The current explosion of AI has precipitated new waves of researchers from various disciplines eager to try their hand at AI research. Data scientists brought into research initiatives face challenges mired with a shared lamentation: the collection, curation, and maintenance of high fidelity and high value datasets is a challenging, expensive, and sometimes even show-stopping initiative. These tasks often consume a majority of data scientists' efforts, minimizing the time they could be spending on the very model development they have spent so many years training to accomplish. Our Medical Imaging AI-focused Lab in Emergency Radiology at Brigham and Women's Hospital faced these challenges at our inception. Despite both excitement and troves of raw data, we lacked the infrastructure and staffing to complete effective AI research. Here we present a methodology and platform built and refined over a three-year effort. We first created a collection of simple and local tools which we developed further into a distributed microservice architecture that is now able to support our AI research efforts at scale. Most significantly, it is designed to run on common-denominator hardware, and to promote collaboration and synergy between multidisciplinary researchers. We have found that this approach not only provides dramatic reduction to our resource needs, both in infrastructure and staffing, but also allows for the collection of higher-fidelity data at lower-cost, which may allow simpler AI models to prove successful. This talk is meant as a guide for the nimble and uninitiated: how to build and accelerate your medical AI lab to solve big problems with big data from the ground up.
Want to sponsor this event? contact us.