🩻 Chest X-ray AI turns 421k scans into a mortality risk signal

Good morning, there. A deep learning model estimated radiographic aging and mortality risk from 421,894 chest X-rays.

This week’s issue points to two important directions in imaging AI. One is visible in a 421,894-patient chest X-ray study, where AI-derived radiographic aging and aging velocity tracked long-term mortality risk. The other is quieter but arguably closer to deployment: GE HealthCare’s FDA-cleared deep learning CT reconstruction tool, which improves image quality without requiring workflow change. Together, they reflect how radiology AI is expanding in two directions at once, toward both predictive infrastructure and invisible operational utility.

If routine imaging starts producing longitudinal risk signals at scale, where does that belong: the radiology report, a downstream screening layer, or a broader population health workflow?

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

• Chest X-ray AI links accelerated aging to mortality risk

• GE HealthCare wins FDA clearance for deep learning CT reconstruction

• AI-enhanced CT heart fat measurement sharpens cardiovascular risk prediction

• AI guidance cuts novice shoulder ultrasound scan time by 34%

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

📅 Chest X-ray AI links accelerated aging to mortality risk

RadAI Slice: A very large chest X-ray AI study suggests routine imaging may capture long-horizon biological risk signals, not just present-day findings.

The details:

• Analyzed 421,894 Korean adults with chest X-rays and follow-up through 2020

• AgeNet found radiographic aging more than 5 years above chronologic age was associated with mortality hazard ratios of 1.26 to 1.52

• Faster radiographic aging velocity of at least 1.5 years per year was linked to mortality rate ratios of 1.51 to 1.71

• Signals held across all-cause, cancer, cardiac, and respiratory mortality over a median 8.5 years

Key takeaway: The bigger signal is not just that AI can estimate 'radiographic aging,' but that routine chest X-rays may become a scalable source of longitudinal risk markers. The open question is less technical novelty than clinical translation: how such outputs would be validated, owned, communicated, and acted on in real workflows.

🛡️ GE HealthCare wins FDA clearance for deep learning CT reconstruction

Image from: Radiology Business

RadAI Slice: GE HealthCare’s latest CT reconstruction clearance is a reminder that some of the most deployable imaging AI creates value without asking users to change behavior.

The details:

• Designed to increase CT resolution without workflow or protocol changes

• Improves visualization of small airways and pulmonary nodules

• Field-upgradable across Revolution Apex and Vibe platforms

• Targets high-detail imaging without increasing scan time or radiation burden

Key takeaway: Reconstruction AI remains one of the clearest adoption paths in imaging: measurable image-quality gains, minimal workflow disruption, and straightforward integration into existing scanner ecosystems.

📈 AI-enhanced CT heart fat measurement sharpens cardiovascular risk prediction

RadAI Slice: Another example of AI extracting more prognostic value from scans already being acquired in routine care.

The details:

• Included about 12,000 adults followed over 16 years using routine coronary calcium CTs

• AI quantified pericardial fat volume alongside conventional scoring

• Pericardial fat improved risk prediction in borderline and intermediate-risk groups

• No extra imaging, contrast, or added scan cost required

Key takeaway: This is the same broader pattern seen in the chest X-ray story: AI is increasing the downstream value of routine imaging by converting existing scans into richer risk signals.

🦾 AI guidance cuts novice shoulder ultrasound scan time by 34%

Image from: Radiology Business

RadAI Slice: A more practical workflow story this week: AI guidance improved speed and consistency for less experienced ultrasound operators.

The details:

• Novice operators reduced scan times by 34% with AI support

• Improved reproducibility and scan-plane accuracy were reported

• Published in Academic Radiology

• Highlights a potential use case in musculoskeletal and technically demanding ultrasound exams

Key takeaway: Not all useful imaging AI needs to be diagnostic. Guidance systems that reduce acquisition variability and training burden may prove just as important in constrained clinical settings.

DOLCHID (2026-03-19)

Modality: CBCT, H&E | Focus: jaw, dental lesions | Task: segmentation, classification

• Size: 262 paired CBCT scans and H&E images from 262 patients

• Annotations: Expert-verified segmentation masks for CBCT and region-of-interest annotations for H&E, all with subtype labels

• Institutions: Wuhan University, University of Sydney, et al.

• Availability:

  Public (figshare link)

• Highlight: First fully paired dental dataset combining CBCT and histopathology for four odontogenic lesion subtypes.

🏛️ FDA Clearances

• DEN240071 - Neuropacs received De Novo FDA clearance for AI-based neurological image analysis and workflow support.

📄 Fresh Papers

• doi:10.1038/s41598-026-46342-y - A dual-encoder hybrid ResTransUNet achieved Dice 0.95 for automated liver CT segmentation on LiTS2017.

• doi:10.2214/AJR.25.33792 - An FDA-cleared Hologic AI system identified 89.8% of true-positive and 32.0% of false-negative breast cancers in DBT.

• doi:10.1007/s10278-026-01925-z - A 3D deep learning model predicted postoperative lung cancer recurrence with AUC up to 0.93 across multicenter CT cohorts.

• doi:10.1007/s10278-026-01931-1 - A systematic review of 296 publications mapped the deep learning ultrasound segmentation landscape and benchmarking gaps.

• Browse 141 new radiology AI studies from last week.

📰 Everything else in Radiology AI last week

• NIH added $12.6M to a USC-led imaging and genomics AI effort for Alzheimer’s research, bringing total funding to $30.7M.

• EFF sued CMS over its AI-powered Medicare prior authorization pilot, raising transparency, bias, and oversight concerns.

• The ALL-RISE trial reported that AI-based FFRangio matched invasive wire-based assessment for coronary flow evaluation.

• Lucida Medical raised $11.4M to expand its AI-based prostate MRI diagnosis platform and pursue FDA clearance.

• Read 13 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.