5 year AI breast risk enters NCCN at 1.7%

Good morning, there. NCCN now recommends mammography AI risk tools and uses a 1.7% 5 year risk threshold.

This is one of the clearest signs yet that imaging AI is moving into routine breast workflow. It matters because AI output may now influence when MRI or ultrasound is added and push risk assessment earlier to age 35.

How would your breast imaging workflow change if AI risk scoring became routine?

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

• NCCN backs AI risk scoring from screening mammograms

• FDA keeps radiology AI under 510k review

• Multicenter LLM trims impression drafting time

• CT angiography AI helps rule in ischemia

• Plus: 1 newly released dataset, 6 FDA approved devices & 4 new papers.

🩺 NCCN backs AI risk scoring from screening mammograms

Image from: Radiology Business

RadAI Slice: This looks like one of the most practice changing breast AI developments of the week.

The details:

• NCCN starts AI risk assessment at age 35, not 40

• AI risk of 1.7% or more can trigger MRI or ultrasound review

• Guidance calls for repeat risk assessment over time

• This creates a direct workflow link from screening to risk based imaging

Key takeaway: AI risk scoring is moving from a supplemental tool toward a guideline linked workflow input for screening escalation and follow up planning.

🏛️ FDA keeps radiology AI under 510k review

RadAI Slice: The message here is clear: imaging AI oversight is not loosening soon.

The details:

• FDA said prior clearance does not prove future manufacturer competence

• At least 255 radiology AI devices received 510k clearance in 2025

• Agency rejected relying on radiologists to catch performance issues

• Decision affects CADe, CADx, CADt, and combined categories

Key takeaway: Evidence, validation, and formal review remain central for imaging AI vendors and for buyers assessing long term product risk.

🧠 Multicenter LLM trims impression drafting time

RadAI Slice: This reporting paper stands out for both scale and workflow relevance.

The details:

• Training used 1.87 million reports from 42 hospitals

• External test included 27471 reports

• Radiologists rated outputs at least as good in 69.0% of comparisons

• Draft time fell by 0.46 minutes per report

Key takeaway: For high volume services, impression generation looks increasingly practical when paired with radiologist review and local validation.

🫀 CT angiography AI helps rule in ischemia

RadAI Slice: This study stands out because it ties AI outputs to a concrete downstream decision point.

The details:

• Analysis included 820 patients and 2339 vessels across 2 studies

• All 1% to 24% stenoses were ruled out in CREDENCE

• Most 70% to 99% stenoses were ruled in with 93% prevalence

• Key thresholds were PAV above 14.7% and ALA below 3.9 mm2

Key takeaway: AI assisted CCTA triage could make coronary CT more actionable when deciding whether ischemia testing can be deferred or escalated.

ENHANCE.PET 1.6k (2026)

Modality: PET/CT | Focus: whole-body, multi-organ | Task: segmentation, multimodal analysis

• Size: 1,683 PET/CT scans from 1,683 participants

• Annotations: CT-derived segmentations for 130 non-pathological regions; organs, cardiac structures, muscles, bones, ribs, vertebrae, and L3 body composition

• Institutions: Medical University of Vienna, University of California Davis, et al.

• Availability:

  public (ScienceDB)

• Highlight: Large multi-center whole-/total-body FDG-PET/CT dataset with verified CT-based labels for 130 healthy anatomical regions.

🏛️ FDA Clearances

• K254207 - AiORTA Plan v2.0 cleared for CTA based aortic aneurysm assessment, segmentation, 3D sizing, and procedure planning in adults.

• K254015 - CoLumbo C Spine cleared as AI software to automatically process cervical spine radiologic images for clinical assessment support.

• K260524 - United Imaging cleared the uMI Panvivo PET series for whole body PET imaging and clinical diagnosis support.

• K252709 - A wireless ultrasound probe scanner cleared to support portable real time imaging in a range of diagnostic settings.

• K253950 - Avatar Medical Vision cleared for 3D CT and MR review on standard displays, autostereoscopic screens, and VR headsets.

• K253029 - RW 1 cleared to process sequential chest fluoroscopy images and visualize dynamic intensity changes for supplementary review.

• Explore last week's 9 radiology AI FDA approvals.

📄 Fresh Papers

• doi:10.1038/s41746-026-02629-y - A prospective multicenter ultrasound model trained on 5258 videos improved focal liver lesion detection for junior and senior sonographers.

• doi:10.1148/ryai.250749 - In 3031 emergency CT angiograms across 10 hospitals, a commercial occlusion AI matched clinical reporting and found missed cases.

• doi:10.1007/s00330-026-12489-5 - A seven hospital fracture AI evaluation found 89% sensitivity and 88% specificity but weaker performance in old fractures and small bones.

• doi:10.1007/s00261-026-05497-y - A hybrid LLM and rule based system tracked 8389 pancreatic cysts across 28489 reports with 96% agreement against adjudicated matching.

• Browse 135 new radiology AI studies from last week.

📰 Everything else in Radiology AI last week

• ACR expanded resources for practices reviewing imaging AI, including a repository of FDA cleared pixel based algorithms.

• In 100 oncologic CT and MRI cases, GPT 4o matched board certified radiologists on follow up recommendation quality.

• The new ARRS president urged radiologists to shape AI adoption now rather than inherit decisions made by others.

• A public radiology leadership rebuttal argued replacement claims ignore technical limits, oversight needs, and weak cost savings.

• Researchers used nearly 100000 head CTs to derive cardiovascular risk signals and estimated coronary calcium from noncardiac imaging.

• 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.