AI brain platform triples UK stroke patient independence

RadAISlice Newsletter

Weekly Updates in Radiology AI

Good morning, there. NHS England’s AI stroke platform tripled functional independence rates.

Efficient AI-driven triage in stroke care dramatically reduced time to treatment, improved functional outcomes, and increased advanced procedure rates nationwide. This broad-scale, real-world impact signals a new era for AI in imaging-driven clinical workflows, highlighting potential benefits for both patients and radiologists.

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

• NHS AI Stroke Platform Triples Functional Independence

• Agentic AI Bridges Imaging Interoperability Gaps

• AI Model Validated for Diverse Lung Cancer Risk Prediction

• AI Breast Imaging Yields High NPV but Raises Recalls

• Plus: 3 newly released datasets, 6 FDA approved devices & 4 new papers.

🚀 Product Update

X-ray Interpreter now supports multiple image uploads 📤. Plus, our new Auto Select feature ⚡ can automatically pick key images from a full DICOMDIR folder, saving you time and effort.

👉 Try it now

🧠 NHS AI Stroke Platform Triples Functional Independence

Image from: Health Imaging

RadAISlice: A national AI brain imaging deployment has transformed stroke recovery and care efficiency.

The details:

• Brainomix 360 used at all NHS stroke centers since summer 2023

• Over 60,000 patient cases analyzed using CT, CTA, CTP, and MRI

• Time to treatment fell from 140 to 79 minutes post-implementation

• Functional independence jumped from 16% to 48% of patients

• Mechanical thrombectomy procedures rose by 50%

Key takeaway: Robust clinical implementation of imaging AI can drive major real-world improvements in patient outcomes, operational efficiency, and advanced care adoption across an entire health system.

🤖 Agentic AI Bridges Imaging Interoperability Gaps

RadAISlice: Agentic AI is emerging as a solution for imaging workflow integration across platforms.

The details:

• Agentic AI acts as credentialed users to automate procedures

• Supports requisition intake, exam retrieval, reminders, and follow-up

• Combines LLMs, vision, and rules for robust, auditable performance

• Reduces manual touches and repeat scans, optimizes scheduling

Key takeaway: Agentic AI delivers true workflow automation and resilience where standards-based integration still fails, enabling next-level efficiency and safety for radiology teams.

🫁 AI Model Validated for Diverse Lung Cancer Risk Prediction

RadAISlice: Sybil AI was validated in a diverse, predominantly Black safety-net hospital population.

The details:

• Validated on over 2,000 CTs at UI Health (Chicago), 2014–2024

• 62% Non-Hispanic Black, 13% Hispanic, high social diversity

• AUC stayed strong through 6-year risk window, even in subgroups

• Consortium advancing to clinical trial for workflow integration

Key takeaway: Equitable screening AI validated in underserved populations signals critical progress against disparities in lung cancer outcomes.

🩺 AI Breast Imaging Yields High NPV but Raises Recalls

RadAISlice: AI-driven breast screening safely rules out negatives but triggers more recalls than radiologists.

The details:

• Transpara AI compared to 11 breast radiologists on >30,000 cases

• NPVs for both: 99.8–99.9%; AI sensitivity up to 94% at higher thresholds

• AI recall rate up to 41.8% vs <7.2% for radiologists

• Urgent need for strategies to limit AI-driven false-positives

Key takeaway: AI can streamline breast screening by ruling out negatives, but clinical gains depend on managing increased recall and false-positives.

HECKTOR2025 (2025-09-03)

Modality: PET/CT | Focus: Head and neck | Task: Segmentation, prognosis prediction

• Size: 1123 scans, 1123 patients

• Annotations: Manual tumor (primary and lymph node) segmentations, radiotherapy dose maps, clinical outcome and biomarker data

• Institutions: Mohamed bin Zayed University of AI, MD Anderson Cancer Center et al.

• Availability:

  public (https://hecktor25.grand-challenge.org/)

• Highlight: Largest publicly available head and neck PET/CT dataset with multimodal annotations and long-term outcome data from 10 centers

African Breast Imaging Dataset (ABID) (2025-08-28)

Modality: Mammogram, US | Focus: Breast | Task: Segmentation, Classification

• Size: ~400 US scans, ~200 mammograms; ~100 patients, longitudinal

• Annotations: BI-RADS scores, segmentation masks, breast composition, lesion characteristics, multi-radiologist labels

• Institutions: Ernest Cook University, MAI Lab et al.

• Availability:

  Public (planned), medRxiv link

• Highlight: First open, longitudinal African dataset with both POC and conventional breast imaging, expert labels.

CRL-2025 Atlas (2025-08-28)

Modality: MRI (T2w, DTI) | Focus: Fetal brain | Task: Segmentation, Parcellation

• Size: 194 MRI scans, 160 fetuses (21-37 weeks gestation)

• Annotations: Detailed tissue segmentation (36 labels), regional parcellation (126 labels), transient white matter compartments to 31 weeks

• Institutions: Boston Children’s Hospital, University of California Irvine, et al.

• Availability:

  Public (Harvard Dataverse link)

• Highlight: High-resolution 4D spatiotemporal fetal brain atlas with unique fine-grained segmentations and open-source tools.

🏛️ FDA Clearances

• K251983 - Brainomix 360 Triage Stroke: FDA-cleared AI triage tool for brain imaging stroke findings, enabling rapid clinical decisions.

• K251484 - CT:VQ by 4DMedical: FDA-cleared CT-based ventilation imaging tool for detailed lung function assessment.

• K252526 - Rapid DeltaFuse by iSchemaView: FDA-cleared image processing for efficient radiology decision support.

• K250947 - VistaSoft 4.0 and VisionX 4.0: FDA-cleared radiology automation software for improved diagnostic image analysis.

• K250788 - Definium Tempo Select by GE: Stationary X-ray system for diagnostic medical imaging.

• K251106 - Sonosite LX and PX: FDA-cleared ultrasound imaging systems for real-time, multi-organ clinical assessment.

• Explore last week's 8 radiology AI FDA approvals.

📄 Fresh Papers

• doi:10.1016/j.artmed.2025.103254 - LoRA-PT offers efficient fine-tuning of transformer UNETR, improving hippocampus segmentation in scarce data settings.

• doi:10.1007/s00259-025-07504-8 - AI-assisted quantification of PET/CT metabolic response predicts survival in metastatic uveal melanoma on tebentafusp therapy.

• doi:10.1038/s41551-025-01497-3 - MedSegX, trained on MedSegDB, achieves state-of-the-art, generalist, open-world medical segmentation—robustly handling OOD data.

• doi:10.1101/2024.10.17.24315675 - YOLOv8-based deep learning model detects early asymptomatic carotid plaques at population scale, improving risk stratification in CV disease.

• Browse 205 new radiology AI studies from last week.

📰 Everything else in Radiology AI last week

• A Finnish review finds most healthcare AI economic studies lack robust evaluation and clear cost reporting.

• Novel imaging and AI reveal principles to control atomic order in 2D MXene nanomaterials for emerging medical devices.

• DreamConnect AI reconstructs and allows language-guided edits of mental imagery from human fMRI scans.

• A literature review underlines inconsistent economic reporting for healthcare AI, risking unreliable investment and procurement decisions.

• Agentic AI automates multi-system radiology tasks where standards fall short, easing manual workflow bottlenecks.

• Read 8 fresh radiology AI news articles for last week.

That's it for today!

👋 We'd love your thoughts. Just hit Reply and tell us what you want to see in the next Slice!

See you next week.