当前位置: X-MOL 学术Pattern Recogn. Lett. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
A hierarchical and multi-view registration of serial histopathological images
Pattern Recognition Letters ( IF 3.9 ) Pub Date : 2021-10-19 , DOI: 10.1016/j.patrec.2021.10.019
Zhineng Chen 1, 2 , Shuai Zhao 3 , Kai Hu 4 , Jing Han 5 , Yuan Ji 5 , Shaoping Ling 6 , Xieping Gao 4
Affiliation  

Serial histopathological images sliced from the same biopsy section exhibit quite similar topological structures. They are usually stained with different dyes and inspected individually for different clinical purposes in the routine workflow of pathology. Recently there is increasing demand for browsing them jointly, e.g., viewing the colocalization of biomarkers. Image registration is recognized as a feasible means to bridge this requirement. However, it is challenging to register the images due to their natural content difference, variation and distortion introduced from slice preparation, the extremely large image size, etc. This letter proposes a novel ierarchical and ulti-iew istration (HMVReg) method to alleviate the difficulties. It is a three-stage coarse-to-fine registration consisting of global rigid registration, multi-view affine transformation and multi-view elastic registration, respectively for eliminating the translational and rotational differences, globally isotropic deformation and locally anisotropic distortion. The stages are complementary and jointly contribute to a stable registration. Experimental results on both private and public data, which cover histopathological images from multiple organ tissues, demonstrate the effectiveness of the proposed HMVReg. It not only well balance the registration accuracy, robustness and speed, where serial but differently stained images are accurately aligned in a reasonable time cost. Moreover, it also shows satisfactory user experience in applying the alignment to biomarker colocalization browsing.

中文翻译:


连续组织病理学图像的分层和多视图配准



从同一活检切片切片的连续组织病理学图像表现出非常相似的拓扑结构。它们通常用不同的染料染色,并在病理学的常规工作流程中针对不同的临床目的单独进行检查。最近,联合浏览它们的需求不断增加,例如查看生物标记的共定位。图像配准被认为是满足这一要求的可行方法。然而,由于切片准备过程中引入的自然内容差异、变化和失真、图像尺寸极大等,配准图像具有挑战性。这封信提出了一种新颖的分层和多视图(HMVReg)方法来缓解困难。它是由全局刚性配准、多视点仿射变换和多视点弹性配准组成的三阶段粗到精配准,分别用于消除平移和旋转差异、全局各向同性变形和局部各向异性畸变。这些阶段是互补的,共同有助于稳定的注册。私人和公共数据的实验结果(涵盖多个器官组织的组织病理学图像)证明了所提出的 HMVReg 的有效性。它不仅很好地平衡了配准精度、鲁棒性和速度,其中连续但不同染色的图像以合理的时间成本准确对齐。此外,在将比对应用于生物标记共定位浏览时,它还表现出了令人满意的用户体验。
更新日期:2021-10-19
down
wechat
bug