Boundary-Aware Fully Convolutional Network for Brain Tumor Segmentation
作者 Haocheng Shen, Ruixuan Wang, Jianguo Zhang, and Stephen J. McKenna
本文来源于MICCAI2017.
目的:Brain Tumor 分割
方法:多任务网络(区域检测+边缘检测)。方法与DCAN类似。
对比方法:FCN & FCN+CRF
数据集:BRATS13 & BRATS15
对比的变形的FCN网络:
此网络来源于Shen, H., Zhang, J., Zheng, W.: Efficient symmetry-driven fully convolutional network for multimodal brain tumor segmentation. In: ICIP (2017, to appear) 是作者的另一篇文章。会议还没开所以还没看到文章。
此网络结构类似VGG-16。是作者所提Boundary-Aware FCN的一个分支,相当于新网络的基础。
Boundary-Aware FCN
在FCN基础上,上采样分为两个分支,一个用于边缘检测,一个用于区域检测,这一点于DCAN类似。不同的是,此网络再次进行了联合训练。
边缘检测相当于二分类,区域检测相当于多分类。
输入八张图片。symmetric intensity difference maps are combined with original slices as input, resulting in 8 input channels
Loss相当于三部分的loss之和:
评测
BRATS13:20 HG patients for training and 10 HGs for testing. (The 10 low-grade
patients were not used.)
BRATS15: 220 annotated HG patients’ images in the training set.
For each patient there were 4 modalities (T1, T1-contrast (T1c), T2 and Flair)
按照比例6:2:2将训练集随机分为132 training, 44 validation,44 test images.
参数设置:Adam optimizer. Learning rate= 0.001. 下采样过程使用 VGG-16 权值初始化。
在三个任务上进行评估:
(1) the complete tumor (including all four tumor structures);
(2) the tumor core (including all tumor structures except edema);
(3) the enhancing tumor region (including only the enhancing tumor structure)
在complete tumor任务上,本文网络有明显效果提升,其他任务并没有明显提升。只采用DCAN(去掉后面的联合训练)Dice会降低15%。将原FCN加深也没有效果提升,说明此网络并不是因为更深才导致效果提升。
评价指标有Dice,Positive predictive value,Sensitivity。