Healthcare AI News 6/21/23

News

Google adds the ability to search for skin conditions to its Lens visual search tool. Users take or upload a photo of their skin condition and the system finds visual matches, which it says is easier than trying to describe the situation with words.

Startup Dandelion Health, which offers de-identified health system data for clinical research, launches a free public service that will evaluate the performance, fairness, and equity of health algorithms. It will initially focus on predictive algorithms for cardiology. The company’s pilot program – which uses data from Sharp HealthCare, Sanford Health, and Texas Health Resources – will evaluate if an algorithm trained on one area’s dataset performs equally well when applied to other populations and if it does so fairly for everyone.

Business

Korea-based AI medical solutions vendor SPASS receives FDA 510(k) clearance for its AI-based detection software for sepsis, anaphylaxis, and hypovolemic shock.

DeepX earns FDA clearance for its digital dermatoscope, which will acquire images for teledermatology review. The company is seeking FDA approval for integrating the images with an AI algorithm that will analyze lesions based on their light transferring properties to provide fast-track diagnosis of skin cancer.

A McKinsey report predicts that generative AI will add $4.4 trillion to the global economy, with leading use cases being in banking, technology, and life sciences.

ChatGPT creator OpenAI is reportedly planning to launch a marketplace for AI models that use its technology, which could raise competitive issues with its partners.  

Research

Researchers find that AI-powered analysis of EHR data can predict pancreatic cancer earlier, identifying heightened risk up to three years before diagnosis. Unexpected predictive symptoms include gallstones, Type 2 diabetes, anemia, and GI symptoms. The researchers believe that 320 of each 1,000 people the AI model identifies as high risk will develop cancer. The federal government doesn’t recommend screening symptom-free people for pancreatic cancer, but targeting AI-identified high-risk patients would make surveillance more affordable and improve long-term survival.

AI algorithms for predicting inflammatory bowel disease have been enhanced to offer personalized treatment recommendations and – by incorporating patient-reported outcomes, sensor data, and biomarkers – to detect early signs of worsening or to predict which treatments will be effective.

Other

Meredith Broussard, MFA, whose 2018 book “Artificial Unintelligence” coined the term “technochauvinism” in describing the belief that technology can solve any problem better than humans, describes her experience with running open-source AI models on her mammograms to see if it would detect the cancer that her doctor has already diagnosed:

The AI that I used did, in fact, work. But it doesn’t diagnose the way that a doctor does. It drew a circle around an “area of concern” on a single flat image and gave me a score between zero and one … I realized that I had expected more—not the Terminator, and not a Jetsons-style robot doctor, but at least a humanlike diagnosis based on my entire medical record. This is pretty typical. We often have imaginary expectations about AI, and the technology fails to live up to what we imagine it can do. It would be really great if we could diagnose more people earlier. It would be great if we could use technology to save more lives from cancer. We are absolutely all united in that goal. But the idea that AI is going to be our salvation for diagnosing all cancers in the next few years is a little bit overblown.

Contacts

Mr. H, Lorre, Jenn, Dr. Jayne.
Get HIStalk updates.
Send news or rumors.
Contact us.