基于改进V-Net的颅内出血病灶分割算法
Improved V-Net-based lesion segmentation algorithm for intracranial hemorrhage
徐睿 1周长才 2宋宇1
作者信息
- 1. 长春工业大学 计算机科学与工程学院,吉林 长春 130102
- 2. 北京银行股份有限公司 济南分行,山东 济南 250000
- 折叠
摘要
针对颅内出血病灶分割不精确问题提出一种改进V-Net算法.用深度可分离卷积去替换普通卷积,加快模型训练速度.在编码器和解码器中分别加入通道注意力机制和混合注意力机制.通过引入SE模块和CBAM模块,强化原始网络的特征提取能力以及自适应调整特征图中不同通道之间的权重,提高模型的性能表现.对比实验结果表明,改进后的 V-Net分割评价指标DSC达到 0.732,比原始V-Net提升 4.4%.
Abstract
An improved V-Net algorithm is proposed to address the inaccurate segmentation of intracranial hemorrhage lesions.The depth-separable convolution is used to replace the normal convolutionto speed up the model training.A channel attention mechanism and a hybrid attention mechanism are added to the encoder and decoder,respectively.By introducing the SE module and CBAM module,the feature extraction capability of the original network is enhanced as well as the adaptive adjustment of the weights between different channels in the feature map to improve the performance of the model.The comparison experimental results show that the improved V-Net segmentation evaluation index DSC reaches 0.732,which is 4.4%better than the original V-Net.
关键词
深度学习/V-Net模型/深度可分离卷积/颅内出血Key words
deep learning/V-Net model/depth separable convolution/intracranial hemorrhage引用本文复制引用
基金项目
吉林省自然科学基金(20220101128JC)
出版年
2024
NETL
NSTL
万方数据