Predicting gastric cancer outcome from resected lymph node histopathology images using deep learning

TLDR

N‑staging is crucial for prognosis and treatment decisions, yet visual inspection of lymph‑node whole‑slides remains the primary method for counting metastatic nodes, and outcomes vary widely even within the same N stage. The study proposes a deep‑learning framework to analyze lymph‑node whole‑slide images, identify lymph nodes and tumor regions, and compute a tumor‑area‑to‑MLN‑area ratio. The framework uses deep learning to detect lymph nodes and tumor regions in WSIs and calculates the T/MLN ratio. The model’s tumor detection matched experienced pathologists and performed similarly on two independent gastric‑cancer cohorts, and the T/MLN ratio emerged as an interpretable independent prognostic factor, indicating deep‑learning can aid pathologists and oncologists in detecting metastatic nodes and identifying new prognostic markers.

Abstract

Abstract N-staging is a determining factor for prognostic assessment and decision-making for stage-based cancer therapeutic strategies. Visual inspection of whole-slides of intact lymph nodes is currently the main method used by pathologists to calculate the number of metastatic lymph nodes (MLNs). Moreover, even at the same N stage, the outcome of patients varies dramatically. Here, we propose a deep-learning framework for analyzing lymph node whole-slide images (WSIs) to identify lymph nodes and tumor regions, and then to uncover tumor-area-to-MLN-area ratio (T/MLN). After training, our model’s tumor detection performance was comparable to that of experienced pathologists and achieved similar performance on two independent gastric cancer validation cohorts. Further, we demonstrate that T/MLN is an interpretable independent prognostic factor. These findings indicate that deep-learning models could assist not only pathologists in detecting lymph nodes with metastases but also oncologists in exploring new prognostic factors, especially those that are difficult to calculate manually.