AI enters clinical practice: cancer screening, FDA clearances, and a Kenya RCT

Today's roundup arrives on a day when artificial intelligence shifted, visibly, from research curiosity to deployed clinical infrastructure — in oncology, radiology, primary care, and rare disease diagnosis alike.

Samsung Bets $110 M on Multi-Cancer Blood Test as Trial Data Divide Experts

Samsung C&T has committed $110 million to GRAIL, the California-based liquid biopsy company, with the aim of commercialising the Galleri multi-cancer blood test across Asia. The investment follows GRAIL's PATHFINDER 2 trial, which enrolled more than 35,000 participants and found that Galleri detected cancer 6.5 times more often than participants' usual care — picking up cancers of the ovary, pancreas, kidney, and other organs before symptoms emerged. The picture is not uniformly positive: the NHS-Galleri trial in England, the largest real-world evaluation to date, did not meet its primary endpoint of demonstrating a significant reduction in Stage III and IV cancer diagnoses, raising important questions about how a laboratory detection advantage translates to measurable population health gains. Galleri remains in commercial roll-out in the United States and now looks to Asian markets while the clinical debate about its real-world impact continues.

FDA Grants Aidoc Breakthrough Status; UpDoc Wins First Patient-Facing LLM Clearance

Two regulatory milestones arrived this week in AI-assisted care. Aidoc, a Tel Aviv-founded AI radiology platform deployed in more than 2,000 hospitals worldwide, received FDA Breakthrough Device designation for its "First Read" tool, which automatically generates preliminary radiology reports covering more than 100 findings on chest X-rays. The designation is not full clearance but accelerates the path to market by giving Aidoc priority FDA interaction and review guidance. In a separate and historically significant decision, UpDoc received FDA clearance for what regulators describe as the first patient-facing large language model medical device: a conversational AI that can recommend insulin dose adjustments for Type 2 diabetes patients within physician-established parameters. Both decisions together mark a week in which the FDA formally acknowledged AI as a participant — not merely a tool — in clinical decision-making.

AI Consult Improves Clinical Quality in 9,600-Patient Kenya RCT (Nature Medicine)

A randomised controlled trial led by the University of Birmingham, published this week in Nature Medicine, assessed a generative AI advisory tool called "AI Consult" across 16 primary-care clinics in Kenya involving approximately 9,600 patients. Clinicians who used the tool showed significantly improved alignment with Kenyan national guidelines for diagnosis and treatment — a process-quality win in a low-resource environment where specialist support is scarce. Short-term patient outcomes, however, did not reach statistical significance, meaning the trial cannot yet demonstrate that better clinical decision-making translated into measurably healthier patients over the follow-up period. The gap between improving how clinicians decide and improving what patients experience is a recognised challenge in implementation science, and the Kenya trial puts hard numbers on it. Researchers noted that a longer follow-up window may be needed before outcome effects become measurable.

ASCO Breakthrough 2026: AI Shifts From Experiment to Standard in Oncology

At the ASCO Breakthrough 2026 meeting in Singapore, oncologists and AI researchers described a sector-wide transition from pilots to embedded deployment. Presentations highlighted AI tools now integrated into multidisciplinary tumour boards, assisting radiologists in staging, and automating extraction of tumour-type data for clinical trial enrollment — one system achieved 98% accuracy on tumour-type extraction from unstructured clinical notes, a benchmark that opens faster pathways for patients to reach trials they qualify for. Experts also described the growing role of multimodal biomarker integration, where AI synthesises imaging, genomics, and pathology data simultaneously. The ASCO sessions framed this not as a future scenario but as a current operational reality already under way in well-resourced cancer centres worldwide.

AlphaFold and MATRIX: Protein Discovery in Minutes, Drug Repurposing at Scale

At a University of Utah symposium on AI in biomedical research, scientists showcased updated demonstrations of AlphaFold and RosettaFold, the protein structure prediction tools that have effectively ended a decades-long bottleneck in understanding how proteins fold. What previously required years of crystallography or cryo-electron microscopy can now be approximated in minutes, dramatically accelerating the earliest stages of drug discovery. Separately, RENCI's MATRIX platform — being used by the "Every Cure" initiative — is systematically mining gene-disease-treatment networks to identify approved drugs that might work against conditions currently lacking any effective therapy. Drug repurposing at this computational scale would have been implausible a decade ago; MATRIX makes it a tractable, ongoing research programme.

OpenAI o3 Finds 18 New Diagnoses in Unsolved Rare Disease Cases at Boston Children's (NEJM AI)

A study published this week in NEJM AI described a retrospective analysis at Boston Children's Hospital in which OpenAI's o3 reasoning model reviewed medical records for 376 children with previously unsolved rare genetic conditions — cases that had already undergone years of specialist evaluation. The model proposed new diagnoses in 18 cases, a yield of roughly 5%, which the authors judged highly significant given that some patients had waited more than a decade for any explanation of their neurodevelopmental or neuromuscular disorder. The findings are retrospective and will require prospective clinical validation before o3 could enter routine diagnostic workflows, but the NEJM AI publication places the result firmly within peer-reviewed clinical literature, signalling that the rare-disease community is taking this evidence seriously.

The content in this roundup is journalistic in nature and does not constitute medical advice. Readers should consult a qualified clinician before making any change to their treatment or care.