基于多编码器的脑肿瘤分割算法研究  

Research on brain tumor segmentation algorithm based on multiple encoders

作　　者：萧飞鹏 宋亚男[1] 徐荣华[1] 罗兆林 Xiao Feipeng;Song Yanan;Xu Ronghua;Luo Zhaolin(Guangdong University of Technology,Guangzhou 510006,China)

机构地区：[1]广东工业大学,广东广州510006

出　　处：《无线互联科技》2024年第8期71-76,86,共7页Wireless Internet Technology

基　　金：广东省本科高校在线开放课程指导委员会研究课题,项目名称:多模态数据融合的学习分析与评价创新,项目编号:2022ZXKC143;广东工业大学高水平大学建设研究生教育创新计划项目,项目名称:实践创新能力培养模式研究,项目编号:2018JGMS-09。

摘　　要：为提高脑肿瘤自动分割的准确率,文章提出一种基于多编码器的脑肿瘤分割算法。文章首先采用多编码器的U型网络结构以深度挖掘不同模态的语义信息,然后在下采样的卷积模块使用空洞卷积提取多尺度特征,并且使用基于通道注意力的融合模块融合不同模态的特征,最后在网络瓶颈层使用Transformer模块对全局特征进行充分感知。该算法在Brain Tumor Segmentation Challenge 2018(BraTS 2018)数据集上进行了实验,模型在整体肿瘤、肿瘤核心、增强肿瘤3个区域中的平均Dice系数分别为0.8907、0.7851、0.7487。实验结果表明,该算法可以有效利用模态特征的差异性,提升分割效果。In order to improve the accuracy of automatic brain tumor segmentation,a multi-encoder based brain tumor segmentation algorithm is proposed.In this paper,the multi-encoder U-shaped network structure is firstly adopted to deeply mine the semantic information of different modes,then the convolution module in the down sampling uses cavity convolution to extract multi-scale features,and the fusion module based on channel attention is used to fuse the features of different modes.Finally,the Transformer module is used at the bottleneck layer of the network to fully perceive the global features.The algorithm is tested on the Brain Tumor Segmentation Challenge 2018(BraTS 2018)dataset,and the average Dice coefficients of the whole tumor,the tumor core and the enhanced tumor are 0.8907,0.7851 and 0.7487,respectively.The experimental results show that the algorithm can effectively utilize the difference of modal characteristics and improve the segmentation effect.

关 键 词：多编码器 空洞卷积 注意力机制 TRANSFORMER 脑肿瘤分割 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]