摘要
目的探讨深度学习重建算法对腰椎MRI的降噪效果及其临床应用中的可行性。方法前瞻性地纳入临床要求行腰椎MR检查的患者53例。使用1.5T MR及脊柱线圈单元行矢状位T1加权成像(T1-Weighted Image,T1WI)、T2加权脂肪抑制成像(T2-Weighted Fat Suppressed Image,T2WI-FS)、短时间反转恢复(Short Time Inversion Recovery,STIR)序列和横断位T2WI序列扫描,获得原始图像(A组),利用常规滤波重建获得图像(B组)和深度学习重建获得图像(C组)。使用定量指标峰值信噪比(Peak Signal to Noise Ratio,PSNR)和图像锐利度评估各序列3组图像的SNR和锐利度。由两位放射医师对图像的整体质量、噪声、对比度和伪影进行评分和分析,两组评分进行一致性检验。结果与A组和B组相比,C组所有腰椎序列的PSNR值均升高(P值均<0.05),T1WI、STIR和横断位T2WI的图像锐利度值升高(P值均<0.05)。C组的T1WI的噪声、对比度分辨率和伪影以及其他序列的整体图像质量、噪声、对比度分辨率和伪影的主观评分均明显高于A、B两组(P值均<0.05)。结论深度学习重建技术在腰椎常规序列图像中的降噪效果良好,提高了图像质量,具有一定的临床应用价值。
Objective To evaluate the image denoising effect in lumbar spine MRI based on deep learning reconstruction algorithm and explore its feasibility in clinical application.Methods A total of 53 patients requiring lumbar MR examination were prospectively included.Each patient underwent sagittal T1-weighted image(T1WI),T2-weighted fat suppression imaging(T2WI-FS),short time inversion recovery(STIR)sequence and transverse T2WI sequence scanning for obtaining the original images(Group A).By using conventional filter reconstruction and Deep Recon technology to reconstruct the images of each sequence,we obtained the conventional filter reconstruction images(Group B)and Deep Recon images(Group C).The quantitative indicators peak signalto-noise ratio(PSNR)and image sharpness were used to evaluate the SNR and sharpness of the three groups of images in each sequence.Two radiologists scored and analyzed the overall quality,noise,contrast resolution,and artifacts of the image.Finally,the scores of the two sets were tested for consistency.Results Compared with Group A and Group B,the PSNR values of all lumbar spine sequences in Group C were increased(P<0.05),and the image sharpness values of T1WI,STIR and transverse T2WI were increased(P<0.05).The subjective evaluation of the image showed that the noise,contrast resolution and artifacts on T1WI of the Group C,as well as the overall image quality,noise,contrast resolution and artifacts of other sequences in Group C were significantly higher than those in groups A and B(P<0.05).Conclusion Deep Recon has good denoising effect in conventional sequence images of lumbar spine,and can improve the image quality,which has certain clinical application value.
作者
张雨
徐旭
肖奕
曾令明
曾文
夏春潮
张树恒
罗曼
李真林
ZHANG Yu;XU Xu;XIAO Yi;ZENG Lingming;ZENG Wen;XIA Chunchao;ZHANG Shuheng;LUO Man;LI Zhenlin(Department of Radiology,West China Hospital,Sichuan University,Chengdu Sichuan 610041,China;Shanghai United Imaging Intelligence Co.,Ltd.,Shanghai 201800,China)
出处
《中国医疗设备》
2021年第10期19-23,共5页
China Medical Devices
基金
四川大学华西医院学科卓越发展135工程项目(ZYGD18019)。
