AI speeds shoulder ultrasound exams for trainees

Good morning, there. AI guidance cut exam time by 34% for junior residents in shoulder ultrasound.

I found the external validation with >8,000 images especially compelling for daily MSK ultrasound practice. Reducing scan duration and standardizing technique could meaningfully impact training, tele-ultrasound, and clinical consistency. This feels like a genuine advance for regions short on expert sonographers or seeking to standardize remote acquisitions.

How would real-time AI guidance alter your MSK ultrasound workflow?

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

• AI Shortens and Standardizes Shoulder Ultrasound Exams

• Radiomics MRI Model Emerges for Lymph Node Risk in Rectal Cancer

• 🩻 Real-World AI Reduces Mammography Workload and Boosts Detection

• Deepfake X-Rays Challenge Radiologists and AI Models

• Plus: 6 FDA approved devices & 4 new papers.

🦾 AI Shortens and Standardizes Shoulder Ultrasound Exams

RadAISlice: AI-driven shoulder ultrasound guidance improves standardization and reduces operator time.

The details:

• 13,000+ exams for training, 8,458 images external validation

• Mean accuracy 94%, F1 scores 0.87–0.99 across structures

• AI cut exam time for residents by 34% (p=0.014)

• Expert review confirmed guidance accuracy

Key takeaway: AI-powered guidance holds real promise to standardize MSK ultrasound, speed training, and may enable effective remote or patient-led exams.

🧠 Radiomics MRI Model Emerges for Lymph Node Risk in Rectal Cancer

RadAISlice: A multicenter MRI-based radiomics model predicted lymphatic spread in rectal cancer patients.

The details:

• 151 patients, prospective design; external validation included

• AUCs 0.90–0.94 for LNM and LVI across train/test cohorts

• Combines DKI habitat radiomics with clinical biomarkers

• SHAP analysis highlighted main predictive features

Key takeaway: Robust, explainable ML models may soon help inform surgical planning or adjuvant therapy discussions for rectal cancer.

🩻 Real-World AI Reduces Mammography Workload and Boosts Detection

Image from: Radiology Business

RadAISlice: A population-based Spanish trial demonstrates real-world safety for autonomous mammography AI.

The details:

• 30,000+ women, prospective design with routine program

• AI safely excluded normal/low-risk studies from radiologist review

• Improved cancer detection rates in workflow

• Findings published in Nature Medicine

Key takeaway: This evidence supports not just AI's technical performance but its operational viability in population screening workflows.

🤖 Deepfake X-Rays Challenge Radiologists and AI Models

Image from: EurekAlert

RadAISlice: Radiologists and top LLMs misidentify a substantial portion of deepfake X-rays.

The details:

• 264 images, 17 radiologists, multicountry design

• Radiologist accuracy ranged from 58% to 92%

• AI models scored 52%–89% detection accuracy

• No link between clinical experience and detection rate

Key takeaway: Synthetic X-rays now rival the realism of true images, raising novel security and integrity threats for clinical and legal workflows.

🏛️ FDA Clearances

• K252261 - InferCare RECIST by Infervision Healthcare: AI-enabled lesion segmentation for CT monitoring and oncology response assessment.

• K254277 - Embrace Neonatal MRI System cleared for detailed imaging in neonates, supporting diagnosis in critical care settings.

• K252249 - Extremity CT Imaging System (Mars Bioimaging): specialized 3D ortho CT for arms/legs, supporting advanced MSK diagnosis.

• K253413 - LiverMultiScan (v6.0; Perspectum): quantitative liver MRI tool, provides advanced tissue characterization for hepatology and radiology.

• K253185 - Lux HD Detectors/DRX-Revolution Mobile X-Ray System (Carestream): digital, mobile X-ray supporting efficient point-of-care imaging.

• K252353 - myray ProXIma X6 (Cefla): dental CT system with 3D reconstruction for detailed jaw/tooth imaging.

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

📄 Fresh Papers

• doi:10.1038/s41598-026-42214-7 - An ensemble AI approach using multi-institutional MRI data improved detection of clinically significant prostate cancer over baseline models.

• doi:10.1016/j.ejrad.2026.112799 - Deep learning reconstruction halved effective dose in pediatric chest CT with lower noise and no loss in diagnostic quality.

• doi:10.1038/s41551-026-01637-3 - A large-scale, contrastive learning-based vision model for cardiac MRI shows strong performance in regression and multi-disease classification.

• doi:10.1186/s13244-026-02250-3 - A multicenter, validated CT-based radiomics nomogram (PNET-PRISM) outperformed clinical models for preoperative grading of pancreatic neuroendocrine tumors.

• Browse 124 new radiology AI studies from last week.

📰 Everything else in Radiology AI last week

• A JMIR policy analysis examined discordance between AI legal requirements for transparency and real-world patient comprehension in medical imaging.

• ICLR 2026 work shows that neural net privacy improvements usually lower model performance; a new method optimizes both for imaging data.

• Researchers validated an interpretable mortality-risk model for ICU sepsis, using SHAP for transparency and multicenter data.

• A deep LASSO model, trained on biopsy and external validated imaging data, stratifies MASLD into four risk subtypes for tailored intervention.

• BiliSeq, a molecular test, doubled the detection of bile duct cancer in a multicenter study when combined with pathology, benefiting imaging-diagnosis clarity.

• Read 11 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.