目的 观察基于深度学习重建(DLR)的超快速扫描方案用于颈椎MR扫描的可行性及其诊断价值.方法 前瞻性对36名受试者分别采用常规(扫描时间6 min 14 s)及超快速(2 min)扫描方案采集颈椎MR矢状位T1WI、脂肪抑制矢状位图像及轴位T2WI;以DLR方法重建超快速MRI.比较不同图像质量主、客观评价结果,以及观察者间诊断椎间盘退行性病变和椎间盘突出的一致性.结果 与常规图像相比,超快速-DLR图像伪影显著减少(P＜0.05).观察者间主观评估图像质量结果的一致性均为良好(Kappa均≥0.60).相比常规图像,超快速-DLR矢状位T1WI、T2WI及轴位T2WI中,脊髓、脑脊液(CSF)和椎体信噪比及脊髓/CSF对比度均显著提高(P均＜0.001).2名医师基于超快速-DLR和常规图像诊断椎间盘退行性病变的Kappa值分别为0.94、1.00,诊断椎间盘突出时分别为0.96、0.98.结论 相比常规扫描方案,超快速-DLR用于颈椎MR检查可缩短扫描时间,而所获图像质量及诊断效能与之相似.
Ultra-fast scanning scheme based on deep learning reconstruction for cervical MR examination
Objective To explore the feasibility and diagnostic value of ultra-fast scanning scheme based on deep learning-based reconstruction(DLR)for cervical MR examination.Methods Thirty-six subjects were prospectively enrolled and underwent both conventional scheme(scan time:6 min 14 s)and ultra-fast scheme(2 min)cervical spine MR scanning to acquire encompassing sagittal T1WI,sagittal adipose suppression T2WI and axial T2WI.The ultra-fast MRI were reconstructed using DLR method.The subjective and objective evaluations on imaging qualities of different MRIs were compared,along with the inter-observer agreement for diagnosing intervertebral disc degeneration and herniation.Results Compared with conventional MRI,artifacts in ultra-fast DLR images significantly reduced(P＜0.05).The subjective evaluation results of MRI had good agreement(all Kappa≥0.60).Compared with conventional MRI,the sagittal T1WI,T2WI and axial T2WI obtained with ultra-fast DLR showed significantly improved signal-to-noise ratio(SNR)of the spinal cord,cerebrospinal fluid(CSF)and vertebral body,as well as the spinal cord/CSF contrast(all P＜0.001).The Kappa value of 2 physicians for diagnosing intervertebral disc degeneration based on ultra-fast DLR and conventional scheme images was 0.94 and 1.00,respectively,of intervertebral disc herniation was 0.96 and 0.98,respectively.Conclusion Compared with conventional scanning scheme,using ultra-fast DLR scheme in cervical MR examination could shorten scanning time while achieve similar image quality and diagnostic accuracy.
NETL
NSTL
万方数据
