Determining breast cancer biomarker status and associated morphological features using deep learning.
Determining breast cancer biomarker status and associated morphological features using deep learning.
复制标题
DOI：
10.1038/s43856-021-00013-3
复制
发表时间：
2021
期刊：
影响因子：
--
通讯作者：
Chen, Po-Hsuan Cameron
中科院分区：
文献类型：
--
作者：
Gamble, Paul;Jaroensri, Ronnachai;Wang, Hongwu;Tan, Fraser;Moran, Melissa;Brown, Trissia;Flament-Auvigne, Isabelle;Rakha, Emad A.;Toss, Michael;Dabbs, David J.;Regitnig, Peter;Olson, Niels;Wren, James H.;Robinson, Carrie;Corrado, Greg S.;Peng, Lily H.;Liu, Yun;Mermel, Craig H.;Steiner, David F.;Chen, Po-Hsuan Cameron
乳腺癌的治疗依赖于生物标志物，包括雌激素受体、孕激素受体和人表皮生长因子受体2（ER/PR/HER2）。虽然现有的评分系统被广泛使用和验证，但它们可能涉及昂贵的准备和可变的解释。此外，组织学和预期生物标志物结果之间的不一致可能促使重复检测，以解决非预期结果的生物学、解释性或技术原因。我们开发了三个独立的深度学习系统（DLS），使用苏木精和曙红染色（H & E）图像作为输入，直接预测病灶组织区域（斑块）和载玻片的ER/PR/HER2状态。使用来自三个数据源的病理学家注释的载玻片对模型进行训练和评价。在贴片水平（> 1.35亿片，181个载玻片）和载玻片水平（n = 3274个载玻片，1249例病例，37个部位）计算测试集的受试者工作特征曲线下面积（AUC）。使用概念激活向量（TCAV）测试、显著性分析和聚集斑块的病理学家审查进行可解释性分析。ER/PR/HER2的贴剂水平AUC分别为0.939（95%CI 0.936 - 0.941）、0.938（0.936 - 0.940）和0.808（0.802 - 0.813）。在玻片水平，ER/PR/HER2的AUC分别为0.86（95%CI 0.84 - 0.87）、0.75（0.73 - 0.77）和0.60（0.56 - 0.64）。可解释性分析显示了已知的生物标志物-组织形态学相关性，包括低级别和小叶组织学与ER/PR阳性的相关性，以及炎性浸润增加与三阴性染色的相关性。这项研究提出了快速乳腺癌生物标志物估计从常规的H & E幻灯片，并建立在现有的病理知识的背景下，通过优先考虑计算学习功能的可解释性先前的进展。乳腺癌的诊断和表征涉及评估肿瘤细胞内部或表面上发现的标志物蛋白。三个最重要的标志物是雌激素受体（ER），孕激素受体（PR）和一种称为HER2的受体。这些标志物的水平可以影响乳腺癌患者在临床上的治疗方式。这项研究探索了机器学习的能力-计算机软件被训练来识别和分类特定的图像特征-以确定这些标记在数字化图像中的状态，而不需要组织染色。我们的研究结果表明，机器学习可以自动预测病理图像中ER、PR和HER2的状态，进一步的测试可以识别出能够实现这些预测的特定图像特征。这种类型的方法可以降低成本和时间表，并能够改善标记物检测的质量控制。Gamble和Jaroensri等人开发了深度学习系统，使用H & E图像预测乳腺癌生物标志物状态。他们的模型能够对ER，PR和HER2进行切片水平和斑块水平的预测，可解释性分析突出了与这些标志物相关的特定组织学特征。
Breast cancer management depends on biomarkers including estrogen receptor, progesterone receptor, and human epidermal growth factor receptor 2 (ER/PR/HER2). Though existing scoring systems are widely used and well-validated, they can involve costly preparation and variable interpretation. Additionally, discordances between histology and expected biomarker findings can prompt repeat testing to address biological, interpretative, or technical reasons for unexpected results. We developed three independent deep learning systems (DLS) to directly predict ER/PR/HER2 status for both focal tissue regions (patches) and slides using hematoxylin-and-eosin-stained (H&E) images as input. Models were trained and evaluated using pathologist annotated slides from three data sources. Areas under the receiver operator characteristic curve (AUCs) were calculated for test sets at both a patch-level (>135 million patches, 181 slides) and slide-level (n = 3274 slides, 1249 cases, 37 sites). Interpretability analyses were performed using Testing with Concept Activation Vectors (TCAV), saliency analysis, and pathologist review of clustered patches. The patch-level AUCs are 0.939 (95%CI 0.936–0.941), 0.938 (0.936–0.940), and 0.808 (0.802–0.813) for ER/PR/HER2, respectively. At the slide level, AUCs are 0.86 (95%CI 0.84–0.87), 0.75 (0.73–0.77), and 0.60 (0.56–0.64) for ER/PR/HER2, respectively. Interpretability analyses show known biomarker-histomorphology associations including associations of low-grade and lobular histology with ER/PR positivity, and increased inflammatory infiltrates with triple-negative staining. This study presents rapid breast cancer biomarker estimation from routine H&E slides and builds on prior advances by prioritizing interpretability of computationally learned features in the context of existing pathological knowledge. Breast cancer diagnosis and characterization involves evaluation of marker proteins found inside or on the surface of tumor cells. Three of the most important markers are estrogen receptor (ER), progesterone receptor (PR) and a receptor called HER2. The levels of these markers can influence how a person with breast cancer is treated in the clinic. This study explored the ability of machine learning – whereby computer software is trained to recognise and classify particular image features - to determine the status of these markers in digitized images, without the need for tissue stains. Our results demonstrate that machine learning can automatically predict the status of ER, PR and HER2 in pathology images and further testing identifies specific image features which enable these predictions. This type of approach may decrease costs and timelines and enable improved quality control in marker detection. Gamble and Jaroensri et al. develop deep learning systems to predict breast cancer biomarker status using H&E images. Their models enable slide-level and patch-level predictions for ER, PR and HER2, with interpretability analyses highlighting specific histological features associated with these markers.
登录
查看更多内容
影响因子：
3.5
作者：
Cohen, David A.;Dabbs, David J.;Bhargava, Rohit
通讯作者：
Bhargava, Rohit
影响因子：
8.8
作者：
Li, CI;Uribe, DJ;Daling, JR
通讯作者：
Daling, JR
影响因子：
15.2
作者：
Wulczyn E;Steiner DF;Moran M;Plass M;Reihs R;Tan F;Flament-Auvigne I;Brown T;Regitnig P;Chen PC;Hegde N;Sadhwani A;MacDonald R;Ayalew B;Corrado GS;Peng LH;Tse D;Müller H;Xu Z;Liu Y;Stumpe MC;Zatloukal K;Mermel CH
通讯作者：
Mermel CH
影响因子：
64.5
作者：
Ciriello G;Gatza ML;Beck AH;Wilkerson MD;Rhie SK;Pastore A;Zhang H;McLellan M;Yau C;Kandoth C;Bowlby R;Shen H;Hayat S;Fieldhouse R;Lester SC;Tse GM;Factor RE;Collins LC;Allison KH;Chen YY;Jensen K;Johnson NB;Oesterreich S;Mills GB;Cherniack AD;Robertson G;Benz C;Sander C;Laird PW;Hoadley KA;King TA;TCGA Research Network;Perou CM
通讯作者：
Perou CM
影响因子：
45.3
作者：
Allison, Kimberly H.;Hammond, M. Elizabeth H.;Wolff, Antonio C.
通讯作者：
Wolff, Antonio C.