基于深度学习的弱监督肝脏分割算法  

WEAKLY SUPERVISED LIVER SEGMENTATION ALGORITHM BASED ON DEEP LEARNING

作　　者：王泽辉 张冰[1] 李垣江[1] 黄炜嘉[1] 张正言[1] 杨魏 Wang Zehui;Zhang Bing;Li Yuanjiang;Huang Weijia;Zhang Zhengyan;Yang Wei(School of Electronic Information,Jiangsu University of Science and Technology,Zhenjiang 212003,Jiangsu,China;Department of Intervention,Jiangsu Provincial People s Hospital,Nanjing 210029,Jiangsu,China)

机构地区：[1]江苏科技大学电子信息学院,江苏镇江212003 [2]江苏省人民医院介入科,江苏南京210029

出　　处：《计算机应用与软件》2022年第12期252-259,共8页Computer Applications and Software

摘　　要：目前,基于深度学习的图像分割方法往往需要大量的标注数据集,尤其在需要精确像素级标注的医学图像上,不仅需要高昂的时间成本,还需要大量的专业领域知识。为此,提出一种基于深度学习的弱监督肝脏分割算法。通过卷积神经网络训练具有边框标注信息的肝脏数据;使用Grad-CAM算法获取肝脏粗略位置,通过改进的区域生长算法结合条件随机场完成对目标区域的数据扩张;将图片经过滤波等算法进一步完善分割区域。在3DIRCADb和自建数据集上的实验结果验证了该算法的有效性。At present, image segmentation methods based on deep learning often require a large number of labeled data sets, especially on medical images that require accurate pixel-level labeling, which not only requires a high time cost but also requires a lot of professional domain knowledge. To solve this problem, a weakly supervised liver segmentation algorithm based on deep learning is proposed. The liver data with frame labeling information was trained by the convolutional neural network. The rough position of the liver was obtained using the Grad-CAM algorithm, and the data expansion of the target area was completed by the improved area growth algorithm combined with the conditional random field. Filtering and other algorithms were used on the image to further improve the segmentation area. Experimental results on 3DIRCADb and self-built data sets verify the effectiveness of the algorithm.

关 键 词：弱监督 肝脏分割 Grad-CAM 区域生长 条件随机场 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]