Mammography AI spots tumours six years early; AMR mapped in landmark study

A Swedish study published in Radiology found that AI can detect signs of breast cancer up to six years before formal diagnosis — while a new King's College London genomic analysis in Cell Genomics projects that antimicrobial resistance could kill 39 million people between now and 2050 if current trends go unchecked.

Saturday, 13 June: AI mammography spots breast cancer signs up to six years early

A retrospective study published in Radiology, the flagship journal of the Radiological Society of North America (RSNA), analysed 88,963 mammograms from 31,394 Swedish patients and found that three commercially available AI-assisted detection systems (AI-CAD) consistently assigned higher risk scores to women who later developed breast cancer. At 90% specificity, the systems flagged potential future cancers in up to 19.7% of women six years before their formal diagnosis, 25.2% four years prior, and 39.3% two years prior. The study is retrospective — it looked back at scans already taken — and cannot yet show whether AI-triggered earlier recall would have improved survival outcomes; a prospective trial would be needed to confirm clinical benefit. Nevertheless, the findings strengthen the case that early mammographic tissue changes are detectable by algorithm years before a radiologist reviewing a single scan would act. The authors call for prospective studies to test whether AI-flagged early-recall pathways can translate this signal into earlier, more treatable diagnoses.

Friday, 12 June: King's College maps 210 AMR threats in 45,000 genomes across 127 countries

A major genomic study led by King's College London, published in Cell Genomics, analysed more than 45,000 bacterial genomes from 16 species designated as WHO critical-priority pathogens across 127 countries. The researchers identified approximately 210 resistance traits most likely to expand globally over the next two decades, projecting that antimicrobial resistance (AMR) could cause 39 million deaths worldwide between 2025 and 2050 if current trends continue. The study's most striking finding is that socioeconomic inequality, overcrowding, and access to clean water and sanitation — not just antibiotic over-use — are among the strongest predictors of future resistance spread. For the Mediterranean and eastern Mediterranean regions the direct risk tier is moderate, but researchers emphasise that AMR travels freely through trade networks, tourism corridors, and the food supply chain, leaving no region insulated. The paper urges a One Health policy response coupling antibiotic stewardship with investment in water, sanitation, and hygiene infrastructure.

Friday, 12 June: UK MHRA launches AI sandbox to reduce adverse drug reactions

The UK Medicines and Healthcare products Regulatory Agency (MHRA) announced the opening of a regulatory AI sandbox — a controlled environment where companies and academic researchers can evaluate AI tools designed to predict how medicines behave in the body and to identify potential adverse effects before they reach patients. Science Minister Lord Vallance made the announcement during London Tech Week on 9 June. The first phase of the programme, opening in summer 2026, will test up to five AI-driven approaches. Adverse drug reactions currently account for approximately 250,000 hospital admissions in England each year and cost the NHS over £2 billion annually. The sandbox is intended to de-risk AI evaluation before tools enter clinical use and to generate early regulatory learning that can feed into formal MHRA guidance. A concurrent MHRA public report found broad patient support for AI in medicines development, provided transparency and safety oversight are maintained.

Friday, 12 June: Time-alignment flaw found in published AI sepsis treatment models

A study led by Shengpu Tang of Emory University, published in npj Digital Medicine, has identified a systematic time-misalignment error embedded in many published reinforcement-learning models designed to guide fluid and vasopressor treatment decisions in sepsis patients. The flaw causes the AI to receive delayed performance feedback, creating the illusion that a patient's condition improved before treatment was actually administered — teaching the algorithm incorrect clinical associations. When the misalignment is corrected, simulations show 8–10% reductions in patient mortality compared with uncorrected models. The team estimates the error may affect nearly half of patient states in affected algorithms and calls for systematic auditing of published clinical AI models before deployment in live care environments. The finding is an important reminder that peer review of a clinical AI study does not constitute a safety clearance for bedside deployment.

Thursday, 11 June: One in five US teens uses AI chatbots for mental health — JAMA Pediatrics

A nationally representative study published in JAMA Pediatrics, led by the RAND Corporation and funded by the US National Institute of Mental Health, found that approximately 19.2% of American adolescents and young adults aged 12–21 — roughly 8.2 million people — have used an AI chatbot for mental health advice. Among those who consulted chatbots, nearly 43% did so at least monthly, and 92% found the guidance at least somewhat helpful. The study raises a significant safeguarding concern: approximately 63% of young users had not disclosed their chatbot use to any adult, clinician, or peer, placing this form of help-seeking entirely outside the formal care system. Young people with the most severe symptoms — including moderate-to-severe depression and suicidality — were more likely to turn to AI tools, raising the risk that unregulated applications could become a substitute for, rather than a supplement to, professional care. The American Psychological Association's Dr. C. Vaile Wright noted there is no professional consensus that AI chatbots can replace therapy, and researchers urge platforms to incorporate crisis detection and referral mechanisms as a baseline safety requirement.

The content above is journalistic reporting and does not constitute medical advice. Readers should consult a qualified healthcare professional before making any change to their treatment or health management.