Radiologists struggle to identify fake images

In a new study, only 41% of radiologists could identify something was odd when asked to diagnose patients based on synthetic images. Even after being told some images were deepfakes, only 75% could identify them.

Radiology deepfakes aren't a problem yet — but it's time to get ready, said co-author Mickael Tordjman, from Mount Sinai in New York:

"In the future, it's very possible that [hackers] will try to inject fake medical data — which will be even worse than stealing, because then you will not be able to differentiate which part of the medical chart is real and which part is synthetic."

Read Katie Palmer's whole interview here

fundraising

Big money for helping health systems with AI

By now, it's not uncommon to hear about health systems using artificial intelligence for clinical documentation or other targeted use cases. Brittany Trang reports on startup Qualified Health which wants to help them take a bigger picture approach to AI with a single platform for building, operating, and monitoring the entirety of a health system's AI. Not a point solution — the whole shebang.

The idea has piqued the interest of high-profile investors: Qualified Health just raised $125 million Series B round led by New Enterprise Associates with participation from Menlo Ventures' Anthology Fund.

Consumers race to bots for advice about symptoms

Rock Health

Last week, I pulled a few recent surveys about how physicians are using AI in their practices. But what about everyday people? Two publications suggest broad and growing use of LLM-based chatbots for health advice.

A Rock Health survey of 8,000 consumers found that 32% of respondents had ever used AI chatbots for health information — twice the number from a year ago. Of these users, 56% searched for a diagnosis based on symptoms. Notably, this survey was conducted before OpenAI and other consumer AI companies announced health-specific chatbot offerings this year. (See above)

Microsoft

Microsoft published an analysis of 500,000 health-related conversations with its Copilot chatbot and found that "nearly one in five conversations involve personal symptom assessment or condition discussion, and even the dominant general information category (40%) is concentrated on specific treatments and conditions, suggesting that this is a lower bound on personal health intent."

Health tech news roundup

Viz.ai has a new partnership with Alnylam Pharmaceuticals to create a new "care pathway" to "accelerate early identification and standardize diagnostic evaluation of patients with cardiac amyloidosis."
AI revenue cycle management startup Adonis raised $40 million led by Quadrille Capital with participation from General Catalyst.
Certuma, a startup that wants to create and Food and Drug Administration-authorized AI doctor that cant treat common conditions like urinary tract infections and sore throats, raised $10 million in seed funding from 8VC, Forbes reports in an interesting profile.

More around STAT

Check out more exclusive coverage with a STAT+ subscription

Read premium in-depth biotech, pharma, policy, and life science coverage and analysis with all of our STAT+ articles.

What we're reading

Meta and YouTube Found Negligent in Landmark Social Media Addiction Case, New York Times
Heart failure is common. A quick MRI could improve diagnosis and guide treatment, STAT