FDA clears AI for multi-disease screening on chest CT (40M/yr reach)

RadAI Slice Newsletter

Weekly Updates in Radiology AI

Good morning, there. The FDA cleared HeartLung's AI-CVD for opportunistic screening of 11 conditions on routine chest CT, with applicability to 40 million annual scans.

I see this as a pivotal step for radiology: automated multi-disease detection on routine CT can broaden the clinical and preventive value of almost every chest scan. This feels especially significant as it integrates cardiovascular, metabolic, and bone health risk assessment and could transition radiology further into population health management. Practices may need to revise workflows and counseling practices to maximize benefit and ensure follow-up.

How would you integrate multi-disease AI findings into your reporting and workflow?

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

• FDA greenlights AI for comprehensive CT-based risk screening

• LLMs outperform RadLex for radiology report language expansion

• Deep learning surpasses radiologists on head and neck cancer ENE detection

• AI maps global levers in cancer outcomes: radiotherapy access tops the list

• Plus: 3 newly released datasets, 3 FDA approved devices & 3 new papers.

🩺 FDA greenlights AI for comprehensive CT-based risk screening

Image from: Cardiovascular Business

RadAI Slice: A newly cleared AI performs 11 opportunistic screenings on chest CTs without extra imaging or dose.

The details:

• FDA 510(k) for HeartLung AI-CVD covers 11 key conditions on a single scan

• Findings assessed: coronary calcium, valves, volumetry, liver and bone density

• Incidental risk flagged for heart disease, stroke, osteoporosis, diabetes, steatosis

• Applies to both new and prior scans, supporting retrospective and routine use

• Estimated US use case: ≈40 million chest CTs/year

Key takeaway: This broadens CT’s preventive utility, ushering radiologists into expanded roles in risk detection and counseling. Practices should plan for workflow changes and multidisciplinary collaboration.

🔤 LLMs outperform RadLex for radiology report language expansion

RadAI Slice: An AJR study finds LLMs recognize more terms and synonyms than RadLex for report processing.

The details:

• LLM generated 208K variants, 69K synonyms, boosting coverage to 81.9% vs 67.5% (RadLex)

• Semantic recall improved to 81.6% (LLM) vs 64% (RadLex), F1: 0.91 vs 0.86

• Tested on chest CT reports from 5 international datasets

• Precision slightly lower (94.8% vs 100%) but tradeoff yielded higher recall

Key takeaway: Automated term expansion with LLMs can enhance NLP, standardized reporting, and multi-site research. Reviewing semantic nuances and safety will be important for future adoption.

🧠 Deep learning surpasses radiologists on head and neck cancer ENE detection

RadAI Slice: External validation shows AI decisively outperforms radiologists for ENE in laryngeal and hypopharyngeal SCC.

The details:

• Tested on 1,954 nodes from 289 patients across 3 sites

• DeepENE AUC: 0.96–0.90 vs radiologists: 0.85–0.66–0.71

• AI sensitivity up to 97% vs 77% (radiologists)

• Prospective trials are planned

Key takeaway: AI tools like DeepENE may deliver more consistent, sensitive pretreatment ENE assessment, aiding precision oncology. Broader clinical validation is the next step.

🌍 AI maps global levers in cancer outcomes: radiotherapy access tops the list

Image from: EurekAlert

RadAI Slice: Machine learning on global data pinpointed radiotherapy access, UHC, and wealth as survival keys.

The details:

• Analyzed data from 185 countries, with focus on radiotherapy, coverage, economics

• Radiotherapy access and UHC led as drivers of improved survival

• Model offers country-specific policy roadmaps and readiness comparisons

• Published in Annals of Oncology

Key takeaway: Infrastructure (including imaging/radiotherapy) and policy remain core to cancer outcomes. This provides actionable evidence for health investment—extending beyond local/regional contexts.

BreastDCEDL (2026-01-08)

Modality: MRI | Focus: Breast | Task: Segmentation, Outcome prediction

• Size: 2,070 patients, ~11,700 3D scans

• Annotations: 1,154 3D tumor segmentation masks (I-SPY), 916 tumor bounding boxes (Duke)

• Institutions: Ariel University, NF Algorithms & AI, et al.

• Availability:

  Public (Zenodo link)

• Highlight: Largest standardized multi-center breast DCE-MRI dataset for deep learning, with harmonized metadata and unified tumor annotations.

HQColon (2026-01)

Modality: CT | Focus: colon | Task: segmentation, anatomical modeling

• Size: 435 CT scans, 315 patients

• Annotations: Expert-validated segmentations: whole colon (gas + fluid) and gas-only masks

• Institutions: University of Copenhagen, Bispebjerg Hospital et al.

• Availability:

  public (OSF)

• Highlight: First open-access, expert-validated CT colon segmentation set including fluid and gas regions

CAPTURE (2026-01)

Modality: X-ray | Focus: Chest, Paediatric | Task: Classification, Segmentation

• Size: 9686 CXRs from children <15 years, collected from ~20 studies worldwide

• Annotations: Radiologist/paediatric TB expert consensus reads: TB-typical, not-typical, normal, unreadable; clinical TB classification

• Institutions: Stellenbosch University, FIND et al.

• Availability:

  Request-only, via agreement with FIND Global Access Policy (link)

• Highlight: First large, global, diverse, expertly-annotated paediatric CXR dataset for TB AI, supporting both algorithm evaluation and training.

🏛️ FDA Clearances

• K253532 - TruSPECT Processing Station: AI-powered denoising for SPECT/CT, validated multicenter, supporting cardiac imaging diagnostics.

• K253844 - AnyScan 3.0 NM Scanner: Multimodal SPECT/PET/CT clinical platform, approved for broad lesion, tumor, and organ imaging in various diseases.

• K253343 - Celebrace dental software: uses patient 3D imaging (CBCT) to automate orthodontic device treatment planning and design.

📄 Fresh Papers

• doi:10.2214/AJR.25.34243 - LLM expansion boosts RadLex term coverage and recall for chest CT radiology NLP across 5 countries; F1: 0.91 vs 0.86 (RadLex).

• doi:10.1038/s41746-025-02222-9 - AI model integrating PET myocardial perfusion markers boosts CAD diagnostic accuracy (AUC 0.83) across multicenter cohorts vs. physicians.

• doi:10.1038/s41598-025-34972-7 - Multicenter open-source 4D flow MRI pipeline quantifies advanced left atrial hemodynamics with robust segmentation and improved disease state discrimination.

• Browse 148 new radiology AI studies from last week.

📰 Everything else in Radiology AI last week

• HeartLung's AI-CVD receives FDA clearance, enabling routine CT to detect multiple diseases in a single scan.

• Survey finds 87% of patients favor AI for imaging diagnosis, but only 28% approve AI triage.

• Google publicly released MedGemma 1.5 and MedASR, vision-language and speech recognition AI models for imaging and healthcare.

• FDA published updated guidance for wellness and clinical decision support software to encourage AI innovation, including in radiology.

• Researchers use AI and advanced microscopy for first visualization of ExHOS, a key cellular nanomachine relevant to cancer and infection.

• 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.