Meeting Summary for HBHI Workgroup on AI and Healthcare 

On May 16, 2025, the HBHI Workgroup on AI and Healthcare held a two-speaker session featuring Dr. Therese Canares, physician-entrepreneur and founder of CurieDx, and Dr. Jenifer Siegelman, radiologist and founder of Asher Informatics. Moderated by Drs. Risa Wolf and Tinglong Dai, the seminar offered an unusually comprehensive view of AI in healthcare: from point-of-care innovation born in the emergency room to governance infrastructure shaped by lessons from radiation safety.

Dr. Therese Canares: From ER Bottlenecks to AI-Driven Home Testing

 

Fireside Chat: From ER Overcrowding to AI-Powered Home Testing: A Physician-Entrepreneur’s Journey

Dr. Canares shared a deeply personal and strategic journey that began in the pediatric emergency department during the COVID-19 pandemic. Families were waiting hours for tests only to be sent home with minor diagnoses. “I wanted to keep families at home when they didn’t need to be here,” she said, describing the catalyst that led her to co-found CurieDx. The startup’s flagship tool, now available on the App Store, uses smartphone images to provide a tonsil analysis and a machine-learning-driven risk score for strep throat.

Key stages in her path included:

  • Feasibility first. Dr. Canares ran a proof-of-concept study with Hopkins computer scientists using a small innovation grant from the Children’s Center. “You don’t leave your job to build something before you test whether it’s even viable.”
  • Commercialization through necessity. “Our clinical research was never going to make it past the abstract. Commercialization became the only route to patient impact.”
  • Team building. Her insight: “Building a company mirrors building a research lab—define the problem, find collaborators, and scale from project to program.”
  • Regulatory acumen. She clarified that the FDA governs only safety, efficacy, and marketing claims. “It’s not about the algorithm. It’s about what you say your product does.”

She stressed that many physicians make the mistake of publishing novel work before consulting a patent office—rendering their own innovation unprotectable due to self-created “prior art.”

 

Pearls for clinician-innovators:

  • Think like a startup founder: define customer personas, attach dollar costs to problems, and focus on urgent, scalable use cases.
  • Don’t confuse idea generation with innovation. “Everyone has ideas. What matters is solving a problem that someone would duct tape a solution to if they had to.”
  • Engage with JHTV and the Carey Business School early. “You may be the best person to lead your innovation forward.”

 

Dr. Jenifer Siegelman: Governing AI by Learning from Radiation Dose Stewardship

 

Talk: An Approach to AI Governance in Medical Imaging: Lessons from Radiation Dose Monitoring

Dr. Siegelman brought a systems lens to AI oversight, drawing on her experience in radiology, pharmaceutical trials, and quality assurance. Her narrative began with early 2000s CT scan variability, including pediatric safety failures that prompted reform through dose monitoring. She likened today’s situation with AI to those pre-reform years: “We have thousands of FDA-cleared AI tools—but no required post-market monitoring.”

Core insights from her governance work:

  • AI variability is invisible but real. Tools trained on patients in Massachusetts may underperform on populations in New Mexico. “We can’t assume generalizability—we have to test for it.”
  • Hospitals need feedback loops. Asher Informatics works with health systems to integrate AI monitoring into clinical workflows, ingesting scan data, AI output, and clinical reports to detect discrepancies and blind spots.
  • FDA oversight is necessary but not sufficient. Dr. Siegelman emphasized that many radiology tools are FDA-cleared only for narrow technical claims (e.g., “displaying an image”), but used off-label in decision-making.
  • Change management plans are the next frontier. Increasingly, FDA expects vendors to have proactive strategies for algorithm evolution and drift mitigation.

Her company, operating as a public benefit corporation, also emphasizes equity. “We evaluate algorithm performance across diverse populations. If a tool only works for one group, that’s not just bias—it’s unsafe.”

 

Future outlook:

  • AI governance is currently a $1B market, projected to grow to $25B by 2035.
  • Radiology is ripe for transformation, but incentives are misaligned. “Hospitals buy robots that reduce length of stay. What’s the ROI for buying AI to help your radiologist see nodules faster?”
  • Despite Geoffrey Hinton’s claims, AI has not eliminated radiologist training. “We don’t have a surplus of radiologists—we have a shortage. AI augments, but it doesn’t yet replace.”

Discussion Highlights

  • On education: Dr. Canares urged medical schools to teach commercialization, regulatory science, and AI literacy. “Being a pure clinician won’t cut it. You have to lead, analyze, or build.”
  • On equity and bias: Dr. Siegelman stressed that most AI developers lack access to real-world demographic metadata. “You can’t fix bias if you can’t see it.”
  • On resistance to automation: Drawing parallels to longshoremen resisting port automation, Dr. Dai raised provocative questions about radiology’s economic structures. Dr. Siegelman responded candidly, noting that radiology’s tight control of credentialing—while a bulwark for safety—may slow responsible task delegation.

Closing Note

Drs. Canares and Siegelman brought rare transparency and nuance to two essential but often siloed areas of AI in healthcare: bedside innovation and enterprise governance. Their shared message was clear—AI’s future in healthcare requires clinician leadership, rigorous oversight, and deep engagement with both the human and technological elements of care.