Development and validation of an AI foundation model for endoscopic diagnosis of esophagogastric junction adenocarcinoma: a cohort and deep learning study.
Source check: Source checked. Not independent scientific confirmation.
Finding: Multicentre study of esophagogastric junction adenocarcinoma staging: 12,302 images/1,546 patients from seven hospitals; DINOv2 and ResNet50 encoders are gated in a mixture-of-experts classifier. Pathology is the reference; patient-disjoint held-out, external, and prospective cohorts plus an endoscopist reader comparison support a bounded diagnostic finding. It does not establish broader clinical benefit; data are from seven Chinese hospitals and image-level training is discussed. The arXiv v1 is the first located result-bearing version; its prospective accuracy was 0.8956, compared with 0.8963 in the later journal version.
Technology: DINOv2 extracts global appearance and ResNet50 local detail; element-wise gating/MoE fuses representations before three-class classification.
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- DINOv2 extracts global appearance and ResNet50 local detail; element-wise gating/MoE fuses representations before three-class classification.
- Test and result
- Results at image and patient levels report accuracy/AUC across held-out, external and prospective sets and comparison with endoscopists.
- Limitation / still unknown
- Histopathology defines early/advanced disease; held-out 112-patient, external 230-patient and prospective 198-patient sets; human reader comparisons. Regional sampling and image-level training limit generalizability; no patient outcome tested.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://arxiv.org/abs/2509.17533v1 · https://pmc.ncbi.nlm.nih.gov/articles/PMC12510009/ · https://pubmed.ncbi.nlm.nih.gov/41079024/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 22 Nov 2026. Review scope: Primary result-bearing material for the stated method, result and evaluation, with bibliographic version information..
No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.