RadAI Slice Newsletter Weekly Updates in Radiology AI |
Good morning, there. A motion-robust MRI method achieved 180μm whole-brain imaging in 17 minutes. I am impressed by a multicenter team that reduced high-res neuro MRI scan time from 34 to 17 minutes. This supports broader use of mesoscale neuroimaging and minimizes motion artifacts. The technique could make advanced brain MR more accessible for research and clinics alike. How would faster, motion-resilient MRI impact your neuroimaging workflow?
Here's what you need to know about Radiology AI last week: Mesoscale MRI Doubles Speed for Motion-Robust Brain Imaging Calls Grow for Stronger FDA Evidence Standards in Radiology AI First FDA-Cleared AI for Mitral Annular Calcification on Routine CT Commercial AI for Fracture Detection: Real-World Prospective Study Plus: 4 newly released datasets, 6 FDA approved devices & 4 new papers.
|
🧾 NEW WEBSITE FEATURE You can now filter our FDA device database by regulatory pathway, review panel, and product code. Explore FDA AI devices → |
🧠 Mesoscale MRI Doubles Speed for Motion-Robust Brain Imaging RadAI Slice: A new neural network-based MRI method reduces scan time for high-res whole-brain imaging. The details: Delivers 180μm isotropic T2 MRI in just 17 minutes at 7T Yields 22% lower reconstruction error than standard approaches Validated in both ex vivo and in vivo datasets with real motion Enables sharper, faster imaging for advanced neuro research
Key takeaway: Advanced neural-MRI workflows could become more efficient and motion-tolerant, bringing high-res imaging into routine research and eventually clinical settings. |
💡 Calls Grow for Stronger FDA Evidence Standards in Radiology AI RadAI Slice: A Dana-Farber review urges the FDA to adopt stricter, more transparent AI validation rules. The details: Retrospective-only validation is routine for most AI tools now Experts seek more real-world, reader, and prospective studies Synthetic data use in FDA clearance is questioned in some cases Public validation database and reporting checklists recommended
Key takeaway: Clearer FDA standards could help drive trustworthy, clinically relevant AI—raising the quality bar for validated radiology tools and increasing patient safety. |
🫀 First FDA-Cleared AI for Mitral Annular Calcification on Routine CT  Image from: Cardiovascular Business RadAI Slice: Mitral annular calcification (MAC) AI is now FDA-cleared for non-gated routine CT scans. The details: Works across a broad range of routine CT protocols First FDA-approved tool specific to MAC detection Development included data from over 25 major centers May boost detection of at-risk cardiovascular patients
Key takeaway: Routine CT scans can now consistently screen for MAC, enabling improved risk assessment during standard imaging and supporting heart disease management. |
🦴 Commercial AI for Fracture Detection: Real-World Prospective Study RadAI Slice: A large prospective study benchmarked three x-ray AI tools across 1,037 adults and 22 regions. The details: Rayvolve AUC: 84.9%; BoneView: 84%; RBFracture: 77.2% Highest sensitivity: Rayvolve at 79.5% Multiple fractures remain a challenging scenario for all tools AI shown to work best as an adjunct, not a replacement
Key takeaway: Radiology AI tools show real utility—but limitations for complex or multiple injuries reinforce the need for expert oversight and continued multicenter validation. |
RadGenome-Chest CT (2025) Modality: CT | Focus: chest, lung | Task: report generation, VQA Size: 25,692 CT volumes from 21,304 patients Annotations: 197-region segmentation masks, region-grounded report sentences, 1.2M region-linked VQA pairs Institutions: Shanghai Jiao Tong University, Shanghai AI Laboratory et al. Availability: Highlight: First large-scale, region-grounded vision-language dataset for 3D chest CT; supports multimodal foundation model benchmarking
|
BIND (2025-10-02) Modality: MRI, CT | Focus: Brain, neurophysiology | Task: Classification, segmentation Size: 1.8M scans from 38,945 patients Annotations: Structured pathology labels from clinical reports (LLM-extracted); session-level reports; some sequence segmentations Institutions: Massachusetts General Hospital, Stanford University, et al. Availability: Highlight: Largest multimodal brain imaging and neurophysiology dataset; integrated with EEG/PSG; LLM-extracted structured clinical findings
|
BreastDCEDL AMBL (2025-10-04) Modality: MRI | Focus: Breast | Task: Classification, Segmentation Size: 133 annotated lesions (89 benign, 44 malignant) from 88 patients Annotations: Manual segmentations for benign and malignant lesions, lesion-level masks in 3D NIfTI format Institutions: Ariel University, The Cancer Imaging Archive Availability: Highlight: First public DCE-MRI dataset with both benign and malignant breast lesion annotations for reproducible AI benchmarking
|
SpineMed-450k (2025-10-03) Modality: CT, MRI, X-ray | Focus: Spine, Orthopedics | Task: QA, report generation Size: ~33,000 images, 834 hospital cases, 463k QA pairs, ~1,000 patients Annotations: Multiple-choice/open-ended QA, reports, dialogue, precise vertebral-level reasoning Institutions: Jilin University, Nanjing University, et al. Availability: Request-only (project page or contact authors, full dataset not directly public as of paper date)
Highlight: Largest multimodal, level-aware, traceable spine AI dataset; integrates real hospital cases and clinician-in-the-loop curation.
|
🏛️ FDA Clearances K250662 - Bunkerhill MAC (BunkerHill Health): FDA-cleared tool for mitral annular calcification on routine CT, aiding broad-scale cardiovascular risk detection. K250766 - LungQ 4 (Thirona BV): CT-based product supports lung analysis with AI-driven insights for diagnostic workflows. K251215 - Philips IntelliSpace Cardiovascular: Multimodal cardiovascular image suite for clinical diagnosis and treatment planning. K251859 - Olympus EVIS EUS Ultrasound GI Videoscope: Advanced endoscopic ultrasound for gastrointestinal assessment. K250039 - HPACS (HealthHub): Radiology imaging platform enhances and manages images for workflow efficiency. K250026 - Shear Wave Quantificational Ultrasound (Wuxi Hisky): Provides quant tissue stiffness from ultrasound, aiding diagnosis and treatment guidance.
|
📄 Fresh Papers doi:10.1183/13993003.00933-2025 - A multinational, prospective ARDS cohort identifies three distinct CT proteomic phenotypes with different treatment responses and outcomes. doi:10.1016/j.jacr.2025.09.032 - A practical four-phase framework is proposed for radiology AI deployment: validation, implementation, value assessment, and post-market surveillance. doi:10.1016/j.acra.2025.09.018 - Prospective dual-center study shows deep learning-based ultra-low-dose CTPA, aided by AI, sustains high PE diagnostic accuracy with 74% dose reduction. doi:10.3174/ajnr.A8977 - Motion-informed deep learning improves 3D brain MRI image quality and morphometric accuracy in cognitively impaired patients. Browse 200 new radiology AI studies from last week.
|
That's it for today! Before you go we’d love to know what you thought of today's newsletter to help us improve the RadAI Slice experience for you. |
|
👋 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. I’m building a reader-curated list of radiology AI tools. which one would you nominate? |
|
