Weekly AI digest

Three things that mattered

• Aidoc receives FDA Breakthrough Device Designation for AI-drafted radiology reports

  Aidoc received FDA Breakthrough Device Designation for First Read, an AI system that generates complete radiology report drafts. This is the first such designation for AI-authored reporting text, enabling an expedited FDA review pathway. If cleared, First Read would be the first FDA-authorized system for full AI-drafted radiology reports, a category-defining regulatory milestone that reframes the competitive landscape for autonomous reporting.

• LLM-assisted structured reporting reaches near-human performance in prospective chest CT evaluation

  A European Radiology study published during the week evaluated a GPT-4-class model adapted for chest CT structured report generation across a 500-report test set. Mean radiologist corrections per report were fewer than 2, with no clinically significant errors in the evaluated cohort. The finding directly addresses the structured reporting bottleneck and provides evidence-level support for supervised LLM integration into reporting workflows.

• Federated learning delivers an 18-point F1 gain for rare thoracic disease detection without data sharing

  An npj Digital Medicine paper demonstrated that federated learning across multiple institutions achieved an F1 score 18 percentage points higher than single-site training for rare thoracic pathologies on chest radiographs, without transferring patient data. The result establishes a validated privacy-preserving methodology for conditions where no single center accumulates sufficient training volume.

Peer-reviewed

• Deep learning pipeline for automated DEXA bone density measurement achieves radiologist-level accuracy in multi-site cohort

  Osteoporosis International, multi-site, 24 June 2026

• GPT-4-class LLM adapted for chest CT structured reporting requires fewer than 2 corrections per report in 500-report evaluation

  European Radiology, 23 June 2026

• CNN trained on 12,000+ pulmonary nodule CT series recommends Fleischner-compliant follow-up with AUC 0.91, non-inferior to radiologists

  Radiology, 25 June 2026

• Vision transformer for retrospective MRI motion artifact detection and k-space correction reduces repeat scan rates in brain imaging

  Magnetic Resonance in Medicine, 22 June 2026

• Multi-center validation of commercial organ segmentation AI across 1,200 abdominal CT scans yields Dice scores above 0.94 for all organs

  Abdominal Radiology, 21 June 2026

• Computer vision models reach LLM-level performance on incidental breast finding labeling in chest CT at lower compute cost

  Journal of Breast Imaging, 26 June 2026

• Federated learning framework for rare thoracic pathology detection achieves F1 18 points above single-site baseline without data transfer

  npj Digital Medicine, 24 June 2026

Industry & regulation

• Aidoc First Read receives FDA Breakthrough Device Designation for AI-drafted radiology reports

  On 25 June, Aidoc announced FDA Breakthrough Device Designation for First Read, the first AI system producing complete radiology report drafts to receive this status. The designation triggers an expedited FDA review pathway, typically reducing authorization timelines by 30 to 50 percent. CEO Elad Walach stated the company is committed to working with the FDA to bring the technology to patients as quickly as possible. Aidoc becomes the first company with an expedited FDA pathway for AI-generated radiology report text.

• GE HealthCare receives CE Mark for AIRx Lumbar Spine automated MRI analysis

  GE HealthCare received CE Mark for AIRx Lumbar Spine, covering automated disc degeneration grading, vertebral height measurement, and stenosis flagging on MRI. The system was validated across three EU sites and targets high-volume spine MRI workflow augmentation.

• Philips expands Radiology Operations Command Center partnership to eight additional NHS trusts

  Philips added eight NHS trusts to its ROCC platform, which combines AI-based worklist triage with centralized radiology workflow management. The expansion positions Philips as the leading enterprise radiology AI infrastructure vendor in the UK public sector.

• RSNA publishes AI Transparency Framework v1.0 with structured vendor certification requirements

  RSNA released its AI Transparency Framework v1.0, establishing requirements for AI vendors seeking RSNA exhibit certification. The framework covers training data provenance, subgroup performance benchmarks, and adverse event reporting pathways. It is the first industry-facing standard from a major US radiology society.

Conferences & voices

The 2nd RCR Global AI Conference (29 to 30 June, London) featured Curtis Langlotz of Stanford and RSNA president as keynote speaker. His presentation addressed a task-based quantitative analysis of AI workforce impact, modeling radiologist demand changes by modality and role. Langlotz promoted the conference actively on LinkedIn during the week of 21 to 27 June. Elad Walach, CEO of Aidoc, commented publicly on the Breakthrough Device Designation, emphasizing the company's commitment to expedited FDA collaboration. ESR updated the iGuide clinical decision support platform with AI-generated referral appropriateness scores for 14 additional categories, bringing the total to over 340 indications, with SMART on FHIR integration confirmed for Q3 2026.

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.