于CS-24组，差异均有统计学意义（ P均<0.05）。 结论:当加速因子均为24时，与CS技术相比，应用DL-CS技术在不改变图像扫描时间的情况下，改善了屏气3D MRCP图像质量。DL-CS技术结合高加速因子33，在确保整体图像质量的同时进一步降低了扫描时间，可提供快速的屏气扫描方案。

abstractsObjective:To explore the improvement of image quality of different acceleration factors in breath-hold three-dimensional magnetic resonance cholangiopancreatography (3D MRCP) using deep learning (DL) and compressed sensing (CS) technology.Methods:A total of 68 patients who underwent upper abdominal 3D MRCP examination at West China Hospital of Sichuan University from March to August 2023 were prospectively included. The patients were subdivided into three groups randomly with the following paramters: CS group with an acceleration factor of 24 (CS-24); DL-CS group with acceleration factors 24 (DL-CS-24) and 33 (DL-CS-33) respectively. The signal-to-noise ratio (SNR), contrast ratio (CR) and contrast-to-noise ratio (CNR) of the three sets of images were measured, and the overall image quality, background suppression, artifacts, and visibility of bile ducts and pancreatic ducts at all levels were subjectively evaluated. Chi-square test and Friedman test were used to perform statistical analysis on the number of unsatisfactory diagnostic images and subjective and objective indicators of the three groups of sequences respectively.Results:The scanning time of the DL-CS-33 group (9 s) was 30% shorter than that of the CS-24 group and DL-CS-24 group (13s). The images of DL-CS-33 group from 68 patients all met the clinical diagnostic requirements and statistically differences were found between the images from CS-24 group and DL-CS-24 group (all P<0.05). There were no statistically differences in SNR, CR, CNR, overall image quality, artifacts, and visibility scores of bile ducts and pancreatic ducts at all levels between the DL-CS-33 group and the CS-24 group (all P>0.05). The SNR, CR, CNR, intrahepatic bile duct, main pancreatic duct and overall image quality of the DL-CS -24 group were better than those of the CS-24 group (all P<0.05). Conclusions:DL-CS technology could improves breath-hold 3D MRCP image quality with the 24 acceleration factor with no additioanl scanning time. DL-CS technology combined with a high acceleration factor of 33 further reduces scanning time while ensuring overall image quality, providing a fast breath-hold scanning solution.