Restaining-based annotation for cancer histology segmentation to overcome annotation-related limitations among pathologists.
Restaining-based annotation for cancer histology segmentation to overcome annotation-related limitations among pathologists.
复制标题
DOI：
10.1016/j.patter.2023.100688
复制
发表时间：
2023-02-10
期刊：
影响因子：
6.5
通讯作者：
Ishikawa, Shumpei
中科院分区：
文献类型：
--
作者：
Komura, Daisuke;Onoyama, Takumi;Shinbo, Koki;Odaka, Hiroto;Hayakawa, Minako;Ochi, Mieko;Herdiantoputri, Ranny Rahaningrum;Endo, Haruya;Katoh, Hiroto;Ikeda, Tohru;Ushiku, Tetsuo;Ishikawa, Shumpei
在过去的几十年里，大量的癌症组织病理学标本被收集和数字化。对肿瘤组织切片中各种细胞的分布进行综合评价可以为了解癌症提供有价值的信息。深度学习适合实现这些目标；然而，广泛、无偏的训练数据的收集受到阻碍，从而限制了准确分割模型的产生。这项研究提出了segpath——最大的注释数据集（>比公开可用的注释大10倍）——用于分割苏木精和伊红（H&E）染色的癌症组织中8种主要细胞类型的切片。SegPath生成管道使用h&e染色切片，然后用精心选择的抗体进行免疫荧光染色。我们发现SegPath与病理学家注释相当或优于病理学家注释。此外，病理学家的注释偏向于典型的形态。然而，在SegPath上训练的模型可以克服这一限制。我们的研究结果为组织病理学中的机器学习研究提供了基础数据集。SegPath是最大的癌症组织学分割注释数据集免疫荧光抑制实现高通量和准确的注释SegPath在形态学上比病理学家的注释偏差更小肿瘤组织由各种细胞类型组成。关于肿瘤组织中各种细胞位置的信息对于识别肿瘤特征至关重要；然而，准确和快速地估计这些信息是具有挑战性的。基于深度学习的分割可以克服这一挑战，但受到训练数据量不足的阻碍。因此，我们创建了训练数据集，通过抗体免疫染色，以前所未有的规模对各种组织或细胞进行分割，识别各种组织/细胞类型。SegPath标注在准确率和形态偏差方面优于手动标注，从而实现更优化的分割模型开发。将SegPath训练的分割模型应用于医院积累的大量癌症组织病理学标本，可以显著影响癌症诊断和获得对癌症研究的额外见解。我们创建了最大规模的癌症组织学图像分割数据集。用抗体免疫染色识别八种组织/细胞类型产生的数据集比传统的人类注释更准确。这些数据集可以为癌症组织学图像开发准确的深度学习模型，这可以帮助计算机辅助诊断、诊断解释和癌症的基础科学。
Numerous cancer histopathology specimens have been collected and digitized over the past few decades. A comprehensive evaluation of the distribution of various cells in tumor tissue sections can provide valuable information for understanding cancer. Deep learning is suitable for achieving these goals; however, the collection of extensive, unbiased training data is hindered, thus limiting the production of accurate segmentation models. This study presents SegPath—the largest annotation dataset (>10 times larger than publicly available annotations)—for the segmentation of hematoxylin and eosin (H&E)-stained sections for eight major cell types in cancer tissue. The SegPath generating pipeline used H&E-stained sections that were destained and subsequently immunofluorescence-stained with carefully selected antibodies. We found that SegPath is comparable with, or outperforms, pathologist annotations. Moreover, annotations by pathologists are biased toward typical morphologies. However, the model trained on SegPath can overcome this limitation. Our results provide foundational datasets for machine-learning research in histopathology. SegPath is the largest annotation dataset for cancer histology segmentation Immunofluorescence restaining enables high-throughput and accurate annotation SegPath is morphologically less biased than pathologists’ annotation Tumor tissue is composed of various cell types. Information on the location of various cells in tumor tissue is essential to identifying tumor features; however, the accurate and quick estimation of this information is challenging. Deep-learning-based segmentation can overcome this challenge but is hindered by the insufficient amount of training data. We therefore created training datasets for the segmentation of various tissues or cells at an unprecedented scale through immunostaining with antibodies that identify various tissue/cell types. SegPath annotation outperforms manual annotation in terms of accuracy and morphological bias, leading to more optimized segmentation model development. Application of the segmentation model trained on SegPath to a large number of cancer histopathology specimens that have been accumulated in hospitals could significantly impact cancer diagnosis and acquisition of additional insight into cancer research. We created the largest-scale datasets for the segmentation of cancer histology images. Immunostaining with antibodies that recognize eight tissue/cell types yields datasets that are more accurate than those of conventional human annotations. These datasets enable the development of accurate deep-learning models for cancer histological images, which could assist in computer-aided diagnosis, interpretation of the diagnosis, and basic science of cancer.
登录
查看更多内容
影响因子：
8.8
作者：
German, Yolla;Vulliard, Loan;Dupre, Loic
通讯作者：
Dupre, Loic
影响因子：
4.6
作者：
Ing N;Huang F;Conley A;You S;Ma Z;Klimov S;Ohe C;Yuan X;Amin MB;Figlin R;Gertych A;Knudsen BS
通讯作者：
Knudsen BS
影响因子：
7.7
作者：
Lal, Shyam;Das, Devikalyan;Kini, Jyoti
通讯作者：
Kini, Jyoti
DOI：
10.1002/path.5028
发表时间：
2018-04
期刊：
The Journal of pathology
影响因子：
--
作者：
Cooper LA;Demicco EG;Saltz JH;Powell RT;Rao A;Lazar AJ
通讯作者：
Lazar AJ
影响因子：
6.5
作者：
Kiuru, Maija;Kriner, Michelle A.;Wong, Samantha;Zhu, Guannan;Terrell, Jessica R.;Li, Qian;Hoang, Margaret;Beechem, Joseph;McPherson, John D.
通讯作者：
McPherson, John D.