2D and 3D radiology imaging—ultrasound, MRIs, CTs, DICOM, NIFTI and x-rays—encompasses a massive amount of data that can be harnessed by AI for medical decision support and analysis.
We offer classification services to identify one or multiple findings from a set of options and precise segmentation services to draw around regions of interest with boxes, lines or polygons. Examples include annotating vessel occlusions in non-contrast CTs and determining fetal sex from ultrasound videos.
Research teams at Brigham and Women's Hospital engaged Centaur Labs in their goal to automate prediction of COVID-19 disease status from point of care ultrasound, funded by a grant from the Massachusetts Life Sciences Center. With mobile-friendly, gamified labeling tools, the Brigham’s experts labeled thousands of ultrasound stills. Centaur is working closely with the Brigham’s machine learning engineering team to refine the model, which is still in development. Learn more here.
Zeta Surgical, a startup aimed at improving neurosurgery with robotic assistance, worked with Centaur Labs to create a training set for their hemorrhage detection algorithm. Centaur Labs segmented areas of bleed on tens of thousands of brain CT scans from hemorrhage patients. Using expert opinions from multiple labelers, Centaur Labs delivered accurate segmentations to refine Zeta Surgical’s algorithm.
Research teams at Vanderbilt University are working with Centaur Labs to create datasets for future AI model development. Centaur trained labelers to segment COVID-related lung lesions in CT scans who annotated lung lesions on thousands of scans. The results from this competition will be used to create a platform to evaluate emerging methods for the segmentation and quantification of lung lesions caused by SARS-CoV-2 infection from CT images.
Predictive Orthopedics, with data from thousands of patients from a large orthopedic practice specializing in partial knee replacement surgery, contracted Centaur Labs to enrich their dataset and train a model. The partnership developed a predictive algorithm that recommended patients for no surgery, partial knee, or total knee replacement with 0.97 AUROC compared to trained physician assessments.