Abstract
Esophageal squamous cell carcinoma (ESCC) lacks a standardized classification system, resulting in inconsistent clinical management and a suboptimal prognosis. This study addresses the urgent need for a robust consensus taxonomy to facilitate precision treatment for ESCC. We employed a network-based approach to elucidate the interconnections among eight existing classification systems, leading to the identification of four distinct consensus molecular subtypes (ECMSs): ECMS1-MET (metabolic), characterized by dysregulated metabolic pathways and NFE2L2 activation; ECMS2-CLS (classical), featuring upregulated cell cycle and canonical signaling pathways; ECMS3-IM (immunomodulatory), marked by robust immune activation and elevated PD-1 expression; and ECMS4-MES (mesenchymal), associated with mesenchymal transition, stromal activation, and VEGF signaling. To improve clinical applicability, we developed an image-based framework (imECMS) that utilizes spatial organization features (SOFs) quantified from autodelineated hematoxylin‒eosin (H&E)-stained whole-slide images through deep learning algorithms. The imECMS classifier assigns patients to one of the four ECMS subtypes, which correlate with distinct molecular characteristics, prognoses, and responses to neoadjuvant chemotherapy and immunotherapy. Validation across multiple independent cohorts confirmed that the imECMS accurately classifies ESCC subtypes from histopathological images, offering a robust and effective tool for precision medicine. In summary, the ECMS/imECMS subtyping systems we developed are the most robust frameworks for ESCC to date, providing clear biological insights and a foundation for clinical stratification and targeted therapies.