July 2026 Cycle Report

Published 31 July 2026

🌌 Mass Annotations in the Pathology Viewer¶

In a previous development cycle, we introduced a proof of concept for rendering over a million point annotations at once in our client-side pathology viewer. We've now extended this approach to support additional annotation types: bounding boxes, lines, and polygons. These new annotation types offer the same performance and usability as points, including:

• Real-time rendering of hundreds of thousands of annotations
• Efficient viewport handling, so only visible annotations are processed
• Hover highlighting and label display
• Two-way selection between the viewer and sidebar

To keep the viewer responsive, more demanding calculations (such as smoothing curved annotations) run in the background rather than on the main thread, so users won't experience any lag while working. At present, polygons are displayed as outlines only. Fully filled polygons at this scale present a more complex rendering challenge, which we plan to address in a future cycle if there's a clear need for it.

✨ Updated algorithm and evaluation method templates¶

We have updated the algorithm and evaluation method templates to use the new invoke API.

The new template makes algorithms compatible with future platform improvements, such as batch-mode for challenge submissions, and enables them to be used as interactive algorithms in reader studies.

Existing algorithms will continue to work, as the platform is fully backwards compatible with the exec API. New algorithms, however, should use the new templates and implement the invoke API to take advantage of the above mentioned benefits.

You can download a fresh template from your algorithm page or your challenge phase settings and try it out. The documentation walks you through the updated workflow.

If you have any questions, please don't hesitate to get in touch: support@grand-challenge.org.

🤖 Creating Algorithm Endpoints¶

For external applications, we have extended the REST API to support creating algorithm endpoints. Algorithm endpoints can now be created and—when the endpoint is up and running—they can be invoked to process inputs directly, without the need to reload the model. The endpoint remains active as long as it continues to receive invocations. After a period of inactivity, it is automatically cleaned up following a short grace period. To prevent cleanup, it can be kept alive by periodically sending it a keep_alive request. We have also updated our GCAPI Python package to simplify creating, managing, and invoking algorithm endpoints.

🎓 Answer scoring in educational reader studies¶

Educational reader studies allow readers to verify their answers against ground truth. Previously, this verification was handled twice: once in CIRRUS (via the Verify button) and once in the Grand Challenge backend (for the leaderboard and statistics pages). This required the ground truth for each question to be shared with CIRRUS so it could calculate the score independently.

To keep the ground truth private, we removed score calculation from CIRRUS. Grand Challenge now calculates and returns the score whenever the reader clicks Save or Verify. This also lays the groundwork for more advanced educational modes, such as an Exam mode, in the future.

✅ Bulk accept/reject shortcut in a reader study case.¶

Reader studies can be used to assess algorithm findings, such as head circumferences in ultrasound images or nodules in lung CT scans. When a study contains many findings, accepting or rejecting each one individually can become cumbersome. To address this, we've added a shortcut that allows readers to accept or reject all pending (unassessed) findings at once. To set this keybinding for your current session, open the keybindings menu by pressing K or clicking the icon in the top-right corner. It is also possible to set this keybinding in the Reader study configuration, see our documentation.

Cover photo by Harli Marten on Unsplash