Hitchhiker's guide to cancer-associated lymphoid aggregates in histology images: manual and deep learning-based quantification approaches

Abstract

images: manual and deep learning-based quantification approaches Authors: ['Karīna Siliņa' 'Francesco Ciompi'] QuanAficaAon of lymphoid aggregates including terAary lymphoid structures with germinal centers in histology images of cancer is a promising approach for developing prognosAc and predicAve Assue biomarkers. In this arAcle, we provide recommendaAons for idenAfying lymphoid aggregates in Assue secAons from rouAne pathology workflows such as hematoxylin and eosin staining. To overcome the intrinsic variability associated with manual image analysis (such as subjecAve decision making, a[enAon span), we recently developed a deep learningbased algorithm called HookNet-TLS to detect lymphoid aggregates and germinal centers in various Assues. Here, we addiAonally provide a guideline for using manually annotated images for training and implemenAng HookNet-TLS for automated and objecAve quanAficaAon of lymphoid aggregates in various cancer types.