🩻 AI flagged 19.7% of cancers 6 years early

Good morning, there. Mammography AI flagged up to 19.7% of cancers 6 years early at 90% specificity.

I see this as a screening workflow signal, not a diagnosis. Earlier risk scores could help tailor intervals, supplemental imaging, and review priorities if validated prospectively.

How would you use long range AI risk scores in breast screening?

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

• 🩻 Mammography AI finds risk years earlier

• Medicare AI prior auth hits resistance

• Fujifilm lung nodule AI clears FDA

• Automated CMR trims scan time

• Plus: 2 newly released datasets, 6 FDA approved devices & 4 new papers.

🩻 Mammography AI finds risk years earlier

RadAI Slice: This Radiology study stood out for linking commercial AI scores to future breast cancer risk.

The details:

• 31,394 women and 88,963 screening mammograms from Sweden

• 3 commercial AI CAD systems tracked 10 year cancer risk

• At 6 years, AI flagged up to 19.7% of cancers at 90% specificity

• AI AUC was 0.63 to 0.67 versus 0.57 for density

Key takeaway: This could shift mammography AI from second reader support toward risk stratification, but prospective screening trials remain the bar.

🏛️ Medicare AI prior auth hits resistance

Image from: HealthExec

RadAI Slice: This policy fight matters because imaging access often turns on fast and transparent authorization.

The details:

• WISeR covers NJ, OH, OK, TX, AZ, and WA from January 2026

• CMS has not disclosed algorithm criteria or vendor model details

• House committee voted to block expansion amid denial concerns

• EFF lawsuit seeks transparency on algorithms and criteria

Key takeaway: For radiology groups, AI prior auth could change denials, scheduling, and revenue cycle work before evidence of fairness is public.

🫁 Fujifilm lung nodule AI clears FDA

RadAI Slice: I see this clearance as another sign that CT triage is moving deeper into enterprise platforms.

The details:

• 510k clearance covers Synapse Lung Nodule AI

• Designed to detect lung nodules on chest CT images

• Fujifilm can pair detection with its Synapse workflow

• Performance metrics were not included in this payload

Key takeaway: The practical question is not only detection accuracy, but how well the alert fits CT reading lists, PACS views, and follow up.

❤️ Automated CMR trims scan time

RadAI Slice: This prospective trial makes workflow automation feel measurable rather than theoretical.

The details:

• Prospective randomized CMR study included 221 patients

• Automated workflow cut exam time from 21.25 to 19.16 minutes

• Scanner idle time fell from 10.12 to 7.80 minutes

• Image quality stayed comparable at 2.74 versus 2.69

• All exams were diagnostically adequate

Key takeaway: For CMR services, small per case time savings can compound, especially when technologist workload limits throughput.

BHD (2026)

Modality: MRI, CT | Focus: brain, dementia | Task: dementia prediction, image/report phenotyping

• Size: 417,341 MRI and 846,077 CT head studies from 830,884 patients; >185 TB.

• Annotations: Radiology reports, linked EHR dementia labels, NLP brain phenotypes, and DICOM-derived sequence/body-part/contrast labels. Manual QC labels for 708 series.

• Institutions: University of Edinburgh, University of Dundee, et al.

• Availability:

  request-only via PBPP/eDRIS: eDRIS

• Highlight: Nationwide Scottish clinical brain imaging cohort linked to EHR and radiology reports in a secure safe haven.

Penn 3T-7T Paired Epilepsy Imaging (2026)

Modality: MRI | Focus: Brain, epilepsy | Task: Image synthesis, seizure localization

• Size: 30 patients; 60 paired 3T/7T MRI sessions with T1w, T2w, FLAIR/rs-fMRI data. iEEG is included for 15 patients.

• Annotations: Clinical metadata, SOZ lateralization, lesional status, and surgical outcome. Curated ictal/interictal iEEG epochs and electrode localizations for 15 patients.

• Institutions: University of Pennsylvania; Penn Epilepsy Center

• Availability:

  Public: Pennsieve

• Highlight: Paired 3T and 7T MRI with iEEG in drug-resistant focal epilepsy. Includes raw BIDS data, processed derivatives, QC, and a 3T-to-7T synthesis use case.

🏛️ FDA Clearances

• K260077 - OrthoGrid Hip AI 4.0 received FDA 510k clearance for automated processing of hip radiographs.

• K261405 - ANDI 2.2 from Imeka Solutions received FDA 510k clearance for automated radiology image processing.

• K252947 - HistoSonics Planning Tool received FDA 510k clearance to support image based ablation planning.

• K260680 - Philips EPIQ and Affiniti ultrasound systems received FDA 510k clearance for diagnostic imaging updates.

• K253103 - Varex received FDA 510k clearance for the Nexus DRF digital X ray and fluoroscopy system.

• K260398 - GE LOGIQ e received FDA 510k clearance for real time pulsed Doppler ultrasound imaging.

• Explore last week's 10 radiology AI FDA approvals.

📄 Fresh Papers

• doi:10.1007/s00330-026-12652-y - A European Radiology review found only 77 of 392 commercial AI validation studies reported subgroup performance.

• doi:10.3174/ajnr.A9473 - An AJNR validation found Aidoc cervical spine CT AI sensitivity of 89.1% versus radiologists at 94.5%.

• doi:10.1007/s00330-026-12670-w - A DICOM label standardization tool cut several CT reading times by 8% to 22%, saving an estimated 270 hours yearly.

• doi:10.1007/s00330-026-12648-8 - European Radiology benchmarked 5 VLMs for ED chest X ray reports and found wide variation in acceptability and hallucinations.

• Browse 188 new radiology AI studies from last week.

📰 Everything else in Radiology AI last week

• ONC highlighted image sharing, information blocking enforcement, and AI ready data access as radiology infrastructure priorities.

• SIIM leaders pointed to ACR SIIM AI guidance, governance, and ROI as imaging AI moves from pilots to deployment.

• Emory reported MyChart imaging links cut patient initiated image requests by 40% to 50%.

• A 20 hospital MSK ultrasound AI study improved novice high quality image rates from 80.7% to 93.3%.

• SIIM data showed several LLMs reached over 95% accuracy on numerical extraction from radiology reports.

• Read 15 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.