Artificial Intelligence-Assisted Whole Slide Image Analysis for Lymph Node Status Prediction in Early Colorectal and Gastric Cancer

• Published in: Digestive endoscopy
• Main Authors: Ichimasa, Katsuro, Kudo, Shin-Ei, Kouyama, Yuta, Takashina, Yuki, Chung, Hyunsoo, Maeda, Yasuharu, Lwin, Wai Phyo, Toya, Yosuke, Hatta, Waku, So, Jimmy Bok Yan, Yeoh, Khay Guan, Nemoto, Tetsuo, Misawa, Masashi
• Format: Journal Article
• Language: English
• Published: Australia 27.09.2025
• Subjects: pathology; stomach neoplasms; lymph nodes; colorectal neoplasms; artificial intelligence
• ISSN: 1443-1661; 1443-1661
• DOI: 10.1111/den.70042

With the widespread use of advanced endoscopic techniques such as endoscopic submucosal dissection, an increasing number of early colorectal cancer (T1 CRC) and early gastric cancer (EGC) cases are now treated with endoscopic resection as the first-line approach. However, the risk of lymph node metastasis (LNM)-approximately 10% in T1 CRC and 5%-10% in EGC-necessitates additional surgical resection in high-risk cases. Current guideline-based risk stratification depends on pathological evaluation of the resected specimens to determine whether further surgery is needed. Yet both T1 CRC and EGC face shared challenges in LNM risk prediction, particularly in terms of accuracy and reproducibility. This review focuses on the latter. The diagnosis of key pathological risk factors, which serve as predictors of LNM, is subject to considerable interobserver variability among pathologists. One potential solution is the application of artificial intelligence (AI)-assisted whole slide image (WSI) analysis, which has been gaining attention in recent studies. AI-assisted models for LNM prediction in T1 CRC and EGC have shown encouraging results, suggesting that WSI-based AI could offer a pathologist-independent strategy to improve diagnostic consistency. However, the field remains in an early stage, with key limitations including small sample sizes and limited external validation. Additional high-quality evidence will be needed to support clinical implementation. Addressing challenges such as stain standardization and image artifacts will also be critical for achieving regulatory approval and broader clinical adoption.