RadAI Slice Newsletter Weekly Updates in Radiology AI |
Good morning, there. A new multimodal machine-learning model predicts cerebral amyloid burden in Alzheimer’s with R² up to 0.64. I found the integration of plasma biomarkers, MRI, and genetic profiles meaningfully boosts Alzheimer’s risk assessment beyond plasma markers alone. This approach improves noninvasive estimation of cerebral Aβ, potentially offering a PET alternative for early AD risk. The result feels important as radiologists seek multimodal, scalable tools for dementia stratification and trial enrichment. Would you incorporate multimodal biomarkers for earlier Alzheimer’s detection in practice?
Here's what you need to know about Radiology AI last week: Multimodal AI enhances Alzheimer’s risk prediction non-invasively AI boosts ED clinicians’ CT head reads to radiologist level AI predicts osteoporosis from lumbar MRI with ViT model AI risk triage model shortens time to cancer diagnosis in real world Plus: 3 newly released datasets, 3 FDA approved devices & 3 new papers.
|
📚 NEW WEBSITE FEATURE You can now save any paper or FDA device and access them later in your personal library. Visit My Library → |
🧬 Multimodal AI enhances Alzheimer’s risk prediction non-invasively RadAI Slice: A multicenter study shows combining MRI, plasma, and genetics improves Alzheimer’s amyloid burden prediction. The details: 150 ADNI and 101 SILCODE participants, global-R² improved from 0.56 to 0.64 with multimodal model Classifier AUC: 0.87 (controls), 0.76 (MCI), 0.95 (AD dementia) External validation across Chinese and U.S. cohorts—consistent results Outperforms plasma-only approaches for PET amyloid status prediction
Key takeaway: Multimodal imaging and biomarkers deliver greater noninvasive accuracy for Alzheimer’s risk, enabling earlier identification and broadening clinical and research relevance. |
🩺 AI boosts ED clinicians’ CT head reads to radiologist level RadAI Slice: A multicenter NHS study found AI assistance increased non-radiologist accuracy on acute CT head scans. The details: 30 general radiologists, emergency doctors, and radiographers, 150 CT heads reviewed in reader study AI assistance increased critical abnormality sensitivity: 82.8%→89.7% (p<0.001) Intracranial hemorrhage sensitivity: 84.6%→91.6% with AI support AI helped ED clinicians match unaided radiologist-level sensitivity
Key takeaway: AI-assisted CT head interpretation meaningfully broadens safe, expedited acute neuroimaging triage to non-radiologist clinicians—impacting workforce, workflow, and patient care. |
🦴 AI predicts osteoporosis from lumbar MRI with ViT model RadAI Slice: A 1,095-patient dual-center study tested MRI+ViT for osteoporosis prediction versus conventional methods. The details: MRI-based ViT model AUC: 0.85 (internal), 0.75 (external) Combined model (clinical, radiomics, ViT): AUC 0.86 (internal), 0.81 (external) ViT features added predictive value over radiomics alone Demonstrated robust performance on multicenter validation
Key takeaway: Automated MRI-based ViT offers a scalable, accurate alternative to DXA, enabling earlier osteoporosis diagnosis directly from lumbar MRI. |
💡 AI risk triage model shortens time to cancer diagnosis in real world RadAI Slice: A prospective, real-world trial used AI to flag high-risk women for same-day breast imaging assessment. The details: 4,145 mammograms, Mirai AI flagged 12.7% as top 10% risk Consent rate 94% for instant reads; 13 biopsies, 6 cancers found Cancer detection rate 60/1,000 in high-risk group; 2.3/1,000 in standard group Median time to result/diagnostic/biopsy cut by 87–99%
Key takeaway: Embedding AI risk models into clinical workflow demonstrably accelerates breast cancer diagnosis and biopsy for high-risk patients—proving real-world, not just theoretical, workflow impact. |
MRE Phantom, Liver, and Brain Dataset (MRE-PLB) (2025) Modality: MRI | Focus: Liver, Brain | Task: Biomechanical parameter inversion, Modulus mapping Size: 1 phantom scan, 1 liver scan, 1 brain scan (all from different subjects) Annotations: Wave images, anatomical images, complex shear modulus maps (storage and loss modulus); all with TWENN algorithm estimates Institutions: Shanghai Jiao Tong University, Ruijin Hospital et al. Availability: Highlight: Includes phantom, liver, and brain MRE data with state-of-the-art neural network-based inversion algorithm (TWENN)
|
CARDIUM (2025-10-20) Modality: Ultrasound, Echo | Focus: Heart (fetal), Maternal health | Task: Classification, Multimodal fusion Size: 6,558 images; 1,103 patients Annotations: CHD diagnosis; maternal clinical variables; 11+ CHD types labeled Institutions: Universidad de los Andes, Fundacion Santa Fe de Bogotá Availability: Highlight: First public multimodal dataset with paired fetal cardiac ultrasound/echo and detailed maternal clinical data for prenatal CHD.
|
Mammo-Bench (2025-10-14) Modality: X-ray | Focus: Breast | Task: Classification, Segmentation Size: 19,731 images from ~6,500 patients Annotations: Disease status, BI-RADS, breast density, abnormality type, molecular subtype, ROI masks Institutions: IIIT Hyderabad, et al. Availability: Highlight: Unified large benchmark from six public datasets, diverse with comprehensive labels and preprocessing.
|
🏛️ FDA Clearances K251355 - X1-FFR by SpectraWAVE: AI converts x-ray angiography to coronary flow simulation, aiding noninvasive CAD assessment. K251805 - syngo.CT Dual Energy by Siemens: Cleared for dual energy CT in routine and advanced tissue characterization workflows. K252199 - AGFA HealthCare Enterprise Imaging: FDA-cleared radiology image management and enhancement platform. Explore last week's 5 radiology AI FDA approvals.
|
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. 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. |
|
|
