[论文解读] Asymmetric Similarity Loss Function to Balance Precision and Recall in Highly Unbalanced Deep Medical Image Segmentation

Seyed Raein Hashemi, Seyed Sadegh Mohseni Salehi+4 more|arXiv (Cornell University)|Mar 28, 2018

Advanced Neural Network Applications参考文献 39被引用 12

一句话总结

该论文提出了一种基于 $F_\beta$ 评分的非对称相似性损失函数，以在高度不平衡的3D医学图像分割中平衡精确率与召回率。通过结合基于补丁的3D密集连接网络、重叠补丁、B样条加权软投票融合以及非对称损失，该方法在MSSEG 2016上达到69.8%的Dice系数，在ISBI 2015上达到65.74%，显著提升了F1和F2得分以及精确率-召回率曲线下方面积（AUPRC）。

ABSTRACT

Fully convolutional deep neural networks have been asserted to be fast and precise frameworks with great potential in image segmentation. One of the major challenges in utilizing such networks raises when data is unbalanced, which is common in many medical imaging applications such as lesion segmentation where lesion class voxels are often much lower in numbers than non-lesion voxels. A trained network with unbalanced data may make predictions with high precision and low recall, being severely biased towards the non-lesion class which is particularly undesired in medical applications where false negatives are actually more important than false positives. Various methods have been proposed to address this problem including two step training, sample re-weighting, balanced sampling, and similarity loss functions. In this paper we developed a patch-wise 3D densely connected network with an asymmetric loss function, where we used large overlapping image patches for intrinsic and extrinsic data augmentation, a patch selection algorithm, and a patch prediction fusion strategy based on B-spline weighted soft voting to take into account the uncertainty of prediction in patch borders. We applied this method to lesion segmentation based on the MSSEG 2016 and ISBI 2015 challenges, where we achieved average Dice similarity coefficient of 69.8% and 65.74%, respectively, using our proposed patch-wise 3D densely connected network. Significant improvement in $F_1$ and $F_2$ scores and the area under the precision-recall curve was achieved in test using the asymmetric similarity loss layer and our 3D patch prediction fusion method. The asymmetric similarity loss function based on $F_\beta$ scores generalizes the Dice similarity coefficient and can be effectively used with the patch-wise strategy developed here to train fully convolutional deep neural networks for highly unbalanced image segmentation.

研究动机与目标

• 为解决高度不平衡的医学图像分割挑战，其中病灶体素远多于非病灶体素。
• 减少深度学习模型对多数类（非病灶）的偏差，该偏差常导致召回率低和假阴性率高。
• 在病灶分割任务中同时提升精确率与召回率，尤其优先考虑召回率以最小化漏诊。
• 开发一种广义Dice系数的损失函数，并通过 $F_\beta$ 基优化更好地契合临床优先事项。
• 将基于补丁的训练与基于B样条加权的软投票融合策略结合，以增强边界预测的可靠性。

提出的方法

• 在大尺寸重叠图像补丁上训练3D密集连接全卷积网络，以提升特征学习能力和鲁棒性。
• 使用补丁选择算法选取补丁，以确保多样性并覆盖病灶区域。
• 采用B样条加权软投票策略融合来自重叠补丁的预测结果，降低补丁边界处的不确定性。
• 引入一种基于 $F_\beta$ 评分的非对称相似性损失函数，以在类别不平衡设置下优先考虑召回率。
• 该损失函数广义化了Dice系数，并在训练过程中端到端优化，以提升对罕见类别的分割性能。
• 对补丁应用内在和外在的数据增强，以增加训练多样性并提升模型泛化能力。

实验结果

研究问题

• RQ1 基于 $F_\beta$ 评分的非对称损失函数是否能有效平衡高度不平衡3D医学图像分割中的精确率与召回率？
• RQ2 基于补丁的训练结合重叠补丁与软投票融合，如何提升分割精度，尤其是在病灶边界处？
• RQ3 与标准Dice损失相比，所提出的损失函数在F1和F2得分以及精确率-召回率曲线下方面积方面，性能提升程度如何？
• RQ4 将非对称损失与3D密集连接网络结合，是否能在MSSEG 2016和ISBI 2015等基准数据集上带来性能提升？

主要发现

• 所提出的非对称相似性损失函数相比标准Dice训练，显著提升了F1和F2得分。
• 在MSSEG 2016挑战数据集上，该方法实现了69.8%的平均Dice相似系数。
• 在ISBI 2015挑战数据集上，该方法达到了65.74%的Dice相似系数。
• 精确率-召回率曲线下方面积显著提升，表明在少数类（病灶）上的性能更优。
• B样条加权软投票融合策略有效降低了补丁边界处的预测不确定性，提升了边界定位精度。
• 基于补丁的训练、非对称损失与密集连接的结合，在评估基准上实现了最先进性能。