A hybrid 2-stage vision transformer for artificial intelligence-assisted 5 class pathologic diagnosis of gastric endoscopic biopsies: a diagnostic tool for guiding gastric cancer treatment

Gastric endoscopic screening is an effective way to decide appropriate gastric cancer (GC) treatment at an early stage, reducing GC-associated mortality rate. Although artificial intelligence (AI) has brought a great promise to assist pathologist to screen digitalized whole slide images, automatic classification systems for guiding proper GC treatment based on clinical guideline are still lacking. We propose an AI system classifying 5 classes of GC histology, which can be perfectly matched to general GC treatment guidance. The AI system was designed to mimic the way pathologist understand slides through multi-scale self-attention mechanism using a 2-stage Vision Transformer network. The AI system performance was evaluated on 876 internal endoscopic slides and 336 external endoscopic slides from clinical cohort. We further evaluated practical usability of the AI system on observation of AI-assisted 6 pathologist performance. The AI system demonstrates clinical capability by achieving class-average diagnostic sensitivity of above 85% for both internal and external cohort analysis. Furthermore, AI-assisted pathologists showed significantly improved diagnostic sensitivity by 10% within 18% saved screening time compared to human pathologists (p-values of 0.006 and 0.030, respectively). The reliable performance of the AI system in multi-center cohort testing and its clinical applicability demonstrate that AI-assisted endoscopic CG screening would help reduce the workload of limited pathologists. Furthermore, the AI system has a great potential for providing presumptive pathologic opinion for deciding proper treatment for early GC patients.