Overview

Grand Challenge is an online hub for building, testing, and publishing machine‑learning solutions in biomedical imaging. The site hosts more than 400 challenges, over 8,000 algorithms, and a community of 138,000 users, all arranged around a clear workflow that starts with data upload and ends with algorithm deployment.

First, you create or join a challenge. You upload DICOM or NIfTI files, optionally strip identifiers, and set visibility and duration. The platform then presents the data through its browser‑based workstations, where users can view slices, apply custom overlays, and annotate slices using bounding boxes, lines, or polygons. Annotators create question sets that clinical experts answer; these answers are stored as Fast Read/Write metadata. Once annotations are finished, the challenge moves into the training phase. Developers download the training set, train a model, and then submit an algorithm container. Submissions are queued on the platform's infrastructure; the system runs the models on the held‑out test set and collects predictions. The outputs are files in NIfTI or CSV format that match the challenge's expected structure, and evaluation metrics, such as Dice similarity, sensitivity and specificity, are computed automatically and shown on a leaderboard.

During the evaluation phase, contributors can view results in the same browser viewer that was used for annotation. The viewer highlights false positives and false negatives, and the interface allows a bulk accept/reject shortcut for review. After the test phase ends, participants can download the final model container for local deployment or further research. Grand Challenge also provides an “invoke API template” that lets teams integrate their algorithms into external pipelines, though detailed parameters are only described in the cycle reports.

The platform is best suited for groups that handle structured imaging data and need a secure, collaborative environment. It works well for research labs, hospitals, and industry teams that value reproducible performance metrics and want to avoid local installation of complex viewers. It is less appropriate for users who need real‑time inference or large‑scale cloud training beyond the supplied infrastructure, and the site does not mention any pricing or resource limits for algorithm execution.

Key features

End‑to‑end challenge workflow: data upload → annotation → training → evaluation → leaderboard
Browser‑based imaging workstations with high‑performance rendering of bounding boxes, lines, and polygons
Custom question sets for expert annotation and immediate performance feedback