AI ultrasound breakthrough predicts delivery timing for expecting mothers

RadAISlice Newsletter

Weekly Updates in Radiology AI

Good morning, there. AI can now predict delivery timing using only standard ultrasound images.

Accurately estimating time to delivery supports clinical decision-making and resource allocation in obstetrics. This innovation may improve pregnancy outcomes, especially in low-resource settings, and marks a milestone for imaging AI in women's health.

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

• AI predicts delivery timing from routine ultrasound images

• Majority of parents trust AI over doctors in pediatric imaging

• AI predicts Alzheimer's protein markers from brain scans

• ChatGPT-4 Turbo monitors post-deployment radiology AI

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

🤰 AI predicts delivery timing from routine ultrasound images

Image from: Health Imaging

RadAISlice: New deep learning models can now forecast pregnant patients’ delivery dates using only ultrasound images.

The details:

• Trained on 2+ million images from thousands of pregnancies

• Achieved R² of 0.95 for term births and 0.92 for all births

• Works across all trimesters and diverse patient groups

• Improved preterm birth prediction with continuous retraining

• Does not require clinical history or other risk factors

Key takeaway: Estimating days until delivery with routine ultrasound empowers clinicians, sharpens risk management, and advances equitable prenatal care through scalable, noninvasive AI.

🤖 Majority of parents trust AI over doctors in pediatric imaging

Image from: Health Imaging

RadAISlice: A new survey shows parents increasingly trust AI to spot abnormalities in pediatric imaging—sometimes even more than physicians.

The details:

• Parents and caregivers surveyed on AI in kids’ healthcare

• Most said AI could be more accurate than doctors for imaging

• Highlights demand for trusted, transparent pediatric AI tools

• Published in the journal Insights Into Imaging

Key takeaway: Parental acceptance will be key to pediatric AI’s integration, potentially speeding adoption and reinforcing AI’s clinical value in family-centered care.

🧠 AI predicts Alzheimer's protein markers from brain scans

RadAISlice: A Boston University AI tool predicts amyloid and tau—hallmarks of Alzheimer's—using imaging, genetics, and clinical data.

The details:

• Model integrates brain scans, cognitive tests, genetics, health history

• Tested on 12,000+ participants from international cohorts

• Performs robust location-specific protein prediction

• May enable earlier, accessible diagnosis and clinical trial enrollment

Key takeaway: AI-enabled, non-invasive Alzheimer's prediction could revolutionize screening and disease staging, accelerating research and personalized care globally.

⚡ ChatGPT-4 Turbo monitors post-deployment radiology AI

RadAISlice: Researchers validated ChatGPT-4 as a scalable tool for monitoring AI models in real-world radiology practice.

The details:

• LLM checked 332,809 head CTs from 37 sites for aidoc ICH AI accuracy

• ChatGPT-4 reached 0.995 accuracy and high PPV/NPV vs radiologist ground truth

• Aidoc's errors more likely than LLM’s extraction issues

• LLM monitoring is cost-effective and reduces manual workload

Key takeaway: LLM-powered monitoring may become essential for ongoing AI QA and safety as radiology AI deployment scales in busy clinics.

🏛️ FDA Clearances

• K250246 - uMR Jupiter MRI system (Shanghai United Imaging) garners FDA clearance for high-quality MR imaging across clinical applications.

• K244016 - Claritas HealthTech’s iPETcertum radiology image-processing platform receives FDA 510(k) for assisting diagnosis.

• K252239 - IMRIS Imaging’s InVision™ 3T Recharge MRI suite receives FDA clearance to boost intraoperative imaging.

• K251038 - Shen Zhen Cambridge-hit’s RiasDR digital radiograph software receives FDA 510(k) for imaging enhancement and workflow.

• K251629 - Medicrea/Medtronic’s UNiD™ Spine Analyzer gets FDA 510(k) to automate spinal segmentation for treatment planning.

• K250664 - UC-CARE’s Navigo Workstation 2.3, an advanced radiology image analysis suite, now FDA cleared.

📄 Fresh Papers

• doi:10.1038/s41598-025-15007-7 - A new vision transformer network achieves 96%+ accuracy for automated Alzheimer’s prediction on augmented brain MRI.

• doi:10.1101/2025.08.08.25333333 - A multimodal deep learning framework combining CXR, ventilator data, and EHR boosts ARDS detection AUROC to 0.86.

• doi:10.1007/s12194-025-00946-7 - Deep learning improves automatic segmentation of prostate CBCT images using treatment planning CT data and domain adaptation.

• doi:10.1186/s13550-025-01303-w - A nnU-Net deep learning approach accurately segments brown adipose tissue on PET/CT, aiding cancer care and metabolism research.

• Browse 237 new radiology AI studies from last week.

📰 Everything else in Radiology AI last week

• Researchers propose using ChatGPT to automatically check for AI drift in radiology models, improving safety.

• Routine use of AI in colonoscopies may cause experienced endoscopists to lose detection skill in non-AI cases.

• RSNA AI Challenge entrants reached high recall and specificity for breast cancer screening in global test sets.

• Heartflow, the cardiac imaging AI firm, raised $364M in a public offering, far exceeding its target.

• United Healthcare will reimburse Elucid’s coronary plaque quantification AI, potentially driving broader adoption in cardiac CT.

• Read 13 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.