一种用于医学图像分割的混合卷积网络
A hybrid convolutional network for medical image segmentation
张眉芳 1李其铿 1谢隆腾1
作者信息
- 1. 福建卫生职业技术学院公共卫生与健康管理学院,福州 350101
- 折叠
摘要
从医学图像中鲁棒性地分割器官是医学图像分析用于疾病诊断的关键技术之一.U-Net是一种用于医学图像分割的鲁棒结构.然而,U-Net采用连续的下采样编码器来捕捉多尺度特征,高层语义特征恢复不足,从而导致上下文信息丢失,无法充分恢复待分割器官特征.对编码器难以捕捉多尺度特征、高层语义特征恢复不足导致上下文信息丢失进行研究,提出了一种新的混合卷积网络来捕捉更多的上下文信息和高级语义特征.混合卷积网络的主要思想是利用提出的混合卷积连接模块从特征编码器提取更多的上下文信息和高级语义特征.多尺度特征提取模块用于连接编码器和解码器子网络,以获得更丰富的多尺度特征图.将提出的方法与最先进的方法在CHASEDB数据集和FRSA数据集上进行了比较.实验结果表明,提出方法的分割效果优于其他分割方法.
Abstract
Robust segmentation of organs from medical images isone of the key techniques in medical image analysis for disease diagnosis.U-Net is a robust structure for medical image segmentation.However,U-Net uses a continuous downsampling encoder to capture multi-scale features,and the high-level semantic features are not sufficiently recovered,which leads to the loss of contextual information and fails to adequately recover the organ features to be segmented.In this paper,a new hybrid convolu-tional network is proposed to capture more contextual information and high-level semantic features.The main idea of the hybrid convolutional network is to extract more contextual information and high-level semantic features from the feature encoder using the proposed hybrid convolutional connectivity module.The multi-scale feature extraction module is used to connect the encoder and decoder sub-networks to obtain richer multi-scale feature maps.The proposed method is compared with the state-of-the-art methods on CHASEDB dataset and FRSA dataset.The experimental results show that the proposed method outperforms other segmentation methods.
关键词
医学图像分割/混合卷积/多尺度特征/上下文信息Key words
medical image segmentation/hybrid convolutional/multi-scale feature/contextual information引用本文复制引用
基金项目
福建省科技计划引导项目(2021H01010037)
福建卫生职业技术学院卫生信息管理专业教学创新团队项目(2022)(JG2022105)
出版年
2024
NETL
NSTL
万方数据