Weekly AI digest

Three things that mattered

• AI detects pancreatic cancer on routine CT, 475 days before diagnosis

  REDMOD, an algorithm developed at the Mayo Clinic and published in Gut, identified 73% of pancreatic cancers up to 475 days before formal clinical diagnosis on routine abdominal CT, versus 39% for radiologists on the same exams, at 90% specificity. A significant step toward sub-clinical detection of a tumor whose survival depends on early diagnosis.

• The ACR builds national infrastructure for AI surveillance

  The ACR published its Assess-AI registry in JACR, a national framework for real-world, post-market monitoring of radiology AI performance. It integrates an LLM-based surrogate labeling module to enable large-scale annotation without systematic manual review, positioning the ACR as de facto steward of AI quality in US radiology.

• Abbott wins simultaneous FDA and CE clearance for AI cardiac imaging

  Abbott obtained FDA 510(k) clearance and the CE mark on the same day for Ultreon 3.0, its AI-augmented intracoronary OCT platform, which adds automated plaque quantification, minimum lumen area estimation, and real-time stent guidance. The dual approval illustrates the converging FDA and CE regulatory paths for cardiovascular imaging devices.

Peer-reviewed

• AI-assisted versus fully automated volumetry of intracranial hematomas: a clinical validation study

  European Journal of Radiology Open · university hospital center, Europe, 25 April 2026

• REDMOD: early AI imaging detection of pancreatic cancer on routine exams, 475 days before clinical diagnosis

  Gut · Mayo Clinic, Rochester, 28 April 2026

• PECertainty: a large language model to quantify diagnostic certainty in pulmonary CT angiography reports

  JMIR Medical Informatics · University of California San Francisco, 28 April 2026

• ACR Assess-AI: a national registry for real-world surveillance of AI algorithm performance

  Journal of the American College of Radiology · American College of Radiology, 29 April 2026

• Meta-analysis of AI for cervical spine fracture detection on CT: sensitivity 0.93, specificity 0.97, AUC 0.99

  Emergency Radiology · hospital institution, 2 May 2026

• CRF-GAN: high-fidelity 3D CT synthesis with optimized memory using conditional adversarial networks

  Journal of Imaging Informatics in Medicine · technical university, Asia, 1 May 2026

• No-code detection of bone metastases on bone scintigraphy with Google Teachable Machine

  Journal of Imaging Informatics in Medicine · university hospital, South Korea, 1 May 2026

• AI enhancement for 70 kVp reduced-iodine cerebral CT angiography: evaluation of DL-ACE and DL-DN

  Korean Journal of Radiology · Asan Medical Center, Seoul, May 2026

Industry & regulation

• Abbott receives simultaneous FDA 510(k) and CE mark for Ultreon 3.0

  28 April 2026. Next-generation AI intracoronary OCT imaging platform with automated plaque quantification, minimum lumen area estimation, and real-time stent guidance; successor to Ultreon 2.0 with improved resolution.

• Perimeter Medical presents Claire, first OCT plus AI system for intraoperative margin assessment, at ASBrS 2026

  FDA PMA cleared in March 2026; clinical results presented 29 April to 3 May 2026 in Orlando. Real-time tissue mapping of the surgical specimen to identify positive margins before closure in breast oncology.

• Viz.ai and the NRHA launch a national initiative for equitable access to diagnostic AI in rural settings

  30 April 2026. Deploying AI imaging analysis (stroke, pulmonary embolism, cardiac) in under-resourced US rural hospitals to reduce disparities in access to time-critical care and enable earlier transfers.

• Hyperfine completes first clinical deployment of Swoop Portable MRI in India, at AIIMS New Delhi

  29 April 2026. The 0.064 T portable MRI integrates AI image-enhancement algorithms for bedside imaging in neuro-critical care; AIIMS is the first Indian institution to deploy it.

• HOPPR launches a narrative AI model for chest radiography with automated structured report generation

  27 April 2026. The MC CXR Narrative Model combines anomaly detection and natural-language clinical text generation; positioned as complementary to existing PACS and RIS systems with API integration. HOPPR was co-founded by Woojin Kim.

Conferences & voices

The week underscored a shift toward governance and real-world accountability. Benoit Rizk, CMIO of the Swiss imaging network 3R, was selected to present a five-year, 400,000-exam AI deployment experience at SIIM 2026, reporting 91% radiologist adoption and a 33% reduction in trauma reporting turnaround. The RSNA held its annual Capitol advocacy day in Washington to defend federal funding for medical imaging research. In opinion writing, Rishi Seth (CEO, Rad AI) argued that replacing radiologists does not remove risk but transfers it to patients, while Woojin Kim framed real-world performance monitoring as the missing layer in AI radiology.

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.

By subscribing you agree to receive the weekly digest by email. Unsubscribe anytime. No sharing with third parties.