Pediatric Moonshot
On Building the AI Superhighway for Healthcare
Pediatric Moonshot
We recently shared our updates with the CEO of one of the leading US children’s hospitals and thought it would be interesting to share them with all of you.
We have proven the ability to deploy a zone in eight hospitals on three continents. Why is it important? As you all know, reliable AI applications can only be developed using real-world data. And, as many of you already know, we are currently building a Distributed AI Lab for Healthcare and Life Sciences that will operate within each of these zones. Our goal is to expand the network to 32 zones, each hosting its local imaging data, reports from the PACS, and medical records from the EMR. These records will become selectively available, in real-time, to authorized AI applications and agents.
Once completed, this infrastructure will give access to 3,000 servers generating a real-time data flow of over 2,000 Terabytes per year. This is creating an unprecedented opportunity to build distributed AI applications in healthcare and life sciences.
For reference, the most recent model from ChatGPT was estimated to have been trained on 1,000 Terabytes of text data (with no imaging), with a speculated hardware/computing cost of $41 to $78 million. We estimated that completing our Distributed AI Lab will require approximately $50 million, and we are pursuing four funding tracks to build it in the next 12 months.
While constructing this data superhighway is foundational, what are these distributed AI applications like?
We have already identified several applications that can answer the question. On the imaging front, we continued to add new research work in cardiology (e.g., Left Ventricular Dysfunction Prediction) and gastroenterology (e.g., Hepatic Steatosis Detection). We have over 140 additional applications, all sharing a common opportunity to transition from the research bench (with minimal data samples) to the bedside, utilizing AI Lab’s distributed/federated learning capabilities and its unique privacy-preserving deployment to the point of care.
We have now realized that a distributed AI cloud infrastructure can speed clinical trial recruiting. The two essential pieces of information — patient's medical record and drug protocol document — remain protected behind their network firewall as of today, but become accessible to the zone’s AI infrastructure and its intelligence.
We prepared a short demo to show it in action:
Finally, as some of you may know, the biopharmaceutical industry is developing increasingly specialized therapeutics. In 2024, over half of the therapeutics approved were for orphan drugs.
An orphan drug is a medication intended to treat a rare disease or condition, typically affecting fewer than 200,000 people in the United States. The challenge is that clinicians struggle to follow all newly approved drugs, as JNJ’s recently published Care Index (https://www.jnj.com/oncology/oncology-care-index) states that nearly three out of four oncologists find the pace of new treatment development overwhelming.
We recognized potential in developing disease-specific AI agents that leverage multimodal biomarkers to continuously assess a patient’s clinical record for a particular disease, enabling the identification of newly approved therapeutics. We have already begun working with rare kidney diseases (FSGS, IgAN, and aHUS) and have been collaborating with leading nephrologists at the University of Michigan, Boston Children’s, Seattle Children’s, and Nemours. Even at this early stage, the opinion is that AI agents are better than a third-year medical student out of the box. These AI agents will continuously run in the background on a privacy-preserving distributed AI database, incrementally improving accuracy and confidence.
We are now expanding into additional conditions, and if you are interested in learning more, please don't hesitate to reach out.
