高加速因子结合深度学习重建在肩关节MRI中的应用价值OA
Value of high acceleration factor combined with deep learning reconstruction in shoulder MRI
目的:探讨高加速因子结合深度学习重建(deep learning reconstruction,DLR)序列在3.0T肩关节MRI中的应用价值.方法:前瞻性纳入2023年2-7月于某医院行肩关节MRI检查的54例患者(共56例次检查).所有患者均使用冠状位脂肪抑制T2加权快速自旋回波(T2-fs-cor)序列和轴位脂肪抑制质子密度加权快速自旋回波(PD-fs-axi)序列完成扫描,并采用常规重建算法和DLR算法进行图像重建,据此将图像分为常规组和DLR组.常规组的广义自动校准部分并行采集(generalized autocalibrating partially parallel acquisitions,GRAPPA)加速因子为2,DLR组的GRAPPA加速因子为3.采用配对样本t检验或配对样本Wilcoxon符号秩和检验比较2组图像客观评价结果的差异,采用卡方检验比较2组图像主观评价结果的差异,采用Kappa检验进行一致性分析.结果:DLR组的信噪比和对比噪声比均优于常规组,常规组的伪影出现率(21.43%)高于DLR组(8.04%),差异有统计学意义(P均<0.01).阅片者对2组图像间的整体图像质量评分、解剖结构评分一致性均为优,常规组阅片者间对不同病变的诊断一致性为优,DLR组阅片者对不同病变的诊断一致性也为优,Kappa值均>0.8;DLR组的整体图像质量评分和解剖结构评分结果均优于常规组,差异有统计学意义(P均<0.01).结论:与常规肩关节MRI序列相比,高加速因子结合DLR序列能够提供更优的图像质量,在疾病评估方面与常规序列具有高度的一致性.
Objective To explore the value of high acceleration factor combined with deep learning reconstruction(DLR)for 3.0T shoulder MRI.Methods Fifty-four patients who underwent shoulder MRI at a hospital from February to July 2023 were prospectively included(56 examinations in total),who were scanned with a coronal fat-suppressed T2-weighted fast spin echo(T2-fs-cor)sequence and an axial fat-suppressed proton density-weighted fast spin echo(PD-fs-axi)sequence,and then divided into a conventional group with routine image reconstruction and a DLR group with DLR algorithm-based reconstruction.The acceleration factor for generalized autocalibrating partially parallel acquisitions(GRAPPA)in the conventional group was 2,and that for the DLR group was 3.The two groups had the objective evaluation results compared by the paired-sam-ples t-test or paired-samples Wilcoxon signed rank test,the subjective evaluation results compared by the Chi-square test and the consistency analysis carried out with the Kappa test.Results The DLR group behaved better than the conventional group in terms of the signal-to-noise ratio and contrast-to-noise ratio,and the rate of artifacts in the con-ventional group(21.43%)was higher than that in the DLR group(8.04%),with the differences being statistically significant(all P<0.01).The two groups both had excellent inter-observer agreement for overall image quality and anatomical structure scoring,and excellent inter-observer diagnostic concordance for various lesions was found in each of the two groups,with all the Kappa values higher than 0.8;the DLR group gained advantages over the conventional group in overall image quality and anatomical structure scoring,with the differences being statistically significant(all P<0.01).Conclusion The sequence of the high acceleration factor combined with DLR contributes to higher image quality when compared with the conventional shoul-der MRI sequences,which has high consistency with conventional sequences in disease evaluation.[Chinese Medical Equipment Journal,2026,47(9):57-62]
李青;曹达;徐露露;祁良;郝绍伟;胡梅
南京医科大学第一附属医院放射科,南京 210029南京医科大学第一附属医院放射科,南京 210029南京医科大学第一附属医院放射科,南京 210029南京医科大学第一附属医院放射科,南京 210029西门子数字医疗科技(上海)有限公司应用培训部,上海 200126南京医科大学第一附属医院放射科,南京 210029
医药卫生
肩关节磁共振成像深度学习重建图像质量加速因子
shoulder jointMRIdeep learning reconstructionimage qualityacceleration factor
《医疗卫生装备》 2026 (9)
57-62,6
国家自然科学基金项目(81701652)
评论