[论文解读] Exploring Separable Attention for Multi-Contrast MR Image Super-Resolution

Chun-Mei Feng, Yunlu Yan+4 more|arXiv (Cornell University)|Sep 3, 2021

Advanced Image Processing Techniques参考文献 30被引用 14

一句话总结

本文提出SANet，一种用于多对比度磁共振图像超分的新型可分离注意力网络，通过利用辅助对比度提升高、低强度区域的重建质量。通过采用高强度优先和低强度分离注意力机制，SANet改善了解剖细节恢复与边缘保持，在fastMRI和体内数据集上实现了最先进性能，PSNR与SSIM得分均更优。

ABSTRACT

Super-resolving the Magnetic Resonance (MR) image of a target contrast under the guidance of the corresponding auxiliary contrast, which provides additional anatomical information, is a new and effective solution for fast MR imaging. However, current multi-contrast super-resolution (SR) methods tend to concatenate different contrasts directly, ignoring their relationships in different clues, e.g., in the high-intensity and low-intensity regions. In this study, we propose a separable attention network (comprising high-intensity priority attention and low-intensity separation attention), named SANet. Our SANet could explore the areas of high-intensity and low-intensity regions in the "forward" and "reverse" directions with the help of the auxiliary contrast, while learning clearer anatomical structure and edge information for the SR of a target-contrast MR image. SANet provides three appealing benefits: (1) It is the first model to explore a separable attention mechanism that uses the auxiliary contrast to predict the high-intensity and low-intensity regions regions, diverting more attention to refining any uncertain details between these regions and correcting the fine areas in the reconstructed results. (2) A multi-stage integration module is proposed to learn the response of multi-contrast fusion at multiple stages, get the dependency between the fused representations, and boost their representation ability. (3) Extensive experiments with various state-of-the-art multi-contrast SR methods on fastMRI and clinical extit{in vivo} datasets demonstrate the superiority of our model.

研究动机与目标

• 解决现有多对比度SR方法在未建模高、低强度区域间依赖关系的情况下直接拼接对比度的局限性。
• 通过显式建模来自辅助MR对比度的互补解剖信息，提升超分质量。
• 开发一种机制，通过注意力驱动的特征精炼，优化高、低强度区域之间不确定边界的重建。
• 通过多阶段融合模块捕获网络各阶段间的融合依赖关系，增强表征学习能力。
• 实现高性能、参数高效且计算开销极小的超分方法。

提出的方法

• 提出一种可分离注意力机制，将注意力分解为高强度优先与低强度分离路径，以精炼强度区域之间的不确定边界。
• 引入多阶段融合模块，在网络多个深度层次上学习并增强多对比度融合的特征响应。
• 使用可学习的注意力权重比（α）平衡目标对比度与辅助对比度的贡献，最优性能出现在α = 0.7时。
• 采用L1损失函数以实现更快收敛与稳定训练，经消融实验验证其有效性。
• 设计可调节残差组数（2至6组）的残差网络，6组时性能最优，且PSNR与SSIM保持稳定。
• 采用渐进式训练策略，包含共享特征提取分支与对比度特异性精炼分支。

Figure 1: Comparison between single-contrast method SMORE [ 9 ] ), multi-contrast method Lyu et al. [ 10 ] , and our SANet on in vivo datasets. We effectively reconstruct the target contrast image by exploring the complementary information between the clear anatomical structure and boundary details

实验结果

研究问题

• RQ1 是否可通过区分高、低强度区域的可分离注意力机制提升多对比度MR超分性能？
• RQ2 辅助对比度信息的引入如何影响细微解剖结构与边缘的重建质量？
• RQ3 注意力机制中目标与辅助对比度贡献的最优平衡比是多少？
• RQ4 多对比度特征的多阶段融合如何增强表征学习与超分性能？
• RQ5 与现有EDSR基模型相比，所提方法是否能在更少参数下实现更优性能？

主要发现

• 在SMS数据集上，SANet在2倍放大下实现PSNR 42.10 dB与SSIM 0.997，优于基线方法。
• 最优注意力权重比α = 0.7时PSNR与SSIM达到最高，而α < 0.3时性能显著下降。
• 当残差组数达到6组时性能趋于稳定，PSNR与SSIM不再提升，表明模型容量充足且无过拟合。
• SANet仅需11M参数（6组残差），显著少于EDSR（43M）与SMORE（双EDSR分支），同时性能更优。
• 联合对目标与辅助对比度进行超分，目标图像PSNR达31.72 dB、SSIM达0.700，辅助图像PSNR达32.3 dB、SSIM达0.705，证明了多对比度融合的鲁棒性。
• 消融实验表明，L1损失与L2损失性能相当但收敛更快，支持其在最终模型中的使用。

Figure 2: Illustration of our SANet framework . Our model is made up of two branches: a target contrast branch and an auxiliary contrast branch. We input the features $\mathbf{x}_{aux}$ and $\mathbf{x}_{tar}$ of two contrasts to the residual groups $\mathcal{R}_{aux}^{l}$ and $\mathcal{R}_{tar}^{l}$