Development and validation of a novel YOLOv5-based artificial intelligence model for gastric mucosal lesion detection
An AI system detects two types of gastric lesions simultaneously with near-expert accuracy, potentially catching early stomach cancers in screening.
Endosmart, a YOLOv5-based AI system trained on 34,979 gastroscopic images, enables simultaneous real-time detection of both diffuse and focal gastric mucosal lesions with AUC up to 0.990 in external validation against senior endoscopists. This addresses a gap where prior AI endoscopy tools handle only one lesion type, potentially improving early gastric cancer detection rates in screening programs.
What the study was
- Study design
- Prospective development and external validation, AI diagnostic model
- Population
- Endoscopy patients at two Chinese tertiary hospitals
- Sample size
- 34979
- Category
- Diagnostics
- Maturity
- Validated
- Journal
- Surg Endosc
Why it surfaced
Large training set (34,979 images), external validation included; simultaneous diffuse+focal lesion detection is novel vs. prior tools; high AUC with endoscopist comparison.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.