[论文解读] PraNet: Parallel Reverse Attention Network for Polyp Segmentation

Deng-Ping Fan, Ge-Peng Ji+5 more|arXiv (Cornell University)|Jun 13, 2020

Advanced Image and Video Retrieval Techniques参考文献 38被引用 109

一句话总结

PraNet 引入并行部分解码器与反向注意力模块，以在结肠镜图像中实现实时、高精度的息肉分割，并提升跨数据集的泛化能力。

ABSTRACT

Colonoscopy is an effective technique for detecting colorectal polyps, which are highly related to colorectal cancer. In clinical practice, segmenting polyps from colonoscopy images is of great importance since it provides valuable information for diagnosis and surgery. However, accurate polyp segmentation is a challenging task, for two major reasons: (i) the same type of polyps has a diversity of size, color and texture; and (ii) the boundary between a polyp and its surrounding mucosa is not sharp. To address these challenges, we propose a parallel reverse attention network (PraNet) for accurate polyp segmentation in colonoscopy images. Specifically, we first aggregate the features in high-level layers using a parallel partial decoder (PPD). Based on the combined feature, we then generate a global map as the initial guidance area for the following components. In addition, we mine the boundary cues using a reverse attention (RA) module, which is able to establish the relationship between areas and boundary cues. Thanks to the recurrent cooperation mechanism between areas and boundaries, our PraNet is capable of calibrating any misaligned predictions, improving the segmentation accuracy. Quantitative and qualitative evaluations on five challenging datasets across six metrics show that our PraNet improves the segmentation accuracy significantly, and presents a number of advantages in terms of generalizability, and real-time segmentation efficiency.

研究动机与目标

• 推动自动、准确的息肉分割以辅助结直肠癌筛查。
• 解决息肉内部外观变异大与息肉边界模糊的问题。
• 开发适用于实时结肠镜视频的快速、可泛化网络。
• 通过新颖的架构组件利用区域线索与边界线索以提高精度。

提出的方法

• 使用并行部分解码器（PPD）聚合高级特征并生成全局息肉图。
• 使用多个反向注意力（RA）模块，通过抹去预测的息肉区域并细化预测来逐步挖掘边界线索。
• 用包括对全局和侧输出图应用加权 IoU 与加权 BCE 的联合损失进行训练。
• 在多输出上采用深度监督以稳定训练并实现端到端学习。
• 在 352x352 输入下保持实时性能（~50 fps）。

实验结果

研究问题

• RQ1 并行解码策略是否能改善用于息肉分割的高级特征聚合？
• RQ2 反向注意力机制是否通过抹去先前预测有效地细化边界？
• RQ3 PraNet 是否能够在多样化息肉数据集上实现实时推理并获得更高的准确性？
• RQ4 PPD 与 RA 如何协同改善学习速度和分割质量？

主要发现

• PraNet 在五个数据集和多项指标上超越了最新方法。
• 在 Kvasir 上，PraNet 的平均 Dice 为 0.898，平均 IoU 为 0.840。
• 在 CVC-612 上，PraNet 的平均 Dice 为 0.899，平均 IoU 为 0.849。
• PraNet 展现出对未见数据集的良好泛化能力，相对于基线有显著提升。
• 对于 352x352 输入，推理速度达到 ~50 fps 的实时性，训练在 ~20 轮（~0.5 小时）内收敛。
• 消融实验显示 PPD 与 RA 对性能有叠加贡献，完整组合（PPD+RA+Backbone）达到最佳结果。