Landmark CUHK study establishes first consensus molecular subtypes for ESCC AI-powered tool imECMS developed to advance and popularise precision medicine
A research team led by Dr Wang Xin (front row, left), Associate Professor, and Dr Cui Heyang (front row, right), Assistant Professor, from the Department of Surgery at CU Medicine, together with collaborators from Shanxi Medical University, Shenzhen Bay Laboratory, Peking University Shenzhen Hospital, and Peking University Cancer Hospital, has established the first robust consensus molecular subtype system for ESCC. The team also developed an innovative AI-powered diagnostic tool, “imECMS,” which overcomes the limitations of costly and complex genetic testing, streamlining clinical application and paving the way for precision medicine.
Esophageal squamous cell carcinoma (ESCC) accounts for about 90% of esophageal cancer cases worldwide, with especially high incidence in Asia. The absence of a standardised classification system has long hindered clinicians from providing precise, personalised treatment. The Faculty of Medicine at The Chinese University of Hong Kong (CU Medicine) and its collaborators have established the first robust consensus molecular subtype system for ESCC and introduced an innovative AI-powered tool, imECMS, that can classify patients using routine pathology slides. This breakthrough overcomes the limitations of costly, complex genetic testing, significantly streamlining clinical application and advancing smarter drug development and improved patient outcomes. The study was published in the renowned journal Signal Transduction and Targeted Therapy, part of the Nature stable.
At present, the complex biological heterogeneity of ESCC has prevented the establishment of a unified classification standard, making it difficult to develop precise, personalised treatment plans and negatively affecting patient prognosis. There is an urgent need for a robust consensus taxonomy to guide ESCC treatment. In response, Dr Wang Xin, Associate Professor in the Department of Surgery at CU Medicine, together with teams from Shanxi Medical University, Shenzhen Bay Laboratory, Peking University Shenzhen Hospital and Peking University Cancer Hospital, has developed the ESCC subtyping system.
First-ever unified classification system for ESCC
The team first conducted similarity network fusion analysis on multi-omics data from 152 ESCC patients – including whole-genome sequencing, RNA sequencing and DNA methylation profiling. Using network analysis and Markov cluster algorithms, they then systematically integrated eight major subtyping systems from the literature. This rigorous approach distilled the field’s complexity into four highly stable, cross-cohort, consistent, consensus molecular subtypes (ECMS 1-4), each with distinct biological characteristics and therapeutic implications.
Dr Wang, co-corresponding author of the study, stated: “ESCC has long suffered from fragmented knowledge, as different research groups developed their own classification systems based on different data types and methodologies, creating fragmented knowledge that could not guide clinical decisions. We needed to integrate these disparate voices into a unified consensus that could actually inform patient care.”
ECMS subtype | Biological characteristics & therapeutic implications |
|---|---|
ECMS1 (Metabolic subtype) | Shows metabolic pathway dysregulation and may respond poorly to certain chemotherapies but has potential sensitivity to FGFR or NTRK inhibitors. |
ECMS2 (Classical subtype) | Characterised by cell cycle activation. These patients tend to respond well to chemotherapy agents, like cisplatin and paclitaxel, and may benefit from CDK or HER2-targeted therapies. |
ECMS3 (Immunomodulatory subtype) | Exhibits strong immune activation and high PD-1 expression. Importantly, these patients do not respond well to standard chemoradiotherapy but show significantly better responses to immune checkpoint inhibitors (anti-PD-1 therapy). |
ECMS4 (Mesenchymal subtype) | Associated with the poorest prognosis and high risk of recurrence. These patients may benefit more from anti-angiogenic therapies or matrix metalloprotease inhibitors than from conventional chemotherapy. |
AI-powered tool imECMS makes precision medicine accessible
Recognising that molecular sequencing is costly and often inaccessible, the team developed an AI-driven diagnostic tool, imECMS, to translate their findings into clinical practice. It can accurately identify all four cancer subtypes using routinely performed H&E-stained pathological slides. By extracting quantifiable spatial organisation features from standard slides, the tool delivers results in minutes, greatly reducing both time and cost compared to RNA sequencing, and eliminating the need for additional patient samples. This enables physicians to develop more precise treatment plans.
While imECMS has performed impressively in validation cohorts and across different Chinese patient groups, it is currently limited by a relatively small sample size and its reliance on the quality of H&E-stained images. The team believes that further optimisation and validation with larger, more diverse data sets – including local patient data – will benefit clinical deployment. Nonetheless, the team remains confident that imECMS has the potential to become an effective tool for cancer subtyping in the future.
The collaborative research team performed similarity network fusion analysis on multi-omics data from ESCC patients, systematically integrating eight major subtyping systems from the literature into four highly stable and cross-cohort consistent consensus molecular subtypes (ECMS1–4). Building on this first unified classification system for ESCC, the team developed an AI-driven diagnostic tool, imECMS, which uses routinely performed H&E-stained pathological slides to accurately identify all four cancer subtypes. This enables physicians to design more precise treatments tailored to each subtype’s distinct biological and therapeutic features.
Professor Brigette Ma, Frances Lee Professor of Precision Oncology and Professor in the Department of Clinical Oncology at CU Medicine, said: “Currently, clinical decisions for ESCC patients are largely empirical. With imECMS, subtype information can be obtained from routine H&E slides, making clinical implementation practical and improving the ability to identify the most effective treatment options for patients.”
Dr Cui Heyang, first author of the study and Assistant Professor at CU Medicine, said: “This transforms subtyping from an expensive research tool into a practical clinical asset. Any hospital with standard pathology capabilities can now access precision insights that were previously limited to well-funded research centres. It’s about democratising precision medicine.”
Professor Yongping Cui of Shanxi Medical University, co-corresponding author of the study, added: “This isn’t just another classification system – it’s a consensus framework that reconciles and transcends previous work. By identifying robust subtype-specific vulnerabilities, we’ve created a roadmap for developing targeted therapies where few exist today. The integration of molecular consensus with AI-powered pathology represents the future of precision oncology. We’re particularly excited about the potential to implement this in resource-limited settings where esophageal cancer burden is highest.”
