Super-resolution musculoskeletal MRI using deep learning.
Super-resolution musculoskeletal MRI using deep learning.
复制标题
DOI：
10.1002/mrm.27178
复制
发表时间：
2018-11
影响因子：
3.3
通讯作者：
Hargreaves BA
中科院分区：
文献类型：
--
作者：
Chaudhari AS;Fang Z;Kogan F;Wood J;Stevens KJ;Gibbons EK;Lee JH;Gold GE;Hargreaves BA
开发一种使用卷积神经网络的超分辨率技术，用于从较厚的输入切片生成薄层膝关节MR图像，并将该方法与替代的通过平面插值方法进行比较。我们实现了一个名为DeepResolve的3D卷积神经网络，以学习相同中心位置的高分辨率薄切片图像和低分辨率厚切片图像之间基于残差的转换。DeepResolve使用124个切片厚度为0.7 mm的双回波稳态（DESS）数据集进行训练，并在17名患者身上进行测试。将地面实况图像与DeepResolve、临床使用的三次插值（TCI）和傅立叶插值（FI）方法沿着最先进的单图像稀疏编码超分辨率（ScSR）进行了比较。使用结构相似性（SSIM）、峰值信噪比（pSNR）和均方根误差（RMSE）图像质量指标对多种薄片下采样因子（DSFs）进行比较。两名肌肉骨骼放射科医生对三个数据集进行了排名，并审查了DeepResolve、TCI和地面实况图像的诊断质量，包括清晰度、对比度、伪影、信噪比和整体诊断质量。Mann-Whitney U检验评价了定量图像指标、阅片者评分和排名之间的差异。Cohen’s Kappa（κ）评价阅片者间可靠性。对于所有DSFs，DeepResolve的SSIM、pSNR和RMSE均显著优于TCI、FI和ScSR（P<0.05，4x和8x ScSR DSFs除外）。在阅片师研究中，DeepResolve在所有图像质量类别和整体图像排名方面均显著优于（P<0.01）TCI。两名阅片员的评分基本一致（κ=0.73）。DeepResolve能够从较低分辨率的较厚切片中分辨出高分辨率的膝关节薄层MRI，实现了比传统方法和最先进方法更优越的上级定量和定性诊断性能。
To develop a super-resolution technique using convolutional neural networks for generating thin-slice knee MR images from thicker input slices, and to compare this method to alternative through-plane interpolation methods. We implemented a 3D convolutional neural network entitled DeepResolve to learn residual-based transformations between high-resolution thin-slice images and lower-resolution thick-slice images at the same center locations. DeepResolve was trained using 124 double-echo in steady-state (DESS) datasets with 0.7mm slice thickness and tested on 17 patients. Ground-truth images were compared to DeepResolve, clinically utilized tricubic interpolation (TCI) and Fourier interpolation (FI) methods, along with state-of-the-art single image sparse-coding super-resolution (ScSR). Comparisons were performed using structural similarity (SSIM), peak signal-to-noise ratio (pSNR), and root-mean-square-errors (RMSE) image quality metrics for a multitude of thin-slice downsampling factors (DSFs). Two musculoskeletal radiologists ranked the three datasets and reviewed the diagnostic quality of the DeepResolve, TCI, and ground-truth images for sharpness, contrast, artifacts, signal-to-noise ratio, and overall diagnostic quality. Mann-Whitney U-Tests evaluated differences between the quantitative image metrics, reader scores, and rankings. Cohen’s Kappa (κ) evaluated inter-reader reliability. DeepResolve had significantly better SSIM, pSNR, and RMSE than TCI, FI, and ScSR for all DSFs (P<0.05, except 4x and 8x ScSR DSFs). In the reader study, DeepResolve significantly outperformed (P<0.01) TCI in all image quality categories and overall image ranking. Both readers had substantial scoring agreement (κ=0.73). DeepResolve was capable of resolving high-resolution thin-slice knee MRI from lower-resolution thicker slices, achieving superior quantitative and qualitative diagnostic performance to both conventionally utilized and state-of-the-art methods.
登录
查看更多内容
DOI：
10.1002/jmri.25508
发表时间：
2017-06
期刊：
Journal of magnetic resonance imaging : JMRI
影响因子：
--
作者：
Bao S;Tamir JI;Young JL;Tariq U;Uecker M;Lai P;Chen W;Lustig M;Vasanawala SS
通讯作者：
Vasanawala SS
影响因子：
19.7
作者：
Duc, Sylvain R.;Pfirrmann, Christian W. A.;Hodler, Juerg
通讯作者：
Hodler, Juerg
影响因子：
3.3
作者：
Hammernik K;Klatzer T;Kobler E;Recht MP;Sodickson DK;Pock T;Knoll F
通讯作者：
Knoll F
影响因子：
27.4
作者：
KELLGREN, JH;LAWRENCE, JS
通讯作者：
LAWRENCE, JS
DOI：
10.1002/jmri.25507
发表时间：
2017-06
期刊：
Journal of magnetic resonance imaging : JMRI
影响因子：
--
作者：
Kijowski R;Rosas H;Samsonov A;King K;Peters R;Liu F
通讯作者：
Liu F