[论文阅读] Efficient semi-supervised gross target volume of nasopharyngeal carcinoma segmentation via

[论文地址] [代码] [MICCAI 21]

Abstract

总目标体积(Gross Target Volume, GTV)分割在鼻咽癌(NasoPharyngeal Carcinoma, NPC)的放疗计划中起着不可替代的作用。尽管卷积神经网络(CNN)在这一任务中取得了良好的性能,但它们依赖于大量的标记图像进行训练,而这些图像的获取是昂贵和耗时的。在本文中,我们提出了一个新颖的框架,采用不确定修正金字塔一致性(URPC)正则化,用于半监督的NPC GTV分割。具体来说,我们扩展了一个骨干分割网络来产生不同尺度的金字塔预测。金字塔预测网络(PPNet)由已标记图像的基础真值和未标记图像的多尺度一致性损失监督,其动机是对同一输入的不同尺度的预测应该是相似和一致的。然而,由于这些预测的分辨率不同,鼓励它们直接在每个像素上保持一致,其鲁棒性较低,可能会失去一些细微的细节。为了解决这个问题,我们进一步设计了一个新颖的不确定性矫正模块,使框架能够逐渐从不同尺度上有意义的、可靠的一致区域中学习。在一个有258张NPC MR图像的数据集上的实验结果表明,在只有10%或20%的图像被标记的情况下,我们的方法通过利用未标记的图像在很大程度上提高了分割性能,而且它也优于五个最先进的半监督分割方法。此外,当只有50%的标记图像时,URPC取得了82.74%的平均Dice分数,接近于完全监督学习。

Method

本文提出了一个使用半监督进行鼻咽癌分割的框架,如下所示:
在这里插入图片描述
主要亮点在于不确定性(uncertainty)的计算方式。传统方法诸如MC-Dropout依赖于对同一张图像进行多次推理,从而较为耗时。本文受分割任务中常用的deep supervision的启发,提出了一种"Multi-Scale Consistency",即各级decoder输出的内容应该是一致的: D s ≈ ∑ j = 0 C p s j ⋅ log ⁡ p s j p c j \mathcal{D}_{s} \approx \sum_{j=0}^{C} p_{s}^{j} \cdot \log \frac{p_{s}^{j}}{p_{c}^{j}} Dsj=0Cpsjlogpcjpsj 使用的是KL散度。 p s p_s ps为各级decoder的预测结果, p c p_c pc为各级decoder的预测结果的平均。这一不确定性有两种作用,直接看最终的无监督损失: L unsup  = 1 S ∑ s = 0 S − 1 ∑ v ( p s v − p c v ) 2 ⋅ w s v ∑ s = 0 S − 1 ∑ v w s v + 1 S ∑ s = 0 S − 1 ∥ D s ∥ 2 \mathcal{L}_{\text {unsup }}=\frac{1}{S} \frac{\sum_{s=0}^{S-1} \sum_{v}\left(p_{s}^{v}-p_{c}^{v}\right)^{2} \cdot w_{s}^{v}}{\sum_{s=0}^{S-1} \sum_{v} w_{s}^{v}}+\frac{1}{S} \sum_{s=0}^{S-1}\left\|\mathcal{D}_{s}\right\|_{2} Lunsup =S1s=0S1vwsvs=0S1v(psvpcv)2wsv+S1s=0S1Ds2 包含两部分。右边这个 1 S ∑ s = 0 S − 1 ∥ D s ∥ 2 \frac{1}{S} \sum_{s=0}^{S-1}\left\|\mathcal{D}_{s}\right\|_{2} S1s=0S1Ds2指的就是直接约束各级输出结果应尽可能一致,而左边这个相当于额外利用这个uncertainty做了一个attention,即选择"各级输出结果一致"的像素来参与监督过程。如果有的位置预测结果是不一致的,意味着该区域不确定性较高,不将其纳入监督。

Semi-supervised classification with graph convolutional networks (GCNs) is a method for predicting labels for nodes in a graph. GCNs are a type of neural network that operates on graph-structured data, where each node in the graph represents an entity (such as a person, a product, or a webpage) and edges represent relationships between entities. The semi-supervised classification problem arises when we have a graph where only a small subset of nodes have labels, and we want to predict the labels of the remaining nodes. GCNs can be used to solve this problem by learning to propagate information through the graph, using the labeled nodes as anchors. The key idea behind GCNs is to use a graph convolution operation to aggregate information from a node's neighbors, and then use this aggregated information to update the node's representation. This operation is then repeated over multiple layers, allowing the network to capture increasingly complex relationships between nodes. To train a GCN for semi-supervised classification, we use a combination of labeled and unlabeled nodes as input, and optimize a loss function that encourages the network to correctly predict the labels of the labeled nodes while also encouraging the network to produce smooth predictions across the graph. Overall, semi-supervised classification with GCNs is a powerful and flexible method for predicting labels on graph-structured data, and has been successfully applied to a wide range of applications including social network analysis, drug discovery, and recommendation systems.
评论 1
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值