AbstractTumor-Infiltrating Lymphocytes (TILs) have strong prognostic and predictive value in breast cancer, but their visual assessment is subjective. To improve reproducibility, the International Immuno-oncology Working Group recently released recommendations for the computational assessment of TILs that build on visual scoring guidelines.

摘要肿瘤浸润淋巴细胞（TIL）在乳腺癌中具有很强的预后和预测价值，但其视觉评估是主观的。为了提高可重复性，国际免疫肿瘤学工作组最近发布了基于视觉评分指南的TIL计算评估建议。

However, existing resources do not adequately address these recommendations due to the lack of annotation datasets that enable joint, panoptic segmentation of tissue regions and cells. Moreover, existing deep-learning methods focus entirely on either tissue segmentation or cell nuclei detection, which complicates the process of TILs assessment by necessitating the use of multiple models and reconciling inconsistent predictions.

然而，由于缺乏能够对组织区域和细胞进行关节，全景分割的注释数据集，现有资源不能充分解决这些建议。。

We introduce PanopTILs, a region and cell-level annotation dataset containing 814,886 nuclei from 151 patients, openly accessible at: sites.google.com/view/panoptils. Using PanopTILs we developed MuTILs, a neural network optimized for assessing TILs in accordance with clinical recommendations. MuTILs is a concept bottleneck model designed to be interpretable and to encourage sensible predictions at multiple resolutions.

我们介绍了PanopTILs，这是一个区域和细胞水平的注释数据集，包含来自151名患者的814886个细胞核，可在以下网址公开访问：sites.google.com/view/PanopTILs。使用PanopTILs，我们开发了MuTILs，这是一种根据临床建议优化用于评估TIL的神经网络。MuTILs是一种概念瓶颈模型，旨在解释并鼓励以多种分辨率进行明智的预测。

Using a rigorous internal-external cross-validation procedure, MuTILs achieves an AUROC of 0.93 for lymphocyte detection and a DICE coefficient of 0.81 for tumor-associated stroma segmentation. Our computational score closely matched visual scores from 2 pathologists (Spearman R = 0.58–0.61, p < 0.001).

使用严格的内部-外部交叉验证程序，MuTILs实现了淋巴细胞检测的AUROC为0.93，肿瘤相关基质分割的骰子系数为0.81。我们的计算得分与两位病理学家的视觉得分非常匹配（Spearman R=0.58-0.61，p<0.001）。

Moreover, computational TILs scores had a higher prognostic value than visual scores, independent of TNM stage and patient age. In conclusion, we introduce a comprehensive open data resource and a modeling approach for detailed mapping of the breast tumor microenvironment..

此外，计算TILs评分比视觉评分具有更高的预后价值，与TNM分期和患者年龄无关。总之，我们介绍了一个全面的开放数据资源和一种建模方法，用于详细绘制乳腺肿瘤微环境。。

IntroductionAdvances in digital imaging of glass slides and machine learning have increased interest in histology as a source of data in cancer studies1,2. Tissue morphology contains important prognostic and diagnostic information and reflects underlying molecular and biological processes. This work presents approaches for the computational discovery of interpretable predictive histologic biomarkers, focusing on invasive breast carcinomas and immune response.

引言载玻片数字成像和机器学习的进展增加了人们对组织学作为癌症研究数据来源的兴趣1,2。组织形态学包含重要的预后和诊断信息，并反映了潜在的分子和生物学过程。这项工作提出了计算发现可解释的预测性组织学生物标志物的方法，重点是浸润性乳腺癌和免疫反应。

Histopathology is a medical field where medical experts (i.e., pathologists) examine stained microscopic tissue sections to make diagnostic decisions, most often from tumor biopsies. While much of medicine relies on the clinical examination of patients, histopathology is a visual-focused field, like radiology, where much of the focus is on visual pattern recognition.The term biomarker refers to a biological feature that we can use to indicate a clinical outcome.

组织病理学是一个医学领域，医学专家（即病理学家）检查染色的显微组织切片以做出诊断决定，最常见的是从肿瘤活检中做出诊断决定。虽然许多医学依赖于患者的临床检查，但组织病理学是一个视觉聚焦领域，就像放射学一样，其中大部分重点是视觉模式识别。术语生物标志物是指我们可以用来指示临床结果的生物学特征。

For example, prognostic biomarkers are biological features associated with good (or bad) prognosis, while predictive biomarkers predict response to therapy in randomized controlled trials3. Typically, when a histologic trait is related to outcomes in cancer, it is incorporated into the grading criteria, though this is not always the case.

例如，预后生物标志物是与良好（或不良）预后相关的生物学特征，而预测性生物标志物预测随机对照试验中对治疗的反应3。通常，当组织学特征与癌症的结果相关时，它被纳入分级标准，尽管情况并非总是如此。

For example, there has been a strong focus on tumor-infiltrating lymphocytes (TILs) as a prognostic and predictive biomarker in breast cancer and other solid tumors in recent years4. This is because TILs infiltration can be a somewhat direct visualization of how well the host (patient) body can respond to the growing tumor by immune cells.The majority of breast cancers are carcinomas.

例如，近年来，人们强烈关注肿瘤浸润淋巴细胞（TILs）作为乳腺癌和其他实体瘤的预后和预测生物标志物4。这是因为TIL浸润可以在某种程度上直接显示宿主（患者）身体对免疫细胞生长的肿瘤的反应程度。大多数乳腺癌是癌症。

Based on morphology, breast carcinomas include many variants; the most common are infiltrating ductal carcinoma (which originates from breas.

根据形态学，乳腺癌包括许多变体；最常见的是浸润性导管癌（起源于乳腺癌）。

1.

1.

Misclassifications of some benign or low-grade tumor nuclei as TILs.

将一些良性或低度恶性肿瘤细胞核错误分类为TIL。

2.

2.

Variations in TILs density in different areas within the slide, which cause inconsistencies in visual scoring. This phenomenon is also a well-known contributor to inter-observer variability in visual TILs scoring10.

幻灯片内不同区域的TILs密度变化，导致视觉评分不一致。这种现象也是视觉TILs评分中观察者间差异的一个众所周知的原因10。

3.

3.

Variable influence of tertiary lymphoid structures on the WSI-level score.

三级淋巴结构对WSI水平评分的可变影响。

Our results show that the most prognostic TILs score variant (nTnS) is derived from dividing the number of TILs cells by the total number of cells within the stromal region. The visual scoring guidelines rely on the nTSa, which is reflected in the slightly higher correlation of the nTSa variant with the visual scores compared to nTnS9.

我们的结果表明，最具预后的TILs评分变异（nTnS）来自将TILs细胞数量除以基质区域内细胞总数。视觉评分指南依赖于nTSa，这反映在nTSa变体与视觉评分的相关性略高于nTnS9。

So why is nTnS more prognostic than nTSa? There are two potential explanations. First, it may be that nTnS is better controlled for stromal cellularity since it would be the same in low- vs. high-cellularity stromal regions if the proportion of stromal cells that are TILs is the same. Second, nTnS may be less noisy since it relies entirely on nuclear assessment at 20x objective, while stromal regions are segmented at half that resolution.Finally, we note that this validation was done only using the TCGA cohort, and future work will include validation on more breast cancer cohorts.

那么为什么NTN比nTSa更具预后？有两种可能的解释。首先，可能是nTnS对基质细胞的控制更好，因为如果TIL的基质细胞比例相同，那么在低细胞与高细胞基质区域中nTnS是相同的。其次，NTN可能噪音较小，因为它完全依赖于20倍目标的核评估，而基质区域以该分辨率的一半分割。最后，我们注意到该验证仅使用TCGA队列进行，未来的工作将包括对更多乳腺癌队列的验证。

In addition, we note that MuTILs cannot distinguish cancer from normal breast tissue at low resolution, which may necessitate manual curation of the analysis region, especially for low-grade cases.MethodsMuTILs model designMuTILs jointly classifies tissue regions and cell nuclei and extends our earlier work on this topic (Fig.

此外，我们注意到MuTILs无法以低分辨率区分癌症和正常乳腺组织，这可能需要手动管理分析区域，尤其是对于低度病例。方法MTILS模型设计Mutils联合对组织区域和细胞核进行分类，并扩展了我们在这个主题上的早期工作（图）。

2)18. It acts as a panoptic segmentation algorithm19; that is, it uses semantic segmentation to delineate tissue regions and instance segmentation to segment and classify individual cell nuclei to enable a holistic, context-aware assessment of TILs. MuTILs comprises two parallel U-Net models20 (each with a depth of 5) for segmenting tissue regions and nuclei at 10X objective and 20X objective magnifications, respectively.

2） 18岁。它充当全景分割算法19；也就是说，它使用语义分割来描绘组织区域，并使用实例分割来分割和分类单个细胞核，以实现对TIL的整体，上下文感知评估。MuTILs包括两个平行的U-Net模型20（每个深度为5），分别以10倍物镜和20倍物镜放大率分割组织区域和细胞核。

Inspired by the HookNet method, information is shared from the tissue region segmentation to inform nucleus segmentatio.

受HookNet方法的启发，从组织区域分割中共享信息以通知细胞核分割。

1.

1.

Number of TILs/Stromal area (nTSa)

TIL/基质面积（nTSa）的数量

2.

2.

Number of TILs/Number of cells in stroma (nTnS)

TILs数量/基质细胞数量（nTnS）

3.

3.

Number of TILs/Number of cells anywhere (nTnA)

TIL数量/任何位置的单元格数量（nTnA）

We obtained these score variants using two aggregation strategies: 1. Globally (aggregating region and nuclear counts from informative tiles) and 2. By saliency-weighted averaging of informative tiles. The saliency score for each tile was obtained using a Euclidean distance transform to identify stroma within 32 microns from the tumor boundary.

我们使用两种聚合策略获得了这些分数变体：1。全球（从信息瓷砖中汇总区域和核计数）和2。。使用欧几里得距离变换获得每个瓷砖的显着性得分，以识别距离肿瘤边界32微米内的基质。

The fraction of image pixels occupied by this peritumoral stroma was used as a saliency score for each tile. The 32 micron distance was determined by visual comparison of 8, 16, 32, and 64 microns and finding 32 to most closely represent the commonly accepted definition of peritumoral stroma.Computational TIL score calibrationA simple linear calibration was used to scale computational scores to a similar range of magnitudes as the visual scores.

该肿瘤周围基质占据的图像像素分数用作每个瓷砖的显着性评分。32微米的距离是通过8,16,32和64微米的视觉比较确定的，发现32最接近于肿瘤周围基质的公认定义。计算TIL分数校准使用简单的线性校准将计算分数缩放到与视觉分数相似的幅度范围。

This calibration procedure first z-scores the visual and computational scores to identify outliers where disagreement is greater than 1.96 standard deviations. The remaining inliers are used to define a scaling factor between computational scores and visual scores using linear regression with no intercept.

该校准程序首先对视觉和计算分数进行z评分，以识别分歧大于1.96标准偏差的异常值。其余的内联用于使用无截距的线性回归定义计算分数和视觉分数之间的比例因子。

This scaling improves interpretability and enables the value of a threshold intended for pathologist TIL scores to be mapped to a corresponding threshold value for computational scores.Clinical outcomes analysisClinical data analysis used progression-free interval (PFI) as the endpoint used per recommendations from Liu et al.

这种缩放提高了可解释性，并使病理学家TIL分数的阈值能够映射到计算分数的相应阈值。临床结果分析临床数据分析使用无进展间隔（PFI）作为根据Liu等人的建议使用的终点。

for TCGA, with progression events including local and distant spread, recurrence, or death29. Kaplan–Meier curves were examined for patient subgroups using a TILs-score threshold of 10% for stromal TILs scores. While different thresholds are used in the literature, a 10% is often the defining threshold for a low TIL-score.

对于TCGA，进展事件包括局部和远处扩散，复发或死亡29。使用基质TILs评分的TILs评分阈值为10%，检查患者亚组的Kaplan–Meier曲线。虽然文献中使用了不同的阈值，但10%通常是低TIL分数的定义阈值。

For the nTnA score variant, a threshold of.

对于nTnA得分变体，阈值为。

Data availability

数据可用性

The PanopTILs dataset is made public at: https://sites.google.com/view/panoptils/.

PanopTILs数据集公开于：https://sites.google.com/view/panoptils/.

Code availability

代码可用性

Relevant code is publicly available at: github.com/PathologyDataScience/MuTILs_Panoptic.

相关代码可在以下网址公开获得：github.com/PathologyDataScience/MuTILs\u Panoptic。

ReferencesAbels, E. et al. Computational pathology definitions, best practices, and recommendations for regulatory guidance: a white paper from the Digital Pathology Association. J. Pathol. 249, 286–294 (2019).Article

参考文献Abels，E。等人，《计算病理学定义，最佳实践和监管指南建议：数字病理学协会的白皮书》。J、 病理学。。文章

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

van der Laak, J., Litjens, G. & Ciompi, F. Deep learning in histopathology: the path to the clinic. Nat. Med. 27, 775–784 (2021).Article

van der Laak，J.，Litjens，G。＆Ciompi，F。组织病理学的深度学习：通往临床的道路。《自然医学》27775-784（2021）。文章

PubMed

PubMed

Google Scholar

谷歌学者

Ballman, K. V. Biomarker: Predictive or Prognostic? J. Clin. Oncol. 33, 3968–3971 (2015).Article

Ballman，K.V。生物标志物：预测还是预后？J、 临床。Oncol公司。333968-3971（2015）。文章

CAS

中科院

PubMed

PubMed

Google Scholar

谷歌学者

Savas, P. et al. Clinical relevance of host immunity in breast cancer: from TILs to the clinic. Nat. Rev. Clin. Oncol. 13, 228–241 (2016).Article

Savas，P。等人。乳腺癌宿主免疫的临床相关性：从TIL到临床。国家修订临床。Oncol公司。。文章

CAS

中科院

PubMed

PubMed

Google Scholar

谷歌学者

Molavi, D. W. The Practice of Surgical Pathology: A Beginner’s Guide to the Diagnostic Process, 2nd ed. (Springer, 2017).Amin, M. B. et al. The Eighth Edition AJCC Cancer Staging Manual: Continuing to build a bridge from a population-based to a more “personalized” approach to cancer staging.

Molavi，D.W.《外科病理学的实践：诊断过程的初学者指南》，第二版（Springer，2017）。Amin，M.B.等人，《AJCC癌症分期手册》第八版：继续搭建一座桥梁，从基于人群的方法到更“个性化”的癌症分期方法。

CA Cancer J. Clin. 67, 93–99 (2017).Article .

癌症J.Clin。67、93-99（2017）。文章。

PubMed

PubMed

Google Scholar

谷歌学者

The Cancer Genome Atlas Network. Comprehensive molecular portraits of human breast tumours. Nature 490, 61–70 (2012).Fallahpour, S., Navaneelan, T., De, P. & Borgo, A. Breast cancer survival by molecular subtype: a population-based analysis of cancer registry data. CMAJ Open 5, E734–E739 (2017).Article .

癌症基因组图谱网络。人类乳腺肿瘤的综合分子肖像。《自然》490,61-70（2012）。Fallahpour，S.，Navaneelan，T.，De，P。＆Borgo，A。分子亚型乳腺癌生存：基于人群的癌症登记数据分析。CMAJ Open 5，E734–E739（2017）。文章。

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Salgado, R. et al. The evaluation of tumor-infiltrating lymphocytes (TILs) in breast cancer: recommendations by an International TILs Working Group 2014. Ann. Oncol. 26, 259–271 (2015).Article

Salgado，R.等人，《乳腺癌中肿瘤浸润淋巴细胞（TILs）的评估：2014年国际TILs工作组的建议》。安科。26259-271（2015）。文章

CAS

中科院

PubMed

PubMed

Google Scholar

谷歌学者

Kos, Z. et al. Pitfalls in assessing stromal tumor infiltrating lymphocytes (sTILs) in breast cancer. NPJ Breast Cancer 6, 17 (2020).Article

Kos，Z.等人。评估乳腺癌中基质肿瘤浸润淋巴细胞（STIL）的缺陷。NPJ乳腺癌6,17（2020）。文章

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Amgad, M. et al. Report on computational assessment of Tumor Infiltrating Lymphocytes from the International Immuno-Oncology Biomarker Working Group. NPJ Breast Cancer 6, 16 (2020).Article

Amgad，M.等人，国际免疫肿瘤学生物标志物工作组关于肿瘤浸润淋巴细胞计算评估的报告。NPJ乳腺癌6,16（2020）。文章

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Amgad, M. et al. Structured crowdsourcing enables convolutional segmentation of histology images. Bioinformatics 35, 3461–3467 (2019).Article

。生物信息学353461-3467（2019）。文章

CAS

中科院

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Amgad, M. et al. Explainable nucleus classification using Decision Tree Approximation of Learned Embeddings. Bioinformatics 38, 513–519 (2022).Article

Amgad，M.等人。使用学习嵌入的决策树近似进行可解释的核分类。生物信息学38513-519（2022）。文章

CAS

中科院

PubMed

PubMed

Google Scholar

谷歌学者

Rudin, C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nat. Mach. Intell. 1, 206–215 (2019).Article

鲁丁（RudinC.）停止解释高风险决策的黑箱机器学习模型，而是使用可解释的模型。自然马赫数。因特尔。1206-215（2019）。文章

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Kundu, S. AI in medicine must be explainable. Nat. Med. 27, 1328 (2021).Article

Kundu，S.AI在医学上必须是可以解释的。《自然医学》271328（2021）。文章

CAS

中科院

PubMed

PubMed

Google Scholar

谷歌学者

Mobadersany, P., et al. Predicting cancer outcomes from histology and genomics using convolutional networks. Proc. Natl Acad. Sci. USA 115, E2970–E2979 (2018)Wei Koh, P. et al. Concept Bottleneck Models. In: Proceedings of the 37th International Conference on Machine Learning, 5338–5348.

Mobadersany，P。等人。使用卷积网络从组织学和基因组学预测癌症结果。程序。国家科学院。科学。美国115，E2970–E2979（2018）Wei Koh，P.等人。概念瓶颈模型。在：第37届国际机器学习会议论文集，5338-5348。

(PMLR, 2020).Amgad, M. et al. Joint region and nucleus segmentation for characterization of tumor infiltrating lymphocytes in breast cancer. In: Medical Imaging 2019: Digital Pathology (eds Tomaszewski, J. E., Ward, A. D.) 129–136 (SPIE, 2019)Kirillov, A., He, K., Girshick, R., Rother, C.& Dollar, P.

（PMLR，2020）。Amgad，M.等。乳腺癌肿瘤浸润淋巴细胞特征的联合区域和细胞核分割。在：医学影像2019：数字病理学（编辑Tomaszewski，J.E.，Ward，A.D。）129-136（SPIE，2019）Kirillov，A.，He，K.，Girshick，R.，Rother，C。＆Dollar，P。

Panoptic Segmentation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 9404–9413 (CVPR, 2019).Ronneberger, O., Fischer, P.& Brox, T. U-Net: Convolutional Networks for Biomedical Image Segmentation. Medical Image Computing and Computer-Assisted Intervention (MICCAI).

全景分割。在：IEEE/CVF计算机视觉和模式识别会议论文集（CVPR）9404-9413（CVPR，2019）。Ronneberger，O.，Fischer，P。＆Brox，T.U-Net：用于生物医学图像分割的卷积网络。医学图像计算和计算机辅助干预（MICCAI）。

Vol. 9351, 234–241 (Springer, LNCS, 2015)van Rijthoven, M., Balkenhol, M., Siliņa, K., van der Laak, J. & Ciompi, F. HookNet: Multi-resolution convolutional neural networks for semantic segmentation in histopathology whole-slide images. Med. Image Anal. 68, 101890 (2021).Article .

第9351234-241卷（Springer，LNCS，2015）van Rijthoven，M.，Balkenhol，M.，Siliņa，K.，van der Laak，J.＆Ciompi，F.HookNet：用于组织病理学全幻灯片图像语义分割的多分辨率卷积神经网络。医学图像肛门。68101890（2021）。文章。

PubMed

PubMed

Google Scholar

谷歌学者

Amgad, M. et al. NuCLS: A scalable crowdsourcing approach and dataset for nucleus classification and segmentation in breast cancer. Gigascience 11, giac037 (2022)Calle, E. E. et al. The American Cancer Society Cancer Prevention Study II Nutrition Cohort. Cancer 94, 2490–2501 (2002).Article .

Amgad，M。等人，《NuCLS：乳腺癌细胞核分类和分割的可扩展众包方法和数据集》。Gigascience 11，giac037（2022）Calle，E.E.等人，《美国癌症协会癌症预防研究II营养队列》。。文章。

PubMed

PubMed

Google Scholar

谷歌学者

Steyerberg, E. W. & Harrell, F. E. Prediction models need appropriate internal, internal–external, and external validation. J. Clin. Epidemiol. 69, 245–247 (2016).Article

Steyerberg，E.W。和Harrell，F.E。预测模型需要适当的内部，内部-外部和外部验证。J、 临床。流行病。69245-247（2016）。文章

PubMed

PubMed

Google Scholar

谷歌学者

Kitware Inc. Python modules to work with large multiresolution images. https://github.com/girder/large_image (2024).Marcolini, A. et al. histolab: A Python library for reproducible Digital Pathology preprocessing with automated testing. SoftwareX 20, 101237 (2022).Article

Kitware Inc.Python模块用于处理大型多分辨率图像。https://github.com/girder/large_image（2024年）。Marcolini，A。等人。histolab：一个Python库，用于通过自动测试进行可重复的数字病理学预处理。SoftwareX 20101237（2022）。文章

Google Scholar

谷歌学者

Gutman, D. A. et al. The Digital Slide Archive: A Software Platform for Management, Integration, and Analysis of Histology for Cancer Research. Cancer Res. 77, e75–e78 (2017).Article

Gutman，D.A.等人，《数字幻灯片档案：用于癌症研究的组织学管理，整合和分析的软件平台》。癌症研究77，e75-e78（2017）。文章

CAS

中科院

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Macenko M. et al. A method for normalizing histology slides for quantitative analysis. In: 2009 IEEE International Symposium on Biomedical Imaging: From Nano to Macro, 1107–1110 (IEEE, 2009).Liu, J. et al. An Integrated TCGA Pan-Cancer Clinical Data Resource to Drive High-Quality Survival Outcome Analytics.

Macenko M.等人。一种用于定量分析的组织学切片标准化方法。发表于：2009 IEEE国际生物医学成像研讨会：从纳米到宏观，1107-1110（IEEE，2009）。Liu，J.等人。一个综合的TCGA泛癌临床数据资源，用于推动高质量的生存结果分析。

Cell 173, 400–416.e11 (2018).Article .

细胞173400–416.e11（2018）。文章。

CAS

中科院

PubMed

PubMed

PubMed Central

公共医学中心

Google Scholar

谷歌学者

Download referencesAcknowledgementsThis work was supported by the U.S. NIH NCI grants U01CA220401 and U24CA19436201, NLM grant R01LM013523, and by the generosity of Ms. Jeanne Lombardo. We acknowledge support from Dr. David Gutman and the American Cancer Society, including Dr. Mia M.

。我们感谢David Gutman博士和包括Mia M博士在内的美国癌症协会的支持。

Gaudet, Dr. Samantha Puvanesarajah, Dr. Lauren Teras, James Hodge, and Elizabeth Bain.Author informationAuthor notesThese authors contributed equally: Shangke Liu, Mohamed Amgad.Authors and AffiliationsDepartment of Pathology, Northwestern University, Chicago, IL, USAShangke Liu, Mohamed Amgad, Deeptej More, Muhammad A.

高德，萨曼莎·普瓦内萨拉杰博士，劳伦·特拉斯博士，詹姆斯·霍奇和伊丽莎白·贝恩。作者信息作者注意到这些作者做出了同样的贡献：刘尚科，穆罕默德·阿姆加德。作者和附属机构西北大学病理学系，芝加哥，伊利诺伊州，美国刘尚科，穆罕默德·阿姆加德，迪普特杰·莫尔，穆罕默德·A。

Rathore & Lee A. D. CooperDepartment of Pathology, GZA-ZNA Ziekenhuizen, Antwerp, BelgiumRoberto SalgadoDivision of Research, Peter MacCallum Cancer Centre, Melbourne, VIC, AustraliaRoberto SalgadoAuthorsShangke LiuView author publicationsYou can also search for this author in.

Rathore＆Lee A.D.CooperDepartment of Pathology，GZA-ZNA Ziekenhuizen，安特卫普，BelgiumRoberto SalgadoDivision of Research，Peter MacCallum Cancer Center，墨尔本，维多利亚州，澳大利亚Aroberto Salgadouthorusshangke LiuView作者出版物您也可以在中搜索这位作者。

PubMed Google ScholarMohamed AmgadView author publicationsYou can also search for this author in

PubMed Google ScholarMohamed AmgadView作者出版物您也可以在

PubMed Google ScholarDeeptej MoreView author publicationsYou can also search for this author in

PubMed Google ScholarDeeptej MoreView作者出版物您也可以在

PubMed Google ScholarMuhammad A. RathoreView author publicationsYou can also search for this author in

PubMed Google Scholarmamuhammad A.RathoreView作者出版物您也可以在

PubMed Google ScholarRoberto SalgadoView author publicationsYou can also search for this author in

PubMed Google ScholarRoberto SalgadoView作者出版物您也可以在

PubMed Google ScholarLee A. D. CooperView author publicationsYou can also search for this author in

PubMed Google ScholarLee A.D.CooperView作者出版物您也可以在

PubMed Google ScholarContributionsM.A.: Idea conception, model implementation, validation, and manuscript review. S.L.: validation, manuscript writing. D.M.: code optimization, execution time measurements. M.A.R.: validation, manuscript writing. R.S.: manual scoring of TILs, manuscript approval.

PubMed谷歌学术贡献。A、 ：想法概念，模型实施，验证和手稿审查。S、 L.：验证，手稿写作。D、 M.：代码优化，执行时间测量。M、 A.R.：验证，手稿写作。R、 S.：TIL的手动评分，手稿批准。

L.A.D.C.: Idea conception, manuscript writing. Authors M.A. and S.L. made equal contributions.Corresponding authorsCorrespondence to.

五十、 A.D.C.：想法概念，手稿写作。作者M.A.和S.L.做出了同样的贡献。通讯作者通讯。

Mohamed Amgad or Lee A. D. Cooper.Ethics declarations

穆罕默德·阿姆加德或李·A·D·库珀。道德宣言

Competing interests

相互竞争的利益

The authors declare the following competing interests: L.A.D.C. has invention disclosures registered at the Northwestern Office of Innovation and New Ventures, consults for Tempus. R.S. has received research support from Merck, Roche, Puma; and travel/congress support from AstraZeneca, Roche, and Merck; and he has served as an advisory board member of BMS and Roche and consults for BMS.

作者声明了以下相互竞争的利益：L.A.D.C.在西北创新与新企业办公室注册了发明披露，Tempus的顾问。R、 美国已获得默克，罗氏，彪马的研究支持；以及阿斯利康，罗氏和默克的旅行/大会支持；他曾担任BMS和Roche的顾问委员会成员，并为BMS提供咨询。

The other authors declare no competing interests..

。。

Additional informationPublisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.Supplementary informationSupplementSupplementary Data 1Rights and permissions

Additional informationPublisher的注释Springer Nature在已发布地图和机构隶属关系中的管辖权主张方面保持中立。补充信息补充数据1权利和权限

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made.

。

The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.

本文中的图像或其他第三方材料包含在文章的知识共享许可中，除非在材料的信用额度中另有说明。如果材料未包含在文章的知识共享许可证中，并且您的预期用途未被法律法规允许或超出允许的用途，则您需要直接获得版权所有者的许可。

To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/..

要查看此许可证的副本，请访问http://creativecommons.org/licenses/by/4.0/..

Reprints and permissionsAbout this articleCite this articleLiu, S., Amgad, M., More, D. et al. A panoptic segmentation dataset and deep-learning approach for explainable scoring of tumor-infiltrating lymphocytes.

转载和许可本文引用本文Liu，S.，Amgad，M.，More，D。等人。用于肿瘤浸润淋巴细胞可解释评分的全景分割数据集和深度学习方法。

npj Breast Cancer 10, 52 (2024). https://doi.org/10.1038/s41523-024-00663-1Download citationReceived: 18 May 2023Accepted: 17 June 2024Published: 28 June 2024DOI: https://doi.org/10.1038/s41523-024-00663-1Share this articleAnyone you share the following link with will be able to read this content:Get shareable linkSorry, a shareable link is not currently available for this article.Copy to clipboard.

npj乳腺癌10,52（2024）。https://doi.org/10.1038/s41523-024-00663-1Download引文接收日期：2023年5月18日接收日期：2024年6月17日发布日期：2024年6月28日OI：https://doi.org/10.1038/s41523-024-00663-1Share本文与您共享以下链接的任何人都可以阅读此内容：获取可共享链接对不起，本文目前没有可共享的链接。复制到剪贴板。

Provided by the Springer Nature SharedIt content-sharing initiative

由Springer Nature SharedIt内容共享计划提供