摘要
伴随着图像处理技术和人工智能技术的飞速发展,医学影像图的自动分割成为了医学家热切关注的话题。研究通过密集跳跃连接和改进残差网络优化的Unet网络,再将其应用于脑部肿瘤影像图像的分割,并通过标准化层FRN进行深度学习模型训练和利用混合损失函数计算模型损失。研究结果表明,相较于其他影像图像分割算法,残差-密集跳跃-Unet网络的Hausdorff距离、敏感度、特异性、Dice四个指标均更佳。Hausdorff距离的最大降低值为1.061,灵敏度、特异性、Dice的提升值0.999,0.092,0.110。残差-密集跳跃连接-Unet网络能有效分割出边界区域,其具有极高的分割效率。
With the rapid development of image processing technology and artificial intelligence technology,automatic segmentation of medical image has become a hot topic of medical experts.The Unet network optimized by dense jump connection and improved residual network is studied,and then applied to the segmentation of brain tumor image.The in-depth learning model training is carried out through the standardized layer FRN and the model loss is calculated by using the mixed loss function.The results show that compared with other image segmentation algorithms,the Hausdorff distance,sensitivity,specificity and dice of residual dense jump Unet network are better.The maximum decrease value of Hausdorff distance is 1.061,and the increase values of sensitivity,specificity and Dice are 0.999,0.092 and 0.110.Residual dense jump connection Unet network can effectively segment the boundary region,and it has high segmentation efficiency.
作者
吴俊
WU Jun(Department of Nursing,Xuancheng Vocational and Technical College,Xuancheng Anhui 242000,China)
出处
《佳木斯大学学报（自然科学版）》
CAS
2023年第2期144-148,共5页
Journal of Jiamusi University：Natural Science Edition
基金
2020年安徽省高等学校质量工程项目(2020zyq95)。
作者简介
吴俊(1984-),男,安徽宣城人,讲师,双学士,研究方向:医学教育教学、肿瘤学。
