🩻 CT AI detected 98% of small pancreatic cancers

Good morning, there. A Radiology study reported 98% sensitivity for CT AI on pancreatic cancers 20 mm or smaller.

I see this as a practical second reader for subtle pancreatic findings on routine CT. The noncontrast result matters because many early clues appear outside pancreas protocols.

Would this change how you review noncontrast abdominal CT?

Here's what you need to know about Radiology AI last week:

• 🩻 CT AI catches smaller pancreatic cancers

• Radiology AI recall risk gets quantified

• RSNA shows why AI integration is standards work

• Mammography AI scores rise years before cancer

• Plus: 6 FDA approved devices & 4 new papers.

🩻 CT AI catches smaller pancreatic cancers

RadAISlice: This Radiology paper stands out for small tumor sensitivity on both contrast and noncontrast CT.

The details:

• 2,251 patients from 2007 to 2022 across multiple institutions

• CECT AUC was 0.99, matching the six reader mean

• NCCT AUC was 0.93 versus 0.91 for readers

• Tumors 20 mm or smaller had 98% sensitivity on CECT and 86% on NCCT

• Analysis took 20 to 22 seconds per patient versus 63 to 78 seconds

Key takeaway: The practical shift is earlier second reader support for pancreatic cancer signs, especially when CT was not ordered for a pancreas indication.

⚠️ Radiology AI recall risk gets quantified

Image from: Radiology Business

RadAISlice: This analysis puts useful numbers behind the safety discussion around cleared imaging AI.

The details:

• 903 FDA authorized health AI devices from 1995 to 2024 were analyzed

• 43 devices were recalled, a 5% overall rate

• Radiology had 692 devices and 30 recalls, a 4.3% rate

• Radiology AI was 52% more likely to face recall than other specialties

• Missing clinical study data and off label use were key risk factors

Key takeaway: This shifts evaluation from clearance alone to evidence quality, intended use discipline, and postmarket monitoring.

🔌 RSNA shows why AI integration is standards work

RadAISlice: This Radiology review makes deployment feel less about models and more about plumbing.

The details:

• Radiology Reimagined has run at RSNA since 2020

• More than 20 vendor partners join each annual demonstration

• Scenarios now span ordering, acquisition, reporting, follow up, billing

• Standards include DICOM, HL7 FHIR, IHE, FHIRcast, and multimedia reporting

Key takeaway: The practice lesson is clear, AI adoption depends on interoperability, not another standalone result box.

🎀 Mammography AI scores rise years before cancer

RadAISlice: This Swedish screening analysis gives a measured look at AI as an early risk signal.

The details:

• 31,394 women and 88,963 screening exams were analyzed

• 12,072 women were later diagnosed with breast cancer

• At 10 years, AI flagged 12.7% to 17.0% of future cancers at 90% specificity

• At 4 years, flagged cancers rose to 24.2% to 25.2%

• AUC was 0.63 to 0.67 versus 0.57 for density

Key takeaway: This supports AI as a longitudinal risk cue, but action on early alerts will need clear supplemental imaging pathways.

🏛️ FDA Clearances

• K261519 - Sonio Suspect gained 510k clearance for AI lesion detection support in suspicious cancer imaging workflows.

• K260497 - DEEPVESSEL Plaque received 510k clearance for automated vascular plaque assessment from radiologic images.

• K253026 - Varian Respiratory Gating 2.1 cleared 510k to reduce breathing motion artifacts during CT acquisition.

• K260780 - Deski HeartFocus cleared 510k as AI support for cardiac image acquisition and quality.

• K253296 - Allengers Digital C ARM with 3D cleared 510k for mobile 2D and 3D fluoroscopic imaging.

• K253893 - Topcon Harmony Referral System with OCT Viewer cleared 510k for ophthalmic OCT image management and viewing.

• Explore last week's 8 radiology AI FDA approvals.

📄 Fresh Papers

• doi:10.1148/radiol.251581 - A Radiology LLM pipeline summarized thoracic oncologic history with 95.5% completeness and 97.9% accuracy.

• doi:10.1016/j.lanwpc.2026.101890 - CRVIA assessed pancreatic cancer vascular invasion across 5 external centers with AUC 0.943 and reader study gains.

• doi:10.1038/s41746-026-02875-0 - OVUCM integrated US, CT, and MRI for ovarian cancer tasks, externally validating AUCs of 0.833 to 0.974.

• doi:10.1002/jmri.70393 - A 4,305 patient bpMRI model improved radiologist csPCa prediction and reclassified many PI RADS 3 cases.

• Browse 335 new radiology AI studies from last week.

📰 Everything else in Radiology AI last week

• In 17,752 chest CTs, a CVM reached 81.9% sensitivity for incidental breast abnormalities versus 39.8% for an LLM.

• TA Associates backed AIRS Medical, whose SwiftMR is used by 1,700 sites and processes 6 million MRI scans a year.

• Radiology leaders now want AI tied to denials, auditability, interoperability, and measurable workflow gains.

• FDA data show radiology still leads cleared health AI, while cardiology has reached 225 algorithms.

• Read 10 fresh radiology AI news articles for last week.

That's it for today!

👋 Quick favor: drag this into your Primary tab so you don’t miss next week. Or just hit Reply with one thought.

See you next week.

P.S. We keep building free tools to accelerate your radiology work. What's the most time-consuming pain point in your day that we should help speed up? Reply and share your take so we keep building around you.