论文标题
AGMR-NET：注意力指导的多尺度恢复框架的中风细分框架
AGMR-Net: Attention Guided Multiscale Recovery framework for stroke segmentation
论文作者
论文摘要
自动和准确的病变细分对于临床估计中风疾病的病变状态并开发适当的诊断系统至关重要。尽管现有方法取得了显着的结果，但该模型的进一步采用受到以下方式的阻碍：（1）阶层间的间隙，正常的脑组织类似于外观的病变。 （2）阶层内不一致，病变不同区域之间存在较大的可变性。为了解决中风细分中的这些挑战，我们提出了一种新颖的方法，即在本文中引导的多尺度恢复框架（AGMR-NET）。首先，采用了编码中的粗粒斑块注意模块，以明确监督的多阶段获得基于贴片的粗粒注意力图，从而通过基于贴片的加权技术实现目标空间上下文显着性表示，从而消除了内在不一致的效果。其次，为了获得更详细的边界分配以解决阶层间的挑战，新设计的跨维特征融合模块可用于捕获全局上下文信息，以进一步指导2D和3D特征的选择性聚集，这可以补偿缺乏2D卷积的边界学习能力。最后，在解码阶段，创新设计的多尺度反卷积上采样而不是线性插值可以增强目标空间和边界信息的恢复。 AGMR-NET在病变的开放数据集解剖图上进行评估，并在施加后（ATLAS）（ATLAS）（ATLAS）上获得最高的骰子相似性系数（DSC）分数（DSC）为0.594，Hausdorff距离，Hausdorff距离，27.005 mm的27.005 mm，以及我们的平均表面距离，我们的平均表面距离具有7.137 mm的效果，这表明了7.137 mm的方法，该方法的范围超出了，该方法的范围是，该方法的范围超过了，该方法的范围超过了，该方法的范围是，该方法的范围是，该方法的范围是，该方法的范围是，该方法的范围超过了，该方法的范围是，该方法的范围是，该方法的范围是，该方法效果均为7.137 mm中风。
Automatic and accurate lesion segmentation is critical for clinically estimating the lesion statuses of stroke diseases and developing appropriate diagnostic systems. Although existing methods have achieved remarkable results, further adoption of the models is hindered by: (1) inter-class indistinction, the normal brain tissue resembles the lesion in appearance. (2) intra-class inconsistency, large variability exists between different areas of the lesion. To solve these challenges in stroke segmentation, we propose a novel method, namely Attention Guided Multiscale Recovery framework (AGMR-Net) in this paper. Firstly, a coarse-grained patch attention module in the encoding is adopted to get a patch-based coarse-grained attention map in a multi-stage explicitly supervised way, enabling target spatial context saliency representation with a patch-based weighting technique that eliminates the effect of intra-class inconsistency. Secondly, to obtain a more detailed boundary partitioning to solve the challenge of the inter-class indistinction, a newly designed cross-dimensional feature fusion module is used to capture global contextual information to further guide the selective aggregation of 2D and 3D features, which can compensate for the lack of boundary learning capability of 2D convolution. Lastly, in the decoding stage, an innovative designed multi-scale deconvolution upsampling instead of linear interpolation enhances the recovery of target space and boundary information. The AGMR-Net is evaluated on the open dataset Anatomical Tracings of Lesions-After-Stroke (ATLAS), achieving the highest dice similarity coefficient (DSC) score of 0.594, Hausdorff distance of 27.005 mm, and average symmetry surface distance of 7.137 mm, which demonstrate that our proposed method outperforms other state-of-the-art methods and has great potential in the diagnosis of stroke.
