Centaur Labs

Centaur Labs labels medical data for healthcare AI teams using a large crowd of vetted experts, and it decides which of their opinions to trust with a scoring system rather than by seniority.

Annotators work through DiagnosUs, a mobile app that turns labeling into something closer to a game. People label cases, get scored against known answers, and earn more for accurate work. Centaur tracks how each person performs over time and, for any given case, blends the reads of only its top performers. A label is not one contractor's guess. It is a quality-weighted answer from the annotators who have proven they are good at that specific task.

The work covers the data types clinical models are built on:

• Radiology, pathology, and dermatology images
• 3D segmentation of DICOM studies
• Classification and region-of-interest measurements
• Named entity recognition in medical and scientific text

For imaging, the platform runs on the OHIF viewer, so annotators get real radiology tooling instead of a generic box-drawing interface. Finished data comes back through an API or as JSON, CSV, or DICOM, ready to train and evaluate models.

Teams reach for it when accuracy on hard, specialized cases matters more than raw speed. A device company validating a diagnostic model needs labels that survive regulatory scrutiny. A digital health startup needs a reliable read on the edge cases its own clinicians argue about. Because the data is clinical, the same performance tracking that scores annotators also surfaces cases where the experts disagree, so a reviewer can settle them before the labels ship.

Common users:

• Medical device and diagnostics companies training and validating models
• Digital health and healthtech teams building clinical AI
• Pharma and life-sciences groups labeling imaging and text
• Teams that need expert medical labels without hiring a clinical annotation staff

Pricing is a managed service, quoted per project by data type, case volume, and how much expert review the work needs.

Best for healthcare AI teams that need high-quality labels on specialized medical data and would rather buy expert judgment than stand up their own annotation team. Not the right fit for a team that wants a self-serve platform to run with its own annotators, or for non-medical data, where a general vision or text vendor is a closer match.