Blog

“The nice thing about AI is that it never gets tired. So as we think about AI replacing radiologists, let's focus on AI replacing the tedious, humdrum, boring stuff that needs to be done very well. That's where AI has the greatest opportunity."
- Dr. Brad Erickson, MD, PhD, Mayo Clinic at fireside chat at RSNA

Paige, the global leader in end-to-end digital pathology solutions and clinical AI applications, was able to improve their model's F1 score from .6 to .83 and complete annotation 10 x faster by working with Centaur Labs to annotate their pathology slide dataset.

Our new Research product enables researchers to use our platform and labeling network for free, and our new APIs allow model developers to integrate our data annotation capabilities into their ML pipelines.

At RSNA we announced our Radiology AI Safety initiative in partnership with deepc and Segmed. Together we’ll make it easier to both identify opportunities to improve models and get the data to quickly retrain them.
From new research and regulatory approvals, to new datasets, here are some of our favorite updates in AI in healthcare.
Whether you're -
...Centaur Labs can help accelerate and improve your data annotation process.
Share the resources you're using, the research you're reading and publishing, and the roles you're hiring for and we’ll share with the community next month.
Until next month,
Erik and the Centaur Labs team
Gamified data labeling enhances model accuracy from 70% to 93% in a case study with Eight Sleep, demonstrating the effectiveness of multimodal annotation.
This blog post highlights how high-fidelity annotation determines the reliability of models in complex scientific and medical domains. It introduces a MedTech case study demonstrating Centaur.ai’s volumetric workflow, expert-driven review, and rigorous quality controls that enabled sub-millimeter cardiac segmentation for advanced simulation and AI training. Readers are encouraged to download the full case study.
Partnered with SciBite to accelerate vocabulary curation, cutting the timeline by over two months through expert crowd-labeling, achieving 90.3–95.1% accuracy.