A Deep Learning Framework for Predicting Prognostically Relevant Consensus Molecular Subtypes in HPV-Positive Cervical Squamous Cell Carcinoma from Routine Histology Images

Despite efforts in human papillomavirus (HPV) prevention and screening, cervical cancer remains the fourth most prevalent cancer among women globally. In this study, we propose an end-to-end deep learning framework to investigate histological correlates of the two consensus molecular subtypes (CMS) of HPV-positive cervical squamous cell carcinoma (CSCC) patients. Analysing three international CSCC cohorts (n=545 patients), we demonstrate that the genomically determined CMS can be predicted from routine H&E-stained histology slides, with our Digital-CMS scores achieving significant patient stratifications in terms of disease-specific survival (TCGA p=0.0022, Oslo p=0.0495) and disease-free survival (TCGA p=0.0495, Oslo p=0.0282). In addition, our extensive analyses reveal distinct tumour microenvironment (TME) differences between the two CMS subtypes of the CSCC cohorts. Notably, CMS-C1 CSCC subgroup has markedly increased lymphocyte presence, whereas CMS-C2 subgroup has high nuclear pleomorphism, an elevated neutrophil-to-lymphocyte ratio, and increased neutrophil density. Analysis of representative histological regions reveals higher degree of malignancy in CMS-C2 patients, associated with poor prognosis. This study introduces a potentially clinically advantageous Digital-CMS score derived from digitised WSIs of routine H&E-stained tissue sections, offers new insights into TME differences impacting patient prognosis and potential therapeutic targets, and identifies histological patterns serving as potential surrogate markers of the two CMS subtypes for clinical application.