Weekly AI digest
Radiology & medical imaging AI · week of 6 to 13 June 2026
14 peer-reviewed papers · 3 industry and regulatory items · conference and KOL highlights. Adapted from my weekly intelligence report. Full report (PDF, FR).
Three things that mattered
AI mammography flags breast cancer years before diagnosis
A retrospective study in Radiology (Karolinska/Lund, 31,394 women) shows three commercial CAD systems (Vara, Lunit, MammoScreen) flag roughly 20% of future cancers six years before clinical diagnosis, rising to about 39% at two years (AUC 0.63 to 0.67, beating density alone). A path toward AI-score risk stratification.A radiology foundation model passes a board exam
Harrison.Rad 1.5 became the first foundation model to pass the Royal College of Radiologists FRCR 2B Short Case (median 86.5 vs a 73.2 threshold), trained on roughly 6 million exams. Alongside it, Vara earned CE class IIb under EU MDR for 3D tomosynthesis and Subtle Medical cleared SubtleHD CT via FDA 510(k).Governance becomes a sector standard
At SIIM 2026, 49 of 72 sessions (68%) addressed AI, and the joint ACR-SIIM Practice Parameter for AI in imaging was framed as the first sector-wide governance standard, with compliance tied to the ACR Recognized Center for Healthcare-AI designation.
Peer-reviewed
Selected from 14 papers indexed in PubMed this week.
AI CAD scores in screening mammography flag breast cancer years before diagnosis
Radiology (RSNA) · Karolinska Institutet / Lund University, 9 June 2026Clinical applications of vision-language foundation models (VLMs)
PLOS Digital Health · University of Oxford / Queen Mary University London, 11 June 2026LLM extraction of opportunistic findings from oncology CT reports
Scientific Reports · University of Udine, 10 June 2026Deep-learning reconstruction (DLIR) for reduced-dose pulmonary CT angiography
Scientific Reports · University of Rostock, 9 June 2026LLM summaries and interactive terms to improve radiology report comprehension
Journal of the American College of Radiology · Emory University, 8 June 2026CT-PET radiogenomics with a multi-task GNN for glioma molecular markers
Neuroradiology · multi-center, 6 June 2026AI for breast cancer differentiation on POCUS vs standard breast ultrasound
Acta Radiologica · Lund University / Skane University Hospital, 10 June 2026Domain-generalized adversarial denoising for scan-wise PET
Physics in Medicine and Biology · Yale / Massachusetts General Hospital, 8 June 2026
Industry & regulation
Harrison.ai launches Harrison.Rad 1.5, first to pass the FRCR 2B exam
9 June 2026. Median 86.5 vs a 73.2 pass threshold; trained on about 6 million imaging exams. Research access only, US and EU clearances pending.Vara receives CE class IIb (EU MDR) for 3D tomosynthesis AI
10 June 2026. Already covers more than 50% of Germany's organized screening program; approval grounded in the PRAIM study (Nature Medicine 2025).Subtle Medical clears SubtleHD CT via FDA 510(k) (K254120)
10 June 2026. Image-quality enhancement through noise reduction and low-contrast detectability; deployed on more than 1,300 scanners.
Conferences & voices
SIIM 2026 (Pittsburgh, 10 to 12 June) confirmed radiology AI's shift from model performance to operations: 68% of sessions addressed AI, with an agenda centered on governance, workflow intelligence, and large-scale deployment. The joint ACR-SIIM Practice Parameter for AI in imaging, published in May 2026, was hailed as the first sector-wide governance standard. Woojin Kim was named a SIIM Fellow. Stanford AIMI's Curtis Langlotz argued that continuous post-deployment monitoring should become a core regulatory requirement, noting that 76% of FDA-authorized AI medical devices are radiology tools yet most lack robust post-market surveillance.
Source articles are indexed in PubMed with verified DOIs. Manuscript-stage work is excluded. The full weekly report is produced in French.
Get the weekly digest
One email a week on what matters in radiology and imaging AI. Researched by a practicing radiologist. Free, no vendor sponsorship.