通过U形网络中的双重自蒸馏进行体积医学图像分割
Volumetric medical image segmentation through dual self-distillation in U-shaped networks
AI总结：
本文提出一种用于体积医学图像分割的U形网络双重自蒸馏（DSD）框架，通过将标签知识蒸馏至解码器层及深层特征蒸馏至浅层，显著提升了多种主干网络的分割精度且几乎不增加参数与训练成本。
摘要：
U形网络及其变体在医学图像分割中展现了卓越的结果。在本文中，我们提出了一种用于体积医学图像分割的U形网络中的新型双重自蒸馏（DSD）框架。DSD将知识从真实的分割标签蒸馏到解码器层。此外，DSD还将知识从单个U形网络的最深解码器和编码器层分别蒸馏到较浅的解码器和编码器层。DSD是一种通用的训练策略，可附加到任何U形网络的主干架构上，以进一步提升其分割性能。我们将DSD附加到几种最先进的U形主干网络上，并在各种公开的3D医学图像分割数据集（心脏子结构、脑肿瘤和海马体）上进行了大量实验，结果表明相较于没有DSD的相同主干网络有显著提升。平均而言，在将DSD附加到U形主干网络后，我们观察到心脏子结构、脑肿瘤和海马体分割的Dice相似系数分别增加了2.82%、4.53%和1.3%，Hausdorff距离分别减少了7.15 mm、6.48 mm和0.76 mm。这些改进是在可训练参数数量和训练时间可忽略不计的增加下实现的。我们提出的DSD框架还在心脏子结构、脑肿瘤和海马体分割上带来了显著的定性改进。源代码公开于https://github.com/soumbane/DualSelfDistillation。
英文摘要：
U-shaped networks and its variants have demonstrated exceptional results for medical image segmentation. In this paper, we propose a novel dual self-distillation (DSD) framework in U-shaped networks for volumetric medical image segmentation. DSD distills knowledge from the ground-truth segmentation labels to the decoder layers. Additionally, DSD also distills knowledge from the deepest decoder and encoder layer to the shallower decoder and encoder layers respectively of a single U-shaped network. DSD is a general training strategy that could be attached to the backbone architecture of any U-shaped network to further improve its segmentation performance. We attached DSD on several state-of-the-art U-shaped backbones, and extensive experiments on various public 3D medical image segmentation datasets (cardiac substructure, brain tumor and Hippocampus) demonstrated significant improvement over the same backbones without DSD. On average, after attaching DSD to the U-shaped backbones, we observed an increase of 2.82\%, 4.53\% and 1.3\% in Dice similarity score, a decrease of 7.15 mm, 6.48 mm and 0.76 mm in the Hausdorff distance, for cardiac substructure, brain tumor and Hippocampus segmentation, respectively. These improvements were achieved with negligible increase in the number of trainable parameters and training time. Our proposed DSD framework also led to significant qualitative improvements for cardiac substructure, brain tumor and Hippocampus segmentation over the U-shaped backbones. The source code is publicly available at https://github.com/soumbane/DualSelfDistillation.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
