Weekly AI digest

Three things that mattered

• Foundation models for CT and MRI reconstruction reach clinical maturity

  TAMP, a foundation model pre-trained on 10.8 million simulated CT projections (Nature Communications), universally improves image quality for low-dose, high-speed, and degraded-protocol CT, adapting to specific clinical scenarios from only a few representative slices. A retrospective MRI study at Mainz (8,183 exams) shows deep-learning reconstruction cuts protocol time by 13% and lifts patient throughput by 7.2%.

• Four FDA 510(k) clearances in one week

  Philips Elevate Plus (ultrasound, with integrated Koios AI), GE HealthCare MIM Contour ProtegeAI+ 2.0 (radiation oncology auto-contouring), SimBioSys TumorSight Viz 1.3 (3D breast surgical planning), and Cercare Perfusion CBCT (K253831, intraoperative cerebral perfusion) all cleared, with MIM's clearance notable for a Predetermined Change Control Plan that allows new models without individual re-clearance.

• Governance and agency frameworks shape the SIIM run-up

  The joint ACR-SIIM Practice Parameter for AI in imaging, adopted in May 2026, continues to structure the field ahead of the SIIM annual meeting in Philadelphia. Thought leaders highlight post-deployment surveillance (Langlotz, Stanford) and a four-level framework for AI agency in the reading room (RSNA Radiology:AI).

Peer-reviewed

• TAMP: a universal-enhancement foundation model for non-ideal computed tomography

  Nature Communications · Southeast University / Case Western Reserve University, 4 June 2026

• Impact of deep-learning reconstruction on MRI workflows: a retrospective analysis

  RoFo · University Medical Center Mainz, 3 June 2026

• Deep-learning prediction of diffusion-FLAIR mismatch without FLAIR in acute stroke

  Scientific Reports · UNIST, South Korea, 2 June 2026

• Machine-learning models for acute ischemic stroke segmentation: systematic review and meta-analysis

  Biomedical Engineering Online · Shahid Beheshti University of Medical Sciences, Tehran, 2 June 2026

• Multimodal AI for early prediction of acute pancreatitis complications on CT

  Abdominal Radiology · Istanbul University / Northwestern University, 2 June 2026

• ZTE MRI with deep-learning reconstruction and chemical-shift correction for osteolytic myeloma lesions

  European Radiology Experimental · UCLouvain / GE HealthCare, 5 June 2026

• Automated AI segmentation of the complete anterior visual pathway on high-resolution MRI

  European Radiology Experimental · University of Genoa / Neurocenter Lugano, 5 June 2026

Industry & regulation

• Philips clears Elevate Plus (FDA 510(k) and CE) for EPIQ Elite and Affiniti ultrasound

  2 June 2026. AI and automation suite with Auto Measure Abdomen (accuracy above 93%), integrated on-cart Koios AI for breast and thyroid lesion classification (Bi-RADS and Ti-RADS, more than 350,000 pathologically proven cases), XRes Pro+, and Super Res MVI Pro. Aims to reduce inter-operator variability.

• GE HealthCare clears MIM Contour ProtegeAI+ 2.0 (FDA 510(k)) for radiation oncology

  4 June 2026. AI auto-contouring with new MR brain and updated CT male pelvis models, direct export to treatment planning systems. Clearance includes a Predetermined Change Control Plan, allowing future anatomy and modality models without individual re-clearance.

• SimBioSys clears TumorSight Viz 1.3 (third FDA 510(k)) for 3D breast surgical planning

  3 June 2026. 3D tumor visualization from conventional breast MRI, with improved AI segmentation, real-time tumor volume estimation, and PACS integration for day-zero results. Validated on more than 1,600 multi-institutional cases, targeting the above-20% re-excision rate after breast-conserving surgery.

• Cercare clears Perfusion CBCT (FDA 510(k), K253831) for intraoperative cerebral perfusion

  28 May 2026. Qualitative cerebral perfusion assessment directly from cone-beam CT in the angiography suite during neurointerventional procedures, providing rCBF, rCBV, MTT, Tmax, OEF, and CMRO2. Validated in a multi-reader study at Inselspital Bern with 13 independent neuroradiologists.

Conferences & voices

The May 2026 ACR-SIIM Practice Parameter for AI in imaging continued to frame the field ahead of the SIIM annual meeting in Philadelphia, where the UVA study on interactive multimedia reporting and new Assess-AI registry results are expected. A retrospective Radiology study at Massachusetts General Hospital (119,662 exams) found that AI slabbing for digital breast tomosynthesis improves specificity without losing sensitivity. Among the voices: Curtis Langlotz (Stanford) announced a 3.8 million dollar ARPA-H grant for ARISE to measure human-plus-AI collaboration; Pranav Rajpurkar (Harvard) co-authored UniBiomed, a universal foundation model for grounded biomedical image interpretation; and the RSNA Radiology:AI editorial team proposed a four-level framework for AI agency in radiology, from observation to execution.

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.