# EndoRare generates rare endoscopy examples from one reference image

Agent contract: 1.2.0

A study reports promising image-test results. Whether the synthetic examples help doctors recognize unfamiliar cases remains unproven.

Canonical: https://brightaifuture.com/discoveries/endorare-one-shot-endoscopy
Format: discovery
Source publication: 2026-10-25
Bright publication: 2026-10-26
Substantive update: None recorded
Evidence and review: Experimental; confidence: unassessed; approved; ai-assisted. Independently reviewed article marked READY; publication explicitly approved by owner October 25, 2026. Static-image and exploratory reader-study limitations retained. No medical images included.

## EndoRare generates rare endoscopy examples from one reference image

How do doctors learn to recognize something they almost never see? For rare findings inside the digestive tract, the shortage of examples creates a practical problem: clinicians need exposure to learn, but the next relevant case may be hard to come by. AI developers face a similar shortage of training images.

EndoRare explores whether generated images can help fill that gap. In an accepted paper published in Nature Communications on 25 October, researchers describe a system that produces new endoscopy examples from a single target reference image. “One-shot” describes that reference, not the system’s entire training history: it builds on a model already trained on routine medical images. [Accepted paper](https://www.nature.com/articles/s41467-026-78492-y_reference.pdf?error=cookies_not_supported)

The aim is to preserve a lesion’s identifying features while varying other aspects of its appearance. The study covers four conditions: calcifying fibrous tumor, juvenile polyposis syndrome, familial adenomatous polyposis and Peutz-Jeghers syndrome. [Accepted study](https://www.nature.com/articles/s41467-026-78492-y)

## What the tests found

One test asked whether synthetic examples helped an image classifier. Averaged equally across the four categories, its precision–recall area under the curve, or PR-AUC, was 0.708 with EndoRare, compared with 0.612 using Textual Inversion, a competing image-generation method. [Supplementary Tables 9, 11, 13 and 15](https://media.springernature.com/original/springer-static/esm/art%3A10.1038%2Fs41467-026-78492-y/MediaObjects/41467_2026_78492_MOESM1_ESM.pdf)

PR-AUC summarizes the trade-off between finding positive cases and making correct positive predictions across different decision thresholds. The result does not mean the classifier was “70.8% accurate,” and it does not establish performance during a live examination.

A separate exploratory study involved 10 novice endoscopists from four institutions, assessed on the same four conditions before and after exposure to generated examples. After pooling each reader’s decisions across those conditions, the authors reported an average F1 score rising from 0.581 to 0.901. F1 combines precision and recall. [Accepted paper, Results and Table 1](https://www.nature.com/articles/s41467-026-78492-y_reference.pdf?error=cookies_not_supported)

That is an encouraging observation, but the readers took the same image test again after two weeks. They received no answers or performance feedback after the first test, and there was no comparison group. This design cannot separate the contribution of synthetic examples from familiarity with the test or practice. It leaves unanswered the question that matters most for education: does any improvement carry over to unfamiliar cases and last?

## How to read the evidence

Methods in brief. Clinicians rated 700 generated outputs across seven methods. They did not know which method produced each image, but did know the images were synthetic. The classifier evaluation used five-fold cross-validation with the same splits across methods. Those folds were technical repeats within the evaluation, not five independent patient cohorts. [Accepted methods](https://www.nature.com/articles/s41467-026-78492-y_reference.pdf?error=cookies_not_supported) · [Supplementary Methods 3.7](https://media.springernature.com/original/springer-static/esm/art%3A10.1038%2Fs41467-026-78492-y/MediaObjects/41467_2026_78492_MOESM1_ESM.pdf)

The next persuasive evidence would include a controlled reader study using new test images, follow-up to measure retention, and classifier evaluation on independent patients. Clear counts of patients, examinations and images, together with documented separation between training and test data, would make the results easier to assess. These are open verification questions, not evidence that overlap occurred.

The researchers have posted [public research code](https://github.com/Jia7878/EndoRare), but its README states that no license has been declared and rights remain reserved. It should not be described as open source. The clinical images are not publicly available, and privacy protection has not been established. [Data access and limitations](https://www.nature.com/articles/s41467-026-78492-y_reference.pdf?error=cookies_not_supported)

For now, EndoRare offers a research direction for a familiar teaching problem: how to provide useful practice when examples are scarce. The evidence concerns static images, not fewer missed diagnoses or better patient outcomes.

## Provenance and history

{
  "dates": {
    "captureDate": "2026-10-25",
    "eventDate": null,
    "lastReviewedDate": "2026-10-25",
    "publicationDate": "2026-10-25"
  },
  "provenance": {
    "origin": "editorial",
    "externalId": "https://www.nature.com/articles/s41467-026-78492-y"
  },
  "revisions": [
    {
      "id": "endorare-one-shot-endoscopy-initial",
      "recordedAt": "2026-10-25",
      "sourceIds": [
        "source:endorare:accepted-paper",
        "source:endorare:study",
        "source:endorare:supplement",
        "source:endorare:repository"
      ],
      "summary": "Initial independently reviewed report on EndoRare, with corrected ten-reader study and scientific limitations."
    }
  ],
  "corrections": []
}

## Original sources

- [EndoRare accepted paper · Nature Communications](https://www.nature.com/articles/s41467-026-78492-y_reference.pdf?error=cookies_not_supported)
- [EndoRare study · Nature Communications](https://www.nature.com/articles/s41467-026-78492-y)
- [Supplementary Methods 3.7 and Tables 9, 11, 13 and 15](https://media.springernature.com/original/springer-static/esm/art%3A10.1038%2Fs41467-026-78492-y/MediaObjects/41467_2026_78492_MOESM1_ESM.pdf)
- [EndoRare public research code and licensing statement](https://github.com/Jia7878/EndoRare)

## Continue exploring