Weekly AI digest

Three things that mattered

• NeuroVFM learns from 5.24 million routine scans

  Kondepudi A et al. (Nature Medicine) introduce "health system learning": training a visual foundation model directly on the uncurated data a health system already produces, rather than on a curated public benchmark. NeuroVFM, built on 5.24 million clinical MRI and CT volumes, generates radiology reports that surpass frontier models on accuracy, clinical triage and expert preference, while reducing hallucinated findings and critical errors.

• Ultrasound transformer beats senior sonographers on enema failure

  Liu J et al. (npj Digital Medicine) prospectively validated a Vision Transformer predicting failure of air-enema reduction in paediatric ileocolic intussusception, across 5,602 children at 14 tertiary hospitals. In a prospective external cohort of 190 patients it reached 93.7 percent accuracy against 74.7 percent for senior and 60.7 percent for junior sonographers.

• Four FDA clearances land in a single week

  CliniComp cleared an EHR-embedded PACS viewer, iHealthScreen cleared iPredict-DR (K253704) for nurse-operable diabetic retinopathy detection, OptraSCAN cleared an integrated digital pathology system with HER2 analysis (K260755), and Philips launched Alturion ultrasound following 510(k) clearance and CE mark. The cluster spans PACS, retina, pathology and ultrasound in seven days.

Peer-reviewed

Selected from 295 papers indexed in PubMed this week.

• A vision transformer deep learning model for assessing pediatric ileocolic intussusception severity using ultrasound images

  npj Digital Medicine, 10 July 2026

• Health system learning enables generalist neuroimaging models

  Nature Medicine, 10 July 2026

• A unified vision-language model for precision oncology and biomarker prediction in neuroblastoma

  Nature Communications, 9 July 2026

• Intelligent surgical workflow recognition-based skill assessment for minimally invasive submandibular gland resection

  npj Digital Medicine, 7 July 2026

• Prediction of microvascular invasion in hepatocellular carcinoma using contrast-enhanced ultrasound and deep learning

  Nature Communications, 10 July 2026

• Magnetic Resonance Imaging-Based Artificial Intelligence in Predicting Prostate Cancer Biochemical Recurrence: Systematic Review and Meta-Analysis

  Journal of Medical Internet Research, 7 July 2026

• Artificial Intelligence-Assisted Screening for Patients With Diabetic Retinopathy and Age-Related Macular Degeneration in Family Medicine and Geriatric and Gerontology Care: Protocol for a Pragmatic Randomized Clinical Trial

  JMIR Research Protocols, 6 July 2026

• A deep learning algorithm for coronary heart disease prediction based on retinal fundus photographs and optical coherence tomography

  BMC Medical Imaging, 10 July 2026

Industry & regulation

• iHealthScreen clears iPredict-DR for autonomous diabetic retinopathy screening

  The FDA granted 510(k) clearance (K253704) to iPredict-DR, which detects more than mild diabetic retinopathy from colour fundus images and can be operated by a nurse or minimally skilled healthcare worker.

• OptraSCAN clears an integrated digital pathology system

  The FDA granted 510(k) clearance (K260755) to the OptraSCAN System, combining the OS-Ultra scanner with ImagePath image management and HER2 image analysis software under a pay-per-use model.

• CliniComp clears a PACS viewer embedded in its EHR

  The FDA cleared CliniComp's PACS Viewer as a Medical Image Management and Processing System, enabling diagnostic-quality viewing and AI-driven prioritisation of high-acuity studies inside a single patient record.

• Philips launches Alturion ultrasound with AI-powered workflows

  Philips released the Alturion ultrasound system in the USA and Europe following 510(k) clearance and CE marking, bundling Elevate Plus AI measurements for abdominal ultrasound to streamline acquisition and improve reproducibility.

• Viz.ai enters neurodegenerative disease with Cortechs.ai

  Viz.ai will integrate the NeuroQuant and NeuroQuant MS suite into its care coordination platform, starting with multiple sclerosis and bringing quantitative MRI analysis to US hospitals and health systems.

• Catalyst MedTech partners with APQ Health on cardiac CT analysis

  Catalyst MedTech will distribute APQ Health's coronary CTA viewer and FDA-cleared Autoplaque quantitative plaque analysis, pairing on-site image review with remote plaque characterisation and reporting.

• Natoe AI scales AI-native teleradiology across US hospitals

  Natoe AI is delivering remote reads built around AI-drafted pre-read reports, with FDA-cleared AI flagging findings before the case opens and a board-certified radiologist signing every report.

Conferences & voices

RSNA News reported on a Radiology: Artificial Intelligence study in which a deep learning model estimating pulmonary nodule malignancy risk matched clinician-level accuracy using as little as 20 percent of its training data, trained on 16,077 annotated nodules from the National Lung Screening Trial and externally tested on the Danish Lung Cancer Screening Trial. Lead author Bogdan Obreja and senior author Colin Jacobs argued that resources are better spent on diverse, well-annotated datasets than on accumulating large volumes of similar cases.

EuSoMII announced the programme for its 2026 Annual Meeting, which will take place in Heraklion, Crete, on 9 and 10 October under the theme "Sustainable Intelligence: Can Radiology AI be Green, Global, and Generational?". The meeting is still three months away, but the programme announcement is itself the signal: society president Daniel Pinto dos Santos amplified the agentic AI teaser and Jacob Visser's confirmed attendance, indicating that the autumn agenda will move past detection towards autonomous agents in the department.

Curt Langlotz confirmed he will speak at RANZCR Intelligence26 in Sydney on 24 and 25 July, a meeting that has been moved to a larger venue after member demand exceeded capacity.

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.