Weekly AI digest

Three things that mattered

• Randomized trial lifts genetic accuracy in retinal disease

  A 300-participant randomized controlled trial of Retina4IRD (Nature Medicine) raised top-5 genetic accuracy before sequencing to 88.5% with AI assistance versus 67.3% for specialists alone (P < 0.001). Downstream management scores also improved (37.7 versus 28.5), making this one of the rare imaging-AI studies with a prospective randomized design and a hard clinical endpoint.

• Auditable chest X-ray foundation model built on concepts

  CLEAR (Nature Biomedical Engineering), trained on over 0.87 million image-report pairs from 239,391 patients, projects chest X-rays into a concept space so every prediction decomposes into named radiological observations. It was validated across the United States, Europe and Asia and targets the auditability that post-market monitoring now expects.

• Speech model tracks Huntington's disease progression

  NDSNet (Lancet Digital Health) estimated clinical scores from speech in Huntington's disease with a relative error of 11.4% and an intraclass correlation of 0.87, similar to clinician ratings (0.847). Predictions were grounded against striatal atrophy on MRI, with a signal already present at presymptomatic stages.

Peer-reviewed

Selected from 266 papers indexed in PubMed this week.

• AI-based clinician decision support system for diagnosis of inherited retinal diseases: a multicenter, randomized trial

  Nature Medicine, 24 July 2026

• CLEAR: an auditable foundation model for radiology grounded in clinical concepts

  Nature Biomedical Engineering, 22 July 2026

• A CNN-based liver ultrasound plane recognition system with multimodal large language model-generated interpretable feedback

  Abdominal Radiology, 22 July 2026

• Interpretable multimodal deep learning for time-resolved survival prediction after hepatocellular carcinoma resection

  npj Digital Medicine, 20 July 2026

• A deep learning model for speech-based prediction of clinical scores in people with Huntington's disease: a longitudinal study with cross-sectional replication

  Lancet Digital Health, 23 July 2026

• Preoperative prediction of positive surgical margins in prostate cancer using multimodal deep learning model: a multicenter study

  Abdominal Radiology, 25 July 2026

• Chinese expert concern and consensus on applications of artificial intelligence in clinical cancer imaging

  Insights into Imaging, 24 July 2026

• Integrating habitat-based radiomics, 3D deep learning, and clinical variables for detection of early-stage esophageal squamous cell carcinoma on noncontrast chest CT: A multicenter study

  European Journal of Radiology, 23 July 2026

Industry & regulation

• CARPL.ai raises 10 million dollar Series A

  CARPL.ai closed a 10 million dollar Series A led by the International Finance Corporation to scale its radiology AI marketplace of more than 300 applications across North America, Latin America, Europe and Asia-Pacific.

• VELMENI cleared for AI dental CBCT imaging

  VELMENI received FDA 510(k) clearance for VELMENI for DENTISTS (V4D) 3D, extending its dental AI portfolio into AI-assisted cone-beam CT segmentation, measurement and reconstruction.

• HOPPR launches a chest CT foundation model

  HOPPR released its EF Chest CT Narrative foundation model, extending its portfolio to a third imaging modality after chest X-ray and mammography.

• Telefonica Tech and Harrison.ai deploy chest X-ray AI in Spain

  Telefonica Tech partnered with Harrison.ai to roll out a CE class IIb chest X-ray AI, trained on more than 780,000 studies and flagging up to 124 findings, across Spain's regional health services.

Conferences & voices

Governance and auditability framed the week. A Chinese multidisciplinary panel of 81 experts reached consensus on 30 statements for AI in cancer imaging in Insights into Imaging, while trade coverage of the EU AI Act argued that AI literacy and post-market monitoring duties will require documented local audits of deployed models rather than paperwork compliance.

At RANZCR's Intelligence26 AI conference, which opened on 24 July, Stanford's Curt Langlotz used the keynote to argue that AI will shift rather than replace radiologist tasks, absorbing rising imaging volume rather than displacing specialists.

Amine Korchi read Sectra's EUR 1.7 million acquisition of Oxipit, the first CE-marked autonomous chest X-ray AI, as a sign that being first to autonomous AI is not the same as being ready for the market, though he still expects autonomous tools to arrive as imaging demand outpaces radiologist supply.

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 radiologist and imaging informatics professional. Free, no vendor sponsorship.

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