Multimodal AI Model Predicts Lymph Node Metastasis in Colorectal Cancer

A multimodal AI model (CT, pathology) predicted lymph node metastasis in colorectal cancer with AUC 0.922 in internal testing, 0.872/0.848 in external cohorts, and 0.845 in a prospective trial. It boosted junior pathologist accuracy from 95.0% to 97.3%.

• DeepLNM integrates H&E-stained whole-slide images, CT scans, and clinical text; it was evaluated across retrospective, external, and prospective cohorts from three hospitals.
• Prospective clinical trial (NCT07239063) performance remained robust at AUC 0.845; the model also generalized to gastric cancer (AUC 0.793).
• In a user study, the model improved junior pathologists' diagnostic accuracy from 95.0% to 97.3%, while reducing reading time and increasing confidence.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.