One-shot synthesis of rare gastrointestinal lesions improves diagnostic accuracy and clinical training
Abstract
Rare gastrointestinal lesions are infrequently encountered in routine endoscopy, limiting data for developing reliable artificial intelligence (AI) models and training novice clinicians. Here we present EndoRare, a one-shot, retraining-free generative framework that synthesizes diverse lesion exemplars from a single reference image. Using language-guided concept disentanglement, EndoRare separates pathognomonic lesion features from non-diagnostic attributes, encoding the former into a learnable prototype embedding while varying the latter to ensure diversity. We validate EndoRare across four rare pathologies: calcifying fibrous tumor, juvenile polyposis syndrome, familial adenomatous polyposis, and Peutz-Jeghers syndrome. Experts judge the synthetic images clinically plausible, and data augmentation significantly improves downstream AI classifiers, increasing true positive rates at low false-positive rates. In an exploratory reader study, exposure to EndoRare-generated cases is associated with improved image-level recognition by novice endoscopists. These findings support the potential use of EndoRare as a data-efficient tool for rare-lesion education and motivate prospective controlled validation.
Related
- Diagnostic Accuracy of a Novel Chemiluminescent Immunoassay for Strongyloides stercoralis in a Non-Endemic Clinical Setting
- Thymus serpyllum Extract Improves Gastrointestinal Symptoms and Modulates Fecal Gut Microbiome Composition in a Multicenter, Randomized, Placebo-Controlled Clinical Trial.
- Diagnostic accuracy of tongue diagnosis for stroke syndrome differentiation in traditional Chinese medicine: a systematic review, meta-analysis, and narrative synthesis of artificial intelligence-assisted models.
- Serum Cystatin 4 combined with routine clinical laboratory indicators: A support vector machine diagnostic model for early screening of gastrointestinal tumors
- Clinical setting-dependent diagnostic accuracy of artificial intelligence and store-and-forward diabetic retinopathy screening: a systematic review and meta-analysis
BibTeX
@article{yu2026oneshot,
title = {{One-shot synthesis of rare gastrointestinal lesions improves diagnostic accuracy and clinical training}},
author = {Jia Yu and Yan Zhu and Peiyao Fu and Tianyi Chen and Zhihua Wang and Fei Wu and Quanlin Li and Pinghong Zhou and Shuo Wang and Xian Yang},
journal = {Nature Communications},
year = {2026},
month = oct,
doi = {10.1038/s41467-026-78492-y},
url = {https://www.nature.com/articles/s41467-026-78492-y},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - One-shot synthesis of rare gastrointestinal lesions improves diagnostic accuracy and clinical training
AU - Jia Yu
AU - Yan Zhu
AU - Peiyao Fu
AU - Tianyi Chen
AU - Zhihua Wang
AU - Fei Wu
AU - Quanlin Li
AU - Pinghong Zhou
AU - Shuo Wang
AU - Xian Yang
PY - 2026
DA - 2026/10/25
JO - Nature Communications
DO - 10.1038/s41467-026-78492-y
UR - https://www.nature.com/articles/s41467-026-78492-y
AB - Rare gastrointestinal lesions are infrequently encountered in routine endoscopy, limiting data for developing reliable artificial intelligence (AI) models and training novice clinicians. Here we present EndoRare, a one-shot, retraining-free generative framework that synthesizes diverse lesion exemplars from a single reference image. Using language-guided concept disentanglement, EndoRare separates pathognomonic lesion features from non-diagnostic attributes, encoding the former into a learnable prototype embedding while varying the latter to ensure diversity. We validate EndoRare across four rare pathologies: calcifying fibrous tumor, juvenile polyposis syndrome, familial adenomatous polyposis, and Peutz-Jeghers syndrome. Experts judge the synthetic images clinically plausible, and data augmentation significantly improves downstream AI classifiers, increasing true positive rates at low false-positive rates. In an exploratory reader study, exposure to EndoRare-generated cases is associated with improved image-level recognition by novice endoscopists. These findings support the potential use of EndoRare as a data-efficient tool for rare-lesion education and motivate prospective controlled validation.
SN - 2041-1723
ER -