基于改进3D UNet的前列腺MR图像分割  

Prostate magnetic resonance image segmentation based on improved 3D UNet

作　　者：桑子江 邵叶秦 许昌炎 SANG Zijiang;SHAO Yeqin;XU Changyan(School of Information Science and Technology,Nantong University,Nantong,Jiangsu 226000,China;School of Transportation and Civil Engineering,Nantong University,Nantong,Jiangsu 226000,China)

机构地区：[1]南通大学信息科学技术学院,江苏南通226000 [2]南通大学交通与土木工程学院,江苏南通226000

出　　处：《光电子．激光》2023年第12期1337-1344,共8页Journal of Optoelectronics·Laser

基　　金：国家自然科学基金(61671255)资助项目。

摘　　要：针对前列腺磁共振(magnetic resonance, MR)图像边缘模糊、对比度较低,灰度值分布不均衡而导致分割精度较差的问题,提出了一种结合双路径注意力(dual path attention, DPA)和多尺度特征聚合(multi-scale feature aggregation, MFA)模块的改进3D UNet网络模型。首先,对数据集进行重采样和裁剪处理以适应模型输入。然后,在3D UNet网络的编码器各层引入DPA并添加残差连接,加强特征的编码能力。同时,在网络解码器中加入MFA模块,以充分利用空间上下文信息,增强语义信息。最后,在公开数据集PROMISE12上进行验证,所提出的模型的Dice系数为89.90%,Hausdorff距离为9.37 mm。相比较于其他模型,所提出模型的分割结果更优,且参数量和运算量更少。An improved 3D UNet network model that combines a dual path attention(DPA)module and a multi-scale feature aggregation(MFA)module is proposed to address the problems such as blurred edges,low contrast,and uneven gray value distribution in magnetic resonance(MR)images of the prostate bringing about the poor segmentation accuracy.Firstly,the dataset is resampled and cropped to fit the model input.Then,DPA and residual connection are added to each layer of the 3D UNet network encoder to enhance the feature coding capability.At the same time,an MFA module is added to the network decoder to make full use of spatial context information and enhance semantic information.Finally,the proposed model is validated on the public dataset PROMISE12,with the Dice coefficient of 89.90%and the Hausdorff distance of 9.37 mm.Compared with other models,the proposed model has better segmentation results,and the number of parameters and the amount of computation are less.

关 键 词：前列腺分割 双路径注意力(DPA) 多尺度特征聚合(MFA) 医学图像分割 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]