基于移位窗口自注意力机制的新生儿脑区域图像分割  

Image Region Segmentation of Neonatal Brain Based on Self-attention Mechanism of Shifted Windows

作　　者：张啸成 王涛 田昕[3] 张永刚[2,4] ZHANG Xiaocheng;WANG Tao;TIAN Xin;ZHANG Yonggang(College of Software,Jilin University,Changchun 130012,China;Key Laboratory of Symbol Computation and Knowledge Engineering of Ministry of Education,Jilin University,Changchun 130012,China;The Second Hospital of Jilin University,Changchun 130062,China;College of Computer Science and Technology,Jilin University,Changchun 130012,China)

机构地区：[1]吉林大学软件学院,长春130012 [2]吉林大学符号计算与知识工程教育部重点实验室,长春130012 [3]吉林大学第二医院,长春130062 [4]吉林大学计算机科学与技术学院,长春130012

出　　处：《吉林大学学报（理学版）》2024年第5期1129-1137,共9页Journal of Jilin University:Science Edition

基　　金：吉林省自然科学基金(批准号:20200201447JC)。

摘　　要：通过改进Swin Transformer编码解码网络,结合跳跃连接和深度监督机制,针对新生儿脑部核磁共振成像(MRI)图像分割面临的低信噪比、组织对比度差等问题,提出一种新的基于移位窗口自注意力机制的新生儿脑区域图像分割方法.该方法经过对MRI图像的预处理后,能实现新生儿脑部图像多功能区域的精确分割,并采用最大连通域算法进一步提高了分割精度.在数据集dHCP上的实验结果表明,该方法优于现有方法,为早期发现和干预新生儿脑部损伤提供了潜在的可能性.By improving the Swin Transformer coding and decoding network,combined with the skip-linking and depth supervision mechanisms,we proposd a new image region segmentation method of neonatal brain based on self-attention mechanism of shifted windows to address the issues of low signal-to-noise ratio and poor tissue contrast in segmentation of nuclear magnetic resonance imaging(MRI)images of the neonatal brain.The method could achieve accurate segmentation of multifunctional regions of the neonatal brain images after preprocessing the MRI images,and further improve the segmentation accuracy by using the maximum connected domain algorithm.The experimental results on the dHCP dataset show that the method is superior to existing methods,providing potential possibilities for early detection and intervention of neonatal brain injury.

关 键 词：脑部图像区域分割 Swin Transformer编码解码网络 新生儿MRI 自注意力 移位窗口 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]