| 摘 要: 扩散峰度成像是一种可以更精确地描述脑组织中水分子非高斯扩散特征的无创技术,然而其弊端是需要长时间地采集扩散加权成像图像(Diffusion Weighted Image,DWI)。针对这一弊端,开发了一种基于深度学习的超快速、高鲁棒性的重建方法。不同于传统的需要大量扩散加权成像图像的模型拟合方法,该方法仅使用少量的扩散加权成像图像就能估计出高质量的扩散峰度成像特征图,并且各项图像评估指标都获得了最佳值。为获取高质量的特征图,传统模型拟合方法至少需要61张DWI图像,而本方法仅用了8张 DWI,大幅(成数量级)减少了扫描次数和扫描时间,有望拓宽扩散峰度成像的应用范围。 |
| 关键词: 扩散峰度成像;扩散加权图像;模型拟合;深度学习 |
|
中图分类号: TP391.5
文献标识码: A
|
| 基金项目: 国家自然科学基金项目(42004105) |
|
| Deep Learning-based Method for Accelerated Acquisition of High-Quality Diffusion Kurtosis Imaging |
|
CHEN Rui, LI Rong, JIA Zijian
|
(School of Health Science and Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China)
ruichenusst2022@163.com; lirong8882022@163.com; jiazijian1989@usst.edu.cn
|
| Abstract: Diffusion Kurtosis Imaging (DKI) is a non-invasive technique that more accurately characterizes the non-Gaussian diffusion properties of water molecules in brain tissue. However, its drawback lies in the prolonged acquisition time required for Diffusion-Weighted Imaging (DWI) scans. To address this issue, a deep learning-based ultra-fast and highly robust reconstruction method has been developed. Unlike traditional mode-l fitting approaches that demand a large number of DWI images, this method estimates high-quality DKI parametric maps using only a minimal set of DWI images while achieving optimal values across various image evaluation metrics. Conventional mode-l fitting methods require at least 61 DWI images to obtain high-quality parametric maps, whereas the proposed method uses only 8 DWI images, significantly (by an order of magnitude) reducing scan repetitions and acquisition time. This breakthrough holds promise for expanding the clinical and research applications of DKI. |
| Keywords: Diffusion Kurtosis Imaging (DKI); Diffusion-Weighted Imaging (DWI); model fitting; deep learning |
