AI model improves early detection of Alzheimer's in diverse groups

RadAI Slice Newsletter

Weekly Updates in Radiology AI

Good morning, there. An AI tool increased sensitivity for undiagnosed Alzheimer's to 81%, nearly doubling traditional models.

I noticed this study stands out for both its clinical scope and its focus on diagnostic equity. This kind of externally validated, EHR-based model maps directly onto workflows in large health systems. Widening the detection net could improve care as new therapeutics emerge and as radiologists and neurologists grapple with underdiagnosis, especially in underrepresented populations.

How would you incorporate an AI like this into your current diagnostic process?

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

• AI detects undiagnosed Alzheimer's across diverse populations

• ESR issues consensus guidance on EU post-market AI device surveillance

• AI-enhanced CT predicts prognosis in large real-world CAD cohort

• Multimodal AI improves breast cancer recurrence prediction post-NCT

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

🧠 AI detects undiagnosed Alzheimer's across diverse populations

RadAI Slice: A UCLA-led team validated an AI tool that boosts Alzheimer's case finding via EHR data.

The details:

• Tested in 97,000+ patients from multiple ethnic backgrounds

• Achieved 77–81% sensitivity vs 39–53% for baseline models

• Validation incorporated genetic markers (APOE ε4)

• Targets equity: addressed Hispanic/Latino and African American disparities

Key takeaway: This approach could shift earlier AD detection at scale, potentially addressing critical gaps in underdiagnosed and marginalized populations, particularly as radiology roles expand in dementia care.

📄 ESR issues consensus guidance on EU post-market AI device surveillance

RadAI Slice: The European Society of Radiology clarifies post-market responsibilities for imaging AI devices.

The details:

• Delphi process with 14 panelists and expert survey

• Covers MDR and new EU AI Act ambiguities for imaging AI

• Stresses radiologist participation in device safety monitoring

• Provides clear PMS/PMCF best practices for clinical deployers

Key takeaway: Radiologists must understand and contribute to post-market monitoring, benefiting safe AI adoption and clarifying ongoing legal/clinical oversight in European imaging settings.

🩺 AI-enhanced CT predicts prognosis in large real-world CAD cohort

RadAI Slice: The FISH&CHIPS study links FFR-CT AI analysis with future cardiac events at population scale.

The details:

• 7,836 patients in the UK, initial pool 90,553

• Severe FFR-CT: 4x MI, 3x mortality risk vs normal

• FFR-CT AI added prognostic value to standard scores

• Independent UK MRC funding; analysis by HeartFlow, not sponsor

Key takeaway: These findings support AI-based physiologic CT tools to improve cardiac risk stratification and guide tailored management in stable CAD, reinforcing CT's evolving role in noninvasive cardiology.

🦠 Multimodal AI improves breast cancer recurrence prediction post-NCT

Image from: EurekAlert

RadAI Slice: AI combined slides, gene data, and clinical risk for better breast cancer recurrence forecasting.

The details:

• 4,000+ TAILORx trial cases

• Outperformed standard recurrence tools (esp. late events)

• Potentially scalable using routine histology + EHR

• Results presented at major US oncologic congress

Key takeaway: This work illustrates real progress in personalized breast cancer care, with multimodal AI opening new doors for risk-adaptive follow-up and therapy in radiology-pathology teams.

BONBID-HIE (2023-07-03)

Modality: MRI | Focus: Neonatal brain | Task: Lesion segmentation, detection

• Size: 133 scans from 133 patients

• Annotations: Manual 3D lesion masks, expert-reviewed

• Institutions: Massachusetts General Hospital, Boston Children's Hospital

• Availability:

  Public (https://hiechallenge.github.io/)

• Highlight: First public dataset with expert-annotated HIE lesions in neonatal diffusion MRI; includes Z-score ADC maps.

CMRxRecon2024 (2024-04)

Modality: MRI | Focus: Cardiac, Vascular | Task: Reconstruction, Generalization

• Size: 330 scans, 330 patients, 200,000+ k-space slices

• Annotations: Image reconstructions from raw k-space data; reference GRAPPA images

• Institutions: Fudan University, Imperial College London et al.

• Availability:

  public (link)

• Highlight: Largest public multi-modality CMR raw dataset with unseen modalities and diverse sampling for universal AI reconstruction.

🏛️ FDA Clearances

• K252366 - a2z-Unified-Triage earned FDA clearance for AI triage of 7 abdominal/pelvic CT findings, showing >80% sensitivity and specificity for each.

• K250694 - Scaida BrainCT-ICH cleared for CT head triage, reaching 87% sensitivity and 89% specificity for ICH detection in multicenter validation.

• K253269 - OEC One CFD mobile C-arm added DL-based trajectory guidance for wire placement, user success rated above 97% in 3,000+ test images.

• K252217 - CT VScore+ (Canon Medical) approved for non-contrast cardiac CT calcium scoring, achieving ICC >0.99 across image vendors and populations.

• Explore last week's 13 radiology AI FDA approvals.

📄 Fresh Papers

• doi:10.1038/s41698-025-01215-x - ADNEX-AI automates ultrasound features for ovarian mass triage, attaining AUC 0.93 in multicenter validation and reducing inter-center variability.

• doi:10.1109/JBHI.2025.3643125 - A knowledge-guided multimodal network integrates ultrasound and clinical data for noninvasive HER2 subtyping in breast cancer.

• doi:10.1161/JAHA.125.043221 - Multicenter chest CT pulmonary vascular radiomics predicts both the diagnosis and 2-year outcomes in PH, outperforming standard risk tools.

• doi:10.2196/80981 - A scoping review charts 67 LLM studies in radiology, finding strong report summarization performance but large clinical validation gaps.

• Browse 166 new radiology AI studies from last week.

📰 Everything else in Radiology AI last week

• RSNA panel debated chest x-ray AI autonomy; cited current hallucination rates, workflow impact, and regulatory needs.

• A White House executive order and HHS AI strategy aim to unify US healthcare AI policy and spur clinical innovation.

• An Irish study finds 67% of patients prefer radiologist-led, AI-assisted reads and want written consent for AI use.

• A large health system allegedly violated consent when using an AI tool to transcribe patient visits, raising privacy and regulatory questions.

• A survey found over half of radiology practices cite faster report turnaround and workflow as the top ROI for imaging AI.

• Read 15 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.