AI in Health, August 28, 2026: Six Stories Reshaping Medicine

This week's health technology headlines span the full arc of medical care — from detecting hidden cardiovascular risk in routine breast imaging to guiding neurosurgeons away from critical brain tissue, and from predicting which cancer drugs work best together to protecting patients from AI masquerading as a therapist. Here are the six stories that stood out.

Mammograms Now Screen for Heart Disease Too

A large retrospective study presented at the European Society of Cardiology (ESC) Congress 2026 in Munich could change how we think about breast imaging. Researchers from Sheba Tel Hashomer hospital and the AI Research Center (ARC) in Israel trained an AI model on 97,364 mammogram examinations from 29,921 women and found that the same scans used to detect breast cancer carry detectable signals for cardiovascular disease.

The algorithm flagged elevated risk for common conditions including heart disease and stroke — without additional imaging, blood tests, or clinical visits. For women who already attend regular mammography screenings, this opens a pathway to cardiovascular risk assessment as a byproduct of an appointment they are already making.

The practical implications are significant: cardiovascular disease remains the leading cause of death among women globally, and it is consistently underdiagnosed compared with men. If AI can extract that risk signal from imaging that millions of women already receive, it could close a meaningful gap in preventive care.

AI Guides Brain Surgery to Protect a Patient's Sight

Surgeons at the National Hospital for Neurology and Neurosurgery (NHNN) in London used an AI system to guide the removal of a brain tumour in what the team describes as a world first. The patient, a British man, retained his vision — an outcome the surgical team credits directly to real-time AI guidance.

The system, developed at UCL's Hawkes Lab and funded through an NIHR Doctoral Fellowship and Google, analyses a live camera feed during the operation. As the surgeon works, the AI highlights critical anatomical structures in the visual field — in this case, the pathways responsible for sight — so they can be avoided in real time. This represents a shift from AI tools that assist with pre-operative planning to systems that participate in the operating room as the surgery unfolds.

The NIHR-backed trial is ongoing, but the case establishes that real-time AI anatomical guidance is now clinically feasible.

AI Predicts Which Cancer Drugs Work Best Together

Two papers published this week in Nature Genetics by researchers from the Chan Zuckerberg Biohub and Columbia University address one of oncology's hardest problems: identifying which drug combinations will work for a given patient's cancer.

The AI maps what the researchers call "conserved malignant cell states" — patterns of cancer cell behaviour that persist across different patients with the same tumour type. By identifying these conserved states, the system predicts which drug combinations are likely to act synergistically rather than redundantly.

In tests on pediatric Diffuse Midline Glioma (DMG), one of the most aggressive brain cancers in children, AI-identified combinations doubled survival in mouse models compared with standard approaches. The researchers report approximately 90% accuracy in predicting synergistic combinations. Moving from mouse models to clinical trials is a long road, but the precision with which the AI can narrow the experimental space has the potential to meaningfully accelerate that journey.

States Begin Banning AI from Posing as Therapists

The United States is starting to regulate a category of AI risk that has received growing attention: mental health chatbots that market themselves as substitutes for licensed therapy.

Tennessee's SB 1580, effective July 1, 2026, prohibits marketing any AI product as a qualified mental health professional and creates a private right of action — meaning individuals who suffer harm can sue. California's SB 243, effective January 1, 2026, takes a complementary approach focused on chatbot safety disclosures. A Kaiser Family Foundation report documents significant variation in how states are approaching this space, with more legislation expected.

The policy push reflects real-world demand: a 2026 American Psychological Association survey found that 77% of psychologists now report patients using AI tools for mental health support between — or instead of — clinical appointments. The laws do not ban mental health chatbots outright; they target deceptive marketing that blurs the line between a consumer app and a licensed professional.

AI Reviewed 114,000 CT Scans Across Six European Hospitals

A study published August 26 in Clinical Lung Cancer describes one of the largest real-world deployments of AI lung nodule detection to date. Across six European hospitals over 14 months, an AI system reviewed 114,644 chest CT scans from 65,344 patients, achieving 99.5% scan coverage.

The AI flagged abnormalities in 62.7% of patients. Among those flagged, 1.6% were subsequently diagnosed with early-stage lung cancer — cases where early detection is most likely to translate into curative treatment. The scale of the deployment and the consistency of coverage across institutions are themselves significant findings: lung cancer is the leading cause of cancer death worldwide, and most cases are still diagnosed at an advanced stage when treatment options are far more limited.

AI Ends Years-Long Diagnostic Odyssey for Rare Disease Patients

Two separate programs are demonstrating that AI can resolve rare disease cases that have stumped clinicians for years.

At Boston Children's Hospital, researchers applied an OpenAI reasoning model to 376 patients with previously unresolved diagnoses. The system correctly identified the condition in 19 of 20 known rare disease cases used for validation, then produced new diagnoses for 18 patients in the unresolved group — including, in at least one case, a second diagnosis that clinicians had missed entirely. At the Children's Hospital of Eastern Ontario (CHEO), a separate algorithm called ThinkRare produced 21 new diagnoses with a 70% success rate across its test cohort.

Rare diseases individually affect small numbers of patients, but collectively they affect an estimated 300 million people worldwide, the majority of them children. The average diagnostic odyssey — the time from first symptom to confirmed diagnosis — lasts four to five years. Tools that can compress that timeline have the potential to change outcomes meaningfully, particularly for conditions where early intervention matters most.