PE AI 99.2% sensitivity; MRI foundation models advance clinical AI

Good morning, there. Radiologist-AI teams reached 99.2% sensitivity for pulmonary embolism on 32,500 CTPA exams.

This study highlights that AI performs best when augmenting, not replacing, radiologists. Gains in acute PE detection show practical clinical benefit.

How comfortable are you integrating AI into acute PE workflows?

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

• Radiologist-AI Collaboration Boosts PE Detection

• MRI Foundation Models & Standardization

• FDA Early Sepsis AI Reduces Mortality

• CT Scout AI Cuts Dose & Overscanning

• FDA Clears Prostate MRI AI

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

🩺 Radiologist-AI Collaboration Boosts PE Detection

Image from: Radiology Business

RadAISlice: Multicenter study demonstrates concrete gains from AI-augmented radiologist diagnosis.

The details:

• 32,500 CTPAs analyzed across multiple centers (2021–2023)

• AI and radiologists agreed in 97.8% of cases

• Radiologists confirmed 85% of AI positives; 15% found by radiologists alone

• AI-informed radiologists achieved 99.2% sensitivity for PE

Key takeaway: AI is most effective as a supportive tool; radiologist oversight remains critical.

🧠 MRI Foundation Models & Standardization

RadAISlice: Foundation models could mark the next step for scalable MRI AI.

The details:

• OmniMRI trained on 80 datasets (19.4M slices), SSIM 0.87 for reconstruction

• Segmentation DICE: 0.77 on benchmarks

• Most MRI AI remains task-specific, limiting broad clinical use

Key takeaway: Standardization is key to generalizable clinical MRI AI.

💡 FDA Early Sepsis AI Reduces Mortality

Image from: EurekAlert

RadAISlice: Demonstrates real-world clinical impact of AI beyond imaging.

The details:

• Identifies sepsis 2–48h earlier than clinicians

• Deployed in multiple hospitals, 18% lower mortality

• Eligible for NTAP Medicare/Medicaid billing

Key takeaway: High-impact FDA approvals set precedent for AI adoption in clinical workflows.

📏 CT Scout AI Cuts Dose & Overscanning

RadAISlice: Addresses a common daily CT workflow challenge.

The details:

• Internal planning captured anatomy in 98% of cases

• External chest coverage 92.9%

• Chest DLP fell 19%, upper abdomen DLP fell 25%

Key takeaway: Automated range planning standardizes setup and lowers dose without reader burden.

🧬 FDA Clears Prostate MRI AI

RadAISlice: Supports structured review and improves workflow consistency.

The details:

• 510k cleared CADe/CADx for men 40+

• Uses T2W, ADC, DWI and prostate segmentation

• Clinical validation on 247 exams

Key takeaway: Adds a cleared, practical option for routine prostate MRI workflows.

OpenPOCUS (April 2026)

Modality: US (POCUS) | Focus: Lung | Task: Image classification, pathology detection

• Size: 1,871 video clips and 324,027 frames from 226 adult patients.

• Annotations: Clip-level expert consensus labels: normal, B-lines, consolidation, both, or indeterminate. Includes lung zone, COVID status, diagnosis, and metadata. No segmentation masks.

• Institutions: Stanford University School of Medicine

• Availability:

  Public: GitHub

• Highlight: Large open-source lung POCUS dataset with multi-device scans, healthy controls, and COVID/non-COVID cases.

🏛️ FDA Clearances

• K254189 - Fujifilm's Synapse 3D v7.2 offers advanced 2D/3D imaging tools for brain, liver, and cardiac CT/MRI—including segmentation and functional quantification.

• K253682 - DeepHealth ProstateAI aids prostate MRI interpretation with validated 3D detection and region segmentation calibrated against multi-institutional datasets.

• K254184 - syngo Application Software (Siemens) adds AI-guided bone removal, enhanced DynaCT image quality, and motion compensation to accelerate advanced clinical workflows in angiography and spine surgery.

• K252503 - Canon's Intelligent NR uses AI for real-time noise reduction in fluoroscopic X-ray, improving contrast-to-noise by 1.1–1.5x.

• K260589 - E-Scopics ES-Series: point-of-care US device for real-time abdominal elastography and quantitative organ tissue analysis.

• K252628 - Cassie (Wesper): cloud-based AI for sleep disorder analysis from PPG and SpO₂, validated vs. PSG with OSA sensitivity ~97%.

• Explore last week's 7 radiology AI FDA approvals.

📄 Fresh Papers

• doi:10.1016/j.radonc.2026.111596 - Multitask deep learning model accurately detects and visualizes radiation-induced temporal lobe injury across endemic and non-endemic regions.

• doi:10.1038/s41698-026-01476-0 - Transformer model fuses CT, dose, and clinical data for acute hematologic toxicity prediction in rectal cancer; AUC 0.83–0.76 external test sets.

• doi:10.1186/s12880-026-02379-z - AI-driven CCTA plaque quantification showed strong agreement (ICC 0.95) with expert annotation for total and calcified plaque burden.

• doi:10.1038/s41598-026-51236-0 - A lightweight CNN (M-Net) achieved 99% brain tumor classification accuracy on three MRI datasets with integrated XAI visualizations.

• Browse 159 new radiology AI studies from last week.

📰 Everything else in Radiology AI last week

• AI avatars increased cancer patient understanding and reduced stress before radiology consults.

• AI in prostate MRI increases efficiency but raises concerns around overdetection and low specificity.

• ISMRM speakers highlighted MRI standardization and large datasets as prerequisites for clinically useful foundation models.

• An open source chest CT framework compares segmentation models without expert ground truth and exposes structure-specific failure modes.

• A chest CT vision language model generated question-driven findings for lung nodules with strong agreement to references.

• 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.