College of Computer Science and Technology, China University of Petroleum (East China), Qingdao, 266580 Shandong, China.
Department of General Practice, Shandong Provincial Third Hospital, Shandong University, Jinan, 250031 Shandong, China.
Comput Math Methods Med. 2022 Mar 23;2022:3812509. doi: 10.1155/2022/3812509. eCollection 2022.
White matter hyperintensities (WMH), also known as white matter osteoporosis, have been clinically proven to be associated with cognitive decline, the risk of cerebral infarction, and dementia. The existing computer automatic measurement technology for the segmentation of patients' WMH does not have a good visualization and quantitative analysis. In this work, the author proposed a new WMH quantitative analysis and 3D reconstruction method for 3D reconstruction of high signal in white matter. At first, the author using ResUnet achieves the high signal segmentation of white matter and adds the attention mechanism into ResUnet to achieve more accurate segmentation. Afterwards, this paper used surface rendering to reconstruct the accurate segmentation results in 3D. Data experiments are conducted on the dataset collected from Shandong Province Third Hospital. After training, the Attention-Unet proposed in this paper is superior to other segmentation models in the segmentation of high signal in white matter and Dice coefficient and MPA reached 92.52% and 92.43%, respectively, thus achieving accurate 3D reconstruction and providing a new idea for quantitative analysis and 3D reconstruction of WMH.
脑白质高信号（WMH），又称脑白质疏松症，已被临床证明与认知能力下降、脑梗死风险和痴呆有关。现有的患者 WMH 分割的计算机自动测量技术缺乏良好的可视化和定量分析。在这项工作中，作者提出了一种新的 WMH 定量分析和 3D 重建方法，用于对脑白质高信号进行 3D 重建。首先，作者使用 ResUnet 实现脑白质高信号的分割，并将注意力机制添加到 ResUnet 中，以实现更准确的分割。之后，本文使用曲面渲染对准确的分割结果进行 3D 重建。数据实验是在山东省立第三医院采集的数据集上进行的。训练后，本文提出的 Attention-Unet 在脑白质高信号的分割和 Dice 系数、MPA 方面均优于其他分割模型，分别达到了 92.52%和 92.43%，从而实现了准确的 3D 重建，为 WMH 的定量分析和 3D 重建提供了新的思路。
