Nikhil Cherian Kurian, Peter H Gann, Neeraj Kumar, Stephanie M McGregor, Ruchika Verma, Amit Sethi
{"title":"Deep Learning Predicts Subtype Heterogeneity and Outcomes in Luminal A Breast Cancer Using Routinely Stained Whole-Slide Images.","authors":"Nikhil Cherian Kurian, Peter H Gann, Neeraj Kumar, Stephanie M McGregor, Ruchika Verma, Amit Sethi","doi":"10.1158/2767-9764.CRC-24-0397","DOIUrl":null,"url":null,"abstract":"<p><strong>Significance: </strong>A deep learning model, trained using transcriptomic data, inexpensively quantifies and fine-maps ITH due to subtype admixture in routine images of LumA breast cancer, the most favorable subtype. This new approach could facilitate exploration of the mechanisms behind such heterogeneity and its impact on selection of therapy for individual patients.</p>","PeriodicalId":72516,"journal":{"name":"Cancer research communications","volume":" ","pages":"157-166"},"PeriodicalIF":2.0000,"publicationDate":"2025-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11770635/pdf/","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Cancer research communications","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1158/2767-9764.CRC-24-0397","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"ONCOLOGY","Score":null,"Total":0}
引用次数: 0
Abstract
Significance: A deep learning model, trained using transcriptomic data, inexpensively quantifies and fine-maps ITH due to subtype admixture in routine images of LumA breast cancer, the most favorable subtype. This new approach could facilitate exploration of the mechanisms behind such heterogeneity and its impact on selection of therapy for individual patients.
肿瘤内异质性(ITH)对精确肿瘤学提出了挑战,但在人群水平上可扩展的空间量化方法尚不存在。基于先前的研究表明PAM50亚型的混合可以使用转录组学数据从组织中测量,我们训练了一个深度神经网络(DNN)来量化Luminal a (LumA)乳腺癌的ITH亚型,从常规染色的整张幻灯片图像中。我们检验了在图像中检测到的亚型混合与肿瘤侵袭性和不良后果相关的假设。在来自TCGA-BRCA队列的680例病例中,我们通过对每个转录组应用矩阵分解来确定对指定亚型的依从性。最纯粹的案例被分成几组进行初始测试、训练和参数调整。230个luma分配的案例进行了最终测试。将图像块输入到对组织学图像进行预训练的DNN中。我们测量了图像中被分类为LumA的肿瘤面积与肿瘤特征和生存率的相关性。在luma指定的病例中,混合物与稍高的er阳性、较低的pr阳性和er相关基因表达、较高的her2阳性、肿瘤大小、分级和TNM分期相关。图像混合与更多的TP53和更少的PIK3CA突变相关。混合病例的无进展生存期明显缩短。我们的研究结果表明,经过训练以识别组织形态学中的基因组相关性的深度学习可以量化和绘制具有临床意义的LumA乳腺癌亚型混合。这种方法的低成本和可扩展性有潜力作为研究ITH的研究工具,并可能提高精确肿瘤学的疗效。