人工智能辅助压缩感知与并行采集技术在肩关节MRI中的对比研究

Comparative use of artificial intelligence-assisted compressed sensing and parallel imaging for shoulder magnetic resonance imaging

目的通过与并行采集(parallel imaging,PI)对比,探讨人工智能辅助压缩感知(artificial intelligence-assisted compressed sensing,ACS)技术对肩关节MRI扫描时间和图像质量的影响,并优化扫描方案。材料与方法前瞻性纳入2023年11月至2024年2月在我院行肩关节MRI检查的70例患者,扫描序列采用快速自旋回波序列包括斜冠状位T1加权成像(oblique coronal T1-weighted,OCor T1WI)、斜冠状位T2加权频率选择脂肪抑制成像(oblique coronal T2-weighted with fat saturation,OCor T2WI-fs)、斜矢状位质子密度(proton density,PD)加权频率选择脂肪抑制成像(oblique sagittal PD-weighted with fat saturation,OSag PDWI-fs)、横断面PD加权频率选择脂肪抑制成像(transverse PD-weighted with fat saturation,Tra PDWI-fs),分别采用ACS和PI两种加速采集技术。比较两种技术的扫描时间。测量冈上肌肌腹和肱骨头的信号强度及背景标准差,并计算信噪比(signal-to-noise ratio,SNR)。采用李克特量表对图像质量进行评分。结果相较于PI,采用ACS缩短了33.5%的扫描时间。采用ACS采集的图像伪影更少,骨骼肌肉的噪声更小,在图像质量主观评分上均高于采用PI的图像,差异均有统计学意义(P均<0.05)。OCor T1WI、OCor T2WI-fs和Tra PDWI-fs序列中采用ACS的图像在冈上肌和肱骨头的SNR均高于采用PI的图像,差异均有统计学意义(P均<0.001)。OSag PDWI-fs序列中图像冈上肌的SNR采用ACS与PI差异无统计学意义(P>0.05),图像肱骨头的SNR采用ACS采集的图像高于PI的图像,差异有统计学意义(P均<0.001)。结论与传统的PI相比,采用ACS在肩关节MRI中可实现更高效且稳定的快速成像方案,提高图像质量,缩短扫描时间,提高患者耐受程度,具有较好的临床应用价值。

Objective:By comparing with parallel imaging(PI),to explore the impact of artificial intelligence-assisted compressed sensing(ACS)technology on the scanning time and image quality of shoulder joint MRI,and optimizes the scanning scheme.Materials and Methods:A total of 70 patients who underwent shoulder MRI in our hospital from November 2023 to February 2024 were prospectively enrolled.The scanning sequences used fast spin echo including oblique coronal T1-weighted(OCor T1WI),oblique coronal T2-weighted with fat saturation(OCor T2WI-fs),oblique sagittal proton density(PD)-weighted with fat saturation(OSag PDWI-fs),and transverse PD-weighted with fat saturation(Tra PDWI-fs),respectively,using two accelerated acquisition technologies:ACS and PI.Compare the scanning time of two technologies.Measure the signal intensity and background standard deviation of the supraspinatus muscle and humeral head,and calculate the signal-to-noise ratio(SNR).Use the Likert scale to rate image quality.Results:Compared to PI,using ACS reduced scanning time by 33.5%.The images obtained using ACS have few artifacts and low noise.The subjective image quality scores are higher than those obtained using PI,and the differences are statistically significant(all P<0.05).The SNR of images using ACS in OCor T1WI,OCor T2WI-fs,and Tra PDWI-fs sequences were higher than those using PI in the supraspinatus muscle and humeral head,and the differences were statistically significant(all P<0.001).The SNR of the supraspinatus muscle in the OSag PDWI-fs sequence using ACS was not significantly different from that of PI(P>0.05),while the SNR of the humeral head in the images obtained using ACS was higher than that of PI,and the difference was statistically significant(all P<0.001).Conclusions:Compared with PI,using ACS in shoulder MRI can achieve a more efficient and stable rapid imaging,improve image quality,shorten scanning time,and increase patient tolerance,which has clinical application value.