[论文解读] MISSFormer: An Effective Medical Image Segmentation Transformer

Xiaohong Huang, Zhifang Deng+2 more|arXiv (Cornell University)|Sep 15, 2021

Advanced Neural Network Applications参考文献 40被引用 170

一句话总结

MISSFormer 引入一个分层 U 形变换器，配备增强变换器块和增强变换器上下文桥，从零训练，在 Synapse 和 ACDC 数据集上实现了最先进的医学图像分割。

ABSTRACT

The CNN-based methods have achieved impressive results in medical image segmentation, but they failed to capture the long-range dependencies due to the inherent locality of the convolution operation. Transformer-based methods are recently popular in vision tasks because of their capacity for long-range dependencies and promising performance. However, it lacks in modeling local context. In this paper, taking medical image segmentation as an example, we present MISSFormer, an effective and powerful Medical Image Segmentation tranSFormer. MISSFormer is a hierarchical encoder-decoder network with two appealing designs: 1) A feed-forward network is redesigned with the proposed Enhanced Transformer Block, which enhances the long-range dependencies and supplements the local context, making the feature more discriminative. 2) We proposed Enhanced Transformer Context Bridge, different from previous methods of modeling only global information, the proposed context bridge with the enhanced transformer block extracts the long-range dependencies and local context of multi-scale features generated by our hierarchical transformer encoder. Driven by these two designs, the MISSFormer shows a solid capacity to capture more discriminative dependencies and context in medical image segmentation. The experiments on multi-organ and cardiac segmentation tasks demonstrate the superiority, effectiveness and robustness of our MISSFormer, the experimental results of MISSFormer trained from scratch even outperform state-of-the-art methods pre-trained on ImageNet. The core designs can be generalized to other visual segmentation tasks. The code has been released on Github: https://github.com/ZhifangDeng/MISSFormer

研究动机与目标

• 解决 CNN 在医学图像分割中建模长距离依赖的局限性
• 提出一个无位置依赖的分层 U 形变换器以实现更准确的分割
• 设计增强变换器块和增强变换器上下文桥以捕获局部与全局上下文
• 在多器官 CT 与心脏 MRI 数据集上评估 MISSFormer 以证明其鲁棒性及从零训练的能力

提出的方法

• 重新设计的前馈网络，命名为 Enhanced Mix-FFN，以提升特征判别力并整合局部/全局上下文
• 增强变换器块，将 LayerNorm、Efficient Self-Attention 与 Enhanced Mix-FFN 相结合，以在降低复杂度的同时建模长距离与局部信息
• 增强变换器上下文桥，通过连接展平后的多层令牌并通过增强变换器块处理来融合多尺度特征
• 分层编码器-解码器，使用重叠的 4×4 块、补丁合并/扩张，以及跳跃连接，形成 U 形结构
• 从零开始在医学数据集上训练，配合标准数据增强，使用 SGD 优化器和多项式学习率策略

实验结果

研究问题

• RQ1 MISSFormer 是否能够在 Synapse 和 ACDC 数据集上从零训练超越现有的医学图像分割方法？
• RQ2 增强变换器块和上下文桥是否在判别力与上下文建模方面优于先前的 Transformer/MLP 基方法？
• RQ3 通过增强变换器上下文桥进行多尺度特征融合如何影响分割精度和边界描绘？
• RQ4 在增强 Mix-FFN 中不同的跳跃连接与递归步骤对收敛与性能有何影响？

主要发现

• MISSFormer 在 Synapse 和 ACDC 数据集上实现了最先进的性能，且常常优于在 ImageNet 上预训练的方法
• 简单的增强 Mix-FFN 与递归跳跃连接相较于基线的 SegFormer 变体提高了训练稳定性与分割精度
• 消融实验显示，加入增强变换器上下文桥与多尺度融合的 MISSFormer 在 Dice 指标（DSC）上获得显著提升且边界描绘改善
• MISSFormer_S（无多尺度桥）相较于 MISSFormer 表现更差，凸显多尺度信息整合的好处
• MISSFormer 在处理困难样本时表现出强边缘预测能力和稳健表现，在 Synapse 上对每个器官的结果具有竞争力或更优，在 ACDC 上的整体鲁棒性也更强