MetAssist 2.0: A Generalizable Artificial Intelligence Framework for Lymph Node Metastasis Detection Across Multiple Cancer Types.

Abstract

Lymph node metastasis assessment is critical for cancer staging, yet the process is labor intensive and prone to variability, particularly when evaluating micrometastases or isolated tumor cells that can directly alter treatment decisions. We present MetAssist 2.0, a modular artificial intelligence system that combines a pathology foundation model with transformer-based segmentation to automate metastasis detection across cancer types. Trained on colorectal and upper gastrointestinal cancers, it was validated on 8144 slides spanning 7 cancer types and 14 multi-institutional cohorts. MetAssist 2.0 achieved at least 90% sensitivity in 13 cohorts and 91% specificity in 11 cohorts, including challenging subtypes such as mucinous adenocarcinoma and tumor deposits. With only 10 annotated slides, the system adapted to unseen cancer types via few-shot fine-tuning. As a triage tool for colorectal cancer, it could reduce pathologist workload by up to 72% with 98% sensitivity and revealed metastases missed in routine reporting. These results demonstrate broad generalizability and near-clinical-grade performance, positioning MetAssist 2.0 for integration into the pathology workflows.