Skip to content

Security: tensorbee/ocelli

Security

SECURITY.md

Security policy

Reporting a vulnerability

Report privately through GitHub's Report a vulnerability button on the Security tab of this repository. Please do not open a public issue for a vulnerability.

Never include patient data in a report. Not in a description, not in a stack trace, not as an attachment. If a defect can only be demonstrated with a real study, say so and describe the shape of the data rather than sending it. We will work out a way to reproduce it that does not move a patient record.

What counts as a security issue here

This is a medical imaging library, so the category is wider than usual. Alongside the ordinary classes, we treat these as security issues:

  • A wrong pixel value. Anything causing the rendered result to differ from the DICOM standard, particularly in the LUT chain, rescale, or the VOI functions. A plausible image with incorrect values is the defect class this project is built to prevent.
  • Wrong geometry. Incorrect spacing, slice ordering, or coordinate transforms, which make a measurement wrong while looking correct.
  • A feature that degrades silently. A capability that cannot run on the resolved hardware tier must report itself unavailable. One that quietly produces a different answer instead is a safety issue, not a papercut.
  • Data escaping the browser. Pixel data is designed never to leave the client. A path that sends it anywhere is a defect regardless of intent.

Supported versions

Pre-1.0 and under active development. Only the latest release is supported. Version 1.0 arrives with feature parity and the semantic-versioning and deprecation policy that goes with it.

This library is not a medical device

Ocelli is a component. It carries no regulatory clearance and makes no clinical claim. An integrator building a cleared product is responsible for that product's validation. What this project provides toward it is measured evidence: a differential corpus, a written tolerance policy, and a published divergence bound rather than a claim of exactness.

There aren't any published security advisories