Powered by collective intelligence, Centaur Labs leverages an on-demand network of medical experts to provide highly accurate yet affordable data labeling.
Case study -- Auggi
Classifying stool images using the Bristol Stool Scale to improve gut health management.Read Case Study
Our clients and partners include large pharmaceutical companies, venture backed startups, and researchers at leading academic institutions. Contact us to hear more customer stories and see how we can solve your labeling needs.
Z Imaging in collaboration with researchers at Brigham and Women’s Hospital and Northeastern University
Identifying and segmenting brain hemorrhages in tens of thousands of CT scans
Centaur designed an image annotation task that required labelers to identify and segment hemorrhages in brain CT scans. This task was released to Centaur's on-demand expert network, which includes verified doctors, medical students, and other medical practitioners who compete to analyze the images most accurately. Powered by collective intelligence, Centaur generated thousands of segmentations on CT scans, combining the answers from only the best labelers to deliver high quality segmentations.
Identifying and segmenting key features in front-of-the-eye images extracted from hundreds of slit lamp videos.
Centaur worked with the client to refine its dataset of anterior segment videos into a usable image format. The Centaur labeling network then annotated pupil and retinal slits in each image. The Centaur team quickly delivered over 25,000 high quality opinions on 6,000 images of eyes, powering the first version of Ocular's initial telehealth eye exam tool.
Classifying thousands of nuanced citation snippets, requiring careful reading from academics with deep understanding of the English language.
Centaur helped the client devise a text classification task that would correctly classify the edge-case data to improve the client's algorithm, and released it to the network of skilled labelers. Centaur's network provided contextual analysis classifications for over 7,000 text citations in a matter of weeks, allowing scite to further refine their algorithm. scite was impressed by the richness of the results received, which would have taken their internal labeling team of PhDs months to perform. scite also used the disagreement scores, compiled from multiple expert opinions, to improve the rules for their internal labeling algorithm.
Identifying malignant and benign lesions and tumors in tens of thousands of skin images.
Centaur designed an image classification task that asked people to distinguish benign and malignant tumors, as well as determine the specific type of lesion (like melanoma, basal cell carcinoma, actinic keratosis) and identify dermoscopic features. Across three labeling tasks—and more than 11,000 skin lesion images—Centaur collected over half a million opinions from its network of experts. The ongoing research is producing new insights for skin lesion feature analysis and demonstrating that the Centaur labeling network achieves classification results superior to individual expert dermatologists.