AI flags COPD on routine ECG with AUC 0.82, external validation

RadAI Slice Newsletter

Weekly Updates in Radiology AI

Good morning, there. An AI model detected COPD using ECG data with an AUC of 0.82 on external validation.

I find this approach remarkable as it turns standard 12-lead ECGs into a scalable tool for early COPD detection. This noninvasive method could enhance screening access, particularly where imaging is scarce. Integrating ECG-based AI into clinical pathways might expand radiology's role in risk stratification and disease monitoring.

How could ECG-AI tools change your approach to early disease detection?

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

• AI Uses Routine ECG for Early, Accurate COPD Detection

• FDA Clarifies LLM and CDS AI Regulation in Radiology

• 🩻 How Much Radiologist Time Does AI Save in Breast Screening?

• FDA Expands AI-Guided Echo to Existing Ultrasound Systems

• Plus: 2 newly released datasets, 5 FDA approved devices & 4 new papers.

🫁 AI Uses Routine ECG for Early, Accurate COPD Detection

RadAI Slice: Deep learning now enables COPD detection from standard ECG with strong validation.

The details:

• Studied 208,000 ECGs, 18,000 COPD cases, 49,000 controls (2006–2023)

• AUCs: internal 0.80, external 0.82 (US), 0.75 (UK Biobank)

• Model highlights P-wave ECG features with explainability tools

• Standard 10-second, 12-lead ECG supports broad, low-cost screening

Key takeaway: This AI-enabled ECG tool offers scalable, non-invasive early COPD detection—potentially allowing radiologists to lead screening in both advanced and resource-limited settings.

📑 FDA Clarifies LLM and CDS AI Regulation in Radiology

RadAI Slice: FDA’s new guidance may speed LLM CDS adoption in radiology practice.

The details:

• January 2024 update clarifies software-as-medical-device rules

• LLMs not directly analyzing images may be exempt under 4 criteria

• Text-based reporting tools likely face fewer regulatory hurdles

• AI image analyzers (CAD, denoising) remain device-regulated

Key takeaway: This policy shift lowers the regulatory bar for text-focused AI, likely accelerating CDS tool development and use—knowledge radiologists should integrate into IT and procurement planning.

🩻 How Much Radiologist Time Does AI Save in Breast Screening?

RadAI Slice: This multicenter paper quantifies time savings when substituting AI for a human breast screening reader.

The details:

• Used Norwegian BreastScreen program data (680,000 women biennially)

• Replacing a second reader with AI cuts screen-reading by 50%

• Total radiologist workload drops from 9% to 4.5% (modest in context)

• Routine, not consensus, reads most affected

Key takeaway: AI as a second reader could relieve some breast radiologist shortages but real-world gains are moderate, emphasizing the importance of targeted workflow analyses before implementation.

🫀 FDA Expands AI-Guided Echo to Existing Ultrasound Systems

Image from: Cardiovascular Business

RadAI Slice: FDA expands AI echocardiography guidance to a wide range of legacy and mobile ultrasound equipment.

The details:

• AI guidance supports handheld, laptop, and cart-based echo units

• No new hardware required—reducing hospital capital barriers

• Scales focused cardiac ultrasound (FoCUS) to novice users

• May offset staffing gaps for cardiac imaging access

Key takeaway: AI-enabled echo guidance on existing equipment can widen cardiac imaging access to non-experts—helpful for hospitals seeking rapid, cost-effective service expansion.

Brain-Mets-Lung-MRI-Path-Segs (2025-11-20)

Modality: MRI | Focus: Brain, Lung metastases | Task: Segmentation, Multimodal analysis

• Size: 111 scans, 103 unique patients

• Annotations: Manual segmentations of enhancing tumor and peritumoral edema; histopathology slide images

• Institutions: Yale School of Medicine, Visage Imaging et al.

• Availability:

  public (TCIA link)

• Highlight: Matched pre-op MRI, expert tumor segmentations, radiomics, and digitized pathology for brain metastases from lung cancer

CURVAS (2025-01)

Modality: CT | Focus: Abdomen, Multi-organ | Task: Segmentation, Uncertainty quantification

• Size: 90 scans from 90 patients

• Annotations: Voxel-wise segmentations of pancreas, liver, and kidneys by 3 raters

• Institutions: Universitätsklinikum Erlangen, Sycai Technologies SL et al.

• Availability:

  Public (Zenodo link)

• Highlight: Multi-rater expert annotations for 3 abdominal organs in contrast CT, designed for calibration and uncertainty analysis.

🏛️ FDA Clearances

• K253122 - Sim&Size received 510(k) as neurovascular planning software supporting device and vessel visualization for treatment planning.

• K252379 - GE's AIR Recon DL MRI image reconstruction: FDA-cleared for high-quality images to improve diagnosis and workflow.

• K252557 - Philips' Lumify portable ultrasound system gains clearance, enhancing point-of-care and real-time detection capabilities in a mobile format.

• K252856 - PeekMed web: FDA-cleared radiology image processing platform for automated analysis, aimed at orthopedic and surgical planning support.

• K252068 - Oxos Medical's MC2 portable X-ray system gets 510(k), offering mobile fluoroscopic and standard radiography at the point of care.

• Explore last week's 6 radiology AI FDA approvals.

📄 Fresh Papers

• doi:10.1016/j.jocmr.2026.102684 - Harvard’s prospective pilot shows AI-enhanced CMR reliably quantifies coronary sinus flow and reserve after exercise, correlating with perfusion imaging.

• doi:10.1016/j.jocmr.2026.102683 - Novel HR-MRI deep learning model for carotid artery segmentation and stenosis evaluation shows high agreement with DSA and multicenter validation.

• doi:10.1158/1078-0432.CCR-25-3080 - Deep learning-derived sarcopenia from CT predicts anti-EGFR therapy benefit in mCRC; high muscle-bone ratio patients showed superior survival (HR 0.41).

• doi:10.1038/s41597-025-06473-9 - New open multi-rater multiorgan abdominal CT dataset (90 scans, 3 annotators) supports research on uncertainty modeling and calibration in abdominal segmentation.

• Browse 152 new radiology AI studies from last week.

📰 Everything else in Radiology AI last week

• GPT-4o outperformed radiologists in CT protocol selection with context-aware LLM prompting, potentially freeing radiologist time.

• Commercial AI software for chest CT/X-ray could assist triage and resource allocation in future pandemics if adapted early.

• Stanford researchers project AI could reduce radiologist hours by up to 49% in five years, but workforce demand remains stable.

• Large Chinese study: XGBoost model beat logistic regression in lung cancer risk prediction, highlighting value for targeted screening.

• LLMs like GPT-o1 automate PI-RADS scoring with high accuracy for high-risk lesions on 146 prostate MRI reports; mid-risk classification remains challenging.

• 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.