基于U-Net的肺肿瘤图像分割算法研究
Research on Segmentation Algorithm of Lung Tumor Image Based on U-Net
刘生智 1雷乔1
作者信息
- 1. 新疆科技学院,新疆库尔勒 841000
- 折叠
摘要
针对肺部CT影像中肿瘤区域复杂多变、边界模糊问题,通过引入注意力机制及优化损失函数,提出一种改进的U-Net网络,力图增强模型对复杂肺部结构中肿瘤区域的识别与分割能力.在临床多模态医学图像数据集上,进行训练,并验证算法的有效性.对比实验结果,本文提出的模型能够实现更高的分割精度(Dice相似系数显著提升)和更低的误分割率,特别是在处理边界模糊、大小不一的肿瘤时表现出色.本模型还表现出良好的泛化能力,在不同患者的CT图像上均能稳定工作.
Abstract
To solve the problem of complex and variable tumor region and blurred boundary in lung CT images,we optimize and design an improved U-Net architecture.The ability of the model to recognize and segment the tumor region in the complex lung structure is enhanced by introducing the attention mechanism and optimizing the loss function.Train on a clinical multimodal medical image dataset and validate the effectiveness of the algorithm.proposes the improved U-Net model by comparing the traditional segmentation method and the standard U-Net model,to achieve higher segmentation accuracy(Dice similarity coefficient is significantly improved)and lower missegmentation rate in lung tumor segmentation technology,especially in the treatment of tumors with blurred boundaries and different sizes.The model also shows good generalization ability and worked stably on CT images of different patients.
关键词
U-Net/肺肿瘤/图像分割算法Key words
U-Net/Lung tumor/Image segmentation algorithm引用本文复制引用
基金项目
新疆科技学院科研基金项目(2023-KYPT23)
出版年
2024
NETL
NSTL
万方数据