Center for Biomedical Imaging, Department of Radiology, New York University Grossman School of Medicine, New York, New York, USA.
Vilcek Institute of Graduate Biomedical Sciences, NYU Langone Health, New York, New York, USA.
NMR Biomed. 2023 Oct;36(10):e4959. doi: 10.1002/nbm.4959. Epub 2023 May 18.
In this work, we introduce a super-resolution method that generates a high-resolution (HR) sodium ( Na) image from simultaneously acquired low-resolution (LR) Na density-weighted MRI and HR proton density, T , and T maps from proton ( H) MR fingerprinting in the brain at 7 T. The core of our method is a partial least squares regression between the HR ( H) images and the LR ( Na) image. An iterative loop and deconvolution with the point spread function of each acquired image were included in the algorithm to generate a final HR Na image without losing features from the LR Na image. The method was applied to simultaneously acquired HR proton and LR sodium data with in-plane resolution ratios between sodium and proton data of 3.8 and 1.9 and the same slice thickness. Four volunteers were scanned to evaluate the method's performance. For the data with a resolution ratio of 3.8, the mean absolute difference between the generated and ground truth HR Na images was in the range of 1.5%-7.2% of the ground truth with a multiscale structural similarity index (M-SSIM) of 0.93 ± 0.03. For the data with a resolution ratio of 1.9, the mean absolute difference was in the range of 4.8%-6.3% with an M-SSIM of 0.95 ± 0.01.
在这项工作中，我们介绍了一种从同时采集的低分辨率 (LR) 钠 ( Na) 密度加权 MRI 和质子 ( H) MR 指纹成像在 7T 大脑中获得的高分辨率 (HR) 质子密度、T 和 T 图中生成高分辨率 (HR) Na 图像的方法。我们方法的核心是 HR ( H) 图像和 LR ( Na) 图像之间的偏最小二乘回归。该算法包括一个迭代循环和每个采集图像的点扩散函数的反卷积，以生成最终的 HR Na 图像，而不会丢失 LR Na 图像中的特征。该方法应用于同时采集的 HR 质子和 LR 钠数据，钠和质子数据的平面内分辨率比为 3.8 和 1.9，且切片厚度相同。对四名志愿者进行了扫描以评估该方法的性能。对于分辨率比为 3.8 的数据，生成的 HR Na 图像与真实 HR Na 图像之间的平均绝对差值在真实 HR Na 图像的 1.5%-7.2%范围内，多尺度结构相似性指数 (M-SSIM) 为 0.93 ± 0.03。对于分辨率比为 1.9 的数据，平均绝对差值在真实 HR Na 图像的 4.8%-6.3%范围内，M-SSIM 为 0.95 ± 0.01。
