基于U型网络的肿瘤病灶分割算法  

Tumor Lesion Segmentation Algorithm Based on U-shaped Network

作　　者：王皓 WANG Hao(School of Computer Science,Beijing University of Posts and Telecommunications,Beijing 100876)

机构地区：[1]北京邮电大学计算机学院,北京100876

出　　处：《计算机与数字工程》2023年第12期2954-2958,共5页Computer & Digital Engineering

摘　　要：脑部肿瘤MRI图像作为一种无介入性的诊断工具,包含极其丰富的病理、生理和解剖信息,为临床诊断工作提供了巨大的支持,在医学研究和疾病治疗中发挥重要作用。传统采用人工阅片的方式需要有经验的医生,这就增加了判断的主观性,容易产生误判的可能。为解决这一问题,提出了一种融合注意力机制模块的U-Net改进网络,使模型学习到更多的内容,可以更好地把训练过程集中在对应的病灶上。在BraTS公共数据集划分出的独立测试集上测试该模型,相关指标高于其他对比分割网络。As a non-interventional diagnostic tool,MRI images of brain tumors contain extremely rich pathological,physio-logical and anatomical information,provide tremendous support for clinical diagnosis,and play an important role in medical re-search and disease treatment.The traditional method of manual reading requires experienced doctors,which increases the subjectivi-ty of judgment and is prone to misjudgment.In order to solve this problem,this paper proposes a U-Net improved network that inte-grates attention mechanism modules,so that the model can learn more content and can better focus the training process on the corre-sponding lesions.The model is tested on an independent test set divided by the BraTS public data set,and the relevant indicators are higher than other comparative segmentation networks.

关 键 词：深度学习 注意力机制 图像分割 U-Net 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]