Blog

Centaur Labs partners with Mayo Clinic spin out Lucem Health to accelerate medical AI development

Alex McFerran, Director of Partnerships
August 1, 2022

Today we’re excited to announce that Centaur Labs is joining the Lucem Health Innovation Collaborative, a partner program designed by Mayo Clinic spin out Lucem Health to move clinical AI/ML innovation to the front lines of healthcare. 

Accelerating the impact of clinical AI 

Whether to advance the development of life saving medicines, bring AI-powered insights into clinical workflows, or improve the experience and affordability of care - the opportunity for AI in healthcare is tremendous. However, the path to realizing this impact is riddled with challenges - both in the development of AI/ML models, and in their adoption by clinical teams. 

In development, clinical AI/ML leaders must first gain access to clinical data in a secure and compliant manner, clean and annotate that data to make it usable, and then build a model with predictive power. Needless to say, many AI/ML projects don’t make it out of this experimental phase. 

The visionary AI/ML leaders that do successfully build accurate clinical AI then face a new set of challenges. They need to both embed their model’s insights into existing clinical tools, processes and workflows, and identify visionary clinicians to inspire trust and adoption. 

Lucem Health, launched with Mayo Clinic and investing partners Commure (a General Catalyst company) and Rally Ventures, was founded to help healthcare visionaries navigate these challenges, and bring AI powered insights from the bench to the front lines of healthcare. 

The medical data annotation bottleneck is hindering clinical AI development

A critical step in clinical AI development is creating training, test and validation datasets. AI learns like humans—by example—and these datasets provide the examples needed to both train models, and then test and validate the accuracy of their predictions. Training an algorithm requires thousands, if not millions of examples, and as a result, teams need large training datasets, with - oftentimes - thousands upon thousands of annotations. Getting accurate annotations of medical datasets at scale has long challenged AI leaders, creating a bottleneck in clinical AI development.

At first, AI/ML teams may attempt to label training data in-house, having team members with clinical training spend hours-upon-hours manually labeling data. Managing this labeling process in-house is time intensive, tedious and the quality of the annotations degrades over time as skilled labelers get tired and disinterested.

Alternatively, teams may try to outsource data labeling to unskilled teams with no clinical training in low wage countries. While this approach may allow teams to give time back to their clinical leaders in the short term, they often discover low quality labels in their quality control process, and decide to do the labeling projects again with a skilled team.  

At Centaur Labs, we’re focused on removing this medical data annotation bottleneck, and offering clinical AI development teams a third way. 

Accurate medical data annotations at scale 

Centaur Labs provides accurate and scalable data annotation for companies developing AI in the medical and life sciences industries. Our innovative approach to medical data annotation leverages our co-founder and CEO Erik Duhaime’s PhD research in collective intelligence for skilled tasks at MIT’s Center for Collective Intelligence.

Instead of assigning an annotation based only on one clinician’s opinion, Centaur collects many opinions from our network of trusted medical experts, comprised of thousands of medical students and professionals globally. We then provide annotations based on the aggregated opinions of the top performing medical experts at that specific annotation task. We continuously assess performance, so we know who in our network is highest performing on any given annotation task.

Today, we collect over 2 million annotations weekly and annotate a range of medical data formats - from medical/scientific text and medical images to videos and medical audio data. We’re working with ML engineers at top-10 pharmaceutical companies, AI startups, medical device companies like Medtronic, and research institutions like Stanford, Memorial Sloan Kettering, and Brigham and Women's Hospital to build accurately annotated datasets at scale, so they can build high quality clinical AI, and get their models into production more rapidly.

“We are excited to welcome Centaur Labs into the Innovation Collaborative to enable clinical AI/ML developers to access accurate medical data annotations as they train and scale their models,” says Sean Cassidy, CEO of Lucem Health. “We know members of the Collaborative are unsatisfied with the current cumbersome and low-quality medical data annotation process, and we see Centaur Labs as enabling members to get the accurate labels they need, so they can focus on building high quality clinical AI”. 

“We’re thrilled to join the Innovation Collaborative and be part of the platform Lucem Health is building,” says Erik Duhaime, co-founder and CEO of Centaur Labs. “We’re already working with members of the Collaborative to annotate their datasets and are excited to support other members as they build and scale their clinical AI efforts.” 

Related posts

December 3, 2020

Our collection of open source datasets for medical AI

Access a collection of dozens of open source image datasets for medical AI across a variety of formats including X-ray, CT, Ultrasound, Whole Slide Imagery, MRI and more

Continue reading →
August 31, 2022

9 most common types of medical text datasets

From SMS to insurance claims to pathology reports and scientific studies, in this post we dig into the most common type of medical text datasets leveraged for NLP in healthcare.

Continue reading →
January 18, 2022

Centaur Labs recognized in 2021 CB Insights Digital Health 150

We’re humbled and honored to be recognized by CB Insights as one of the top 150 digital health startups in the world!

Continue reading →