Weekly AI digest

Three things that mattered

• Deep learning detects colorectal cancer on routine non-contrast CT at scale

  COCA (DAMO Academy / Guangdong Provincial People's Hospital) was validated prospectively on 27,433 consecutive patients, reaching 86.6 to 88.2% sensitivity at 99.5 to 99.8% specificity. Against radiologists alone, it improved detection sensitivity by 20.4% and specificity by 5.4%, opening a path to large-scale opportunistic screening without bowel preparation.

• Philips clears the first spectral CT with AI reconstruction at the detector

  The Philips Spectral CT Verida received FDA 510(k) clearance as the first spectral CT with deep-learning reconstruction integrated at the detector. It reconstructs 145 images per second, visualizes a full exam in under 30 seconds, handles up to 270 exams per day, and reports 80% AI noise reduction.

• The ACR formalizes the AI lifecycle for imaging departments

  The ACR consolidated two complementary programs: ARCH-AI, a recognition program for safe and effective AI use, and Assess-AI, an NRDR registry that continuously monitors real-world algorithm performance against national benchmarks. Per the ACR, the FDA has now authorized more than 1,400 algorithms for clinical use, around 80% in medical imaging.

Peer-reviewed

• COCA: deep-learning colorectal cancer detection on non-contrast CT, international multicenter study of 27,433 consecutive patients

  Annals of Oncology · DAMO Academy (Alibaba Group) / Guangdong Provincial People's Hospital, 21 April 2026

• DeepFAN: transformer model for collaborative AI to radiologist classification of indeterminate lung nodules on CT

  Nature Cancer · Peking Union Medical College Hospital / Deepwise Healthcare / University of Hong Kong, 22 April 2026

• Gadolinium-free deep learning on cine cardiac MRI to predict adverse left ventricular remodeling after STEMI

  International Journal of Cardiology · Beijing Anzhen Hospital, Capital Medical University / BAAI, 22 April 2026

• Calcified coronary plaque detection: CCTA vs thin-slice non-contrast CT with automatic deep-learning registration

  European Radiology · Charite Universitatsmedizin Berlin / DKFZ Heidelberg, 20 April 2026

• Generalist LLM with integrated speech recognition for radiology dictation: prospective multicenter comparison

  European Radiology · AP-HP Hopital Saint-Louis / Hopital Europeen Georges Pompidou / CHU de Nancy, 20 April 2026

• Early prediction of carotid plaque response to statins via deep learning on longitudinal multimodal ultrasound

  Insights into Imaging · Taizhou University Affiliated Wenling Hospital / Macao Polytechnic University, 21 April 2026

• Open-source LLMs for automatic pneumonia classification in pediatric chest radiograph reports

  Pediatric Emergency Care · Northwestern University Feinberg School of Medicine / Lurie Children's Hospital, 21 April 2026

Industry & regulation

• Philips clears Spectral CT Verida via FDA 510(k), first spectral CT with AI reconstruction at the detector

  16 April 2026. Dual-layer Nano-panel Precise detector, 145 images per second, full exam visualized in under 30 seconds, up to 270 exams per day, 80% AI noise reduction. Follows a CE mark obtained in 2025.

• AIRS Medical SwiftMR earns FDA 510(k) to run alongside OEM deep-learning reconstruction

  15 April 2026. On a GE HealthCare 3T scanner already running AIR Recon DL, SwiftMR cut a routine brain exam from 15 to 9 minutes. Deployed in more than 1,700 imaging centers across over 40 countries.

• Sectra completes acquisition of Oxipit, maker of the first CE class IIb autonomous chest radiograph system

  15 April 2026. Oxipit's ChestLink autonomously identifies high-confidence normal chest radiographs and removes them from the radiologist worklist. Financial terms were not disclosed.

• HOPPR launches MC CXR, a structured narrative model for chest radiographs

  23 April 2026. The MC Chest Radiography Narrative model turns chest radiographs into structured narrative text for downstream reporting workflows, supported by HOPPR's Forward Deployed Services.

• ScreenPoint Medical raises 16 million dollars to expand Transpara internationally

  20 April 2026. Includes 14 million dollars from Insight Partners and Siemens Healthineers plus 2 million dollars in research grants, alongside MASAI final results in The Lancet, a Nature Medicine study (up to 63.6% workload reduction), and an npj Digital Medicine paper. Transpara is deployed in more than 30 countries and has analyzed over 12 million mammograms.

• ACR formalizes ARCH-AI and Assess-AI for AI evaluation and real-world monitoring

  22 April 2026. ARCH-AI (ACR Recognized Center for Healthcare-AI) operationalizes best practices for safe and effective AI use; Assess-AI, an NRDR registry, continuously monitors real-world algorithm performance against national benchmarks. Per the ACR, the FDA has authorized more than 1,400 algorithms, around 80% in medical imaging.

Conferences & voices

The week's strongest voice came from Dr Benoit Rizk, Medical Director of Innovation at the 3R Swiss Imaging Network, in a video interview published by AuntMinnie Europe on 24 April. Drawing on five years of large-scale deployment across 20 private imaging centers, he addressed generative-AI sycophancy, cybersecurity, and shifting professional roles, arguing that success requires governance and regulatory alignment with the EU AI Act and MDR, not algorithms alone. He will present these lessons at SIIM 2026, with perspectives on EU AI Act compliance, MDR alignment, and EHDS requirements.

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.