8th World Congress of Pediatric Cardiology
The 8th World Congress of Pediatric Cardiology and Cardiac Surgery is being held in Washington, DC this week. It’s the largest meeting ever with over 5500 registrations from 117 countries. The Pediatric Moonshot was accepted for the poster session. The image above was printed on a 3 foot x 4 foot sheet of paper. It tells a simple, but we think compelling story.
Pediatric clinical expertise is scarce and geographically concentrated in 500 children’s hospitals in the world.
AI in medicine could be the solution to reducing healthcare inequity, lower cost and improving outcomes. But it’s not performing even in adult medicine. “..studies have shown that the performance of many radiologic AI models worsens when they are applied to patients who differ from those
used for model development…” from The Current and Future State of AI Interpretation of Medical Images Pranav Rajpurkar, Ph.D., and Matthew P. Lungren, M.D., M.P.
So, why is AI in medicine not performing? ChatGPT and many consumer facing AI applications have been trained on large diverse data sets which result in increasing accuracy. But this is not the case for AI in medical imaging.
But the data is available. The 500 children’s hospitals generate over 6,000,000 Terabytes of labeled data a year. Inspired by Dr. Anthony Chang we launched the Pediatric Moonshot in 2020.
PEDIATRIC MOONSHOT MISSION
Reduce healthcare inequity, lower cost and improve outcomes for children nationally and globally, by creating privacy-preserving, real-time applications based on access to data from 1,000,000 healthcare machines in all 500 children’s hospitals in the world.
Most consumer AI applications have been trained using a centralized architecture. But the traditional centralized architectures will not work for AI in medicine. Let’s take the 6,000,000 TB of echo data and put them in a centralized architecture in say Ireland. Who will pay for the data transfer cost from Vatican City, Chicago or San Francisco to Ireland? You’ll never build a real-time application for the globe running in Ireland and aggregating lots of data with no particular purpose is counter the first principles of privacy management — purpose limiation.
We need a new decentralized, in the building edge cloud service to access large amounts of diverse data. BevelCloud has developed and deployed edge zones in eight children’s hospitals on three continents as part of the Vanguard program.
The architecture is optimized for decentralized learning, so you can train the AI application without the data ever leaving the children’s hospitals. We are in the process of launching the first phase of the Gemini program, which will focus on building an AI research lab for pediatric cardiology. Initially this will be six sites on two continents, 100 twinned ultrasounds with 100 distributed edge servers. Phase 1 will have access to more data than the NIH’s Imaging Data Commons project. Phase 2 will expand to over thirty two children’s hospitals on four contienets and include all imaging machines to support AI application training in neuro-radiology, orthopedics and emergency medicine.
Technology alone is not enough. We have the support of many clinicans, researchers and administrators. With our combined efforts we will be able to reduce healthcare inequity, lower costs and improve the quality of care for children everywhere, even those who are not geographically or socially lucky.
