Every story in this issue with a short summary, so you can decide what to read before leaving the site. Summaries are written by AI (Claude) only from each publisher's own description and checked by Jev for anything not in that description. They are not full translations. Follow "Read the original" for the full story.
A scholarly exchange in Pediatric Radiology concerns a deep learning model for detecting neonatal pneumothorax. It highlights the gap between overall algorithmic accuracy and clinically meaningful performance.
Read the original →A study of 202 patients shows automated CT body composition measurements remain highly consistent across filtered back projection, iterative, and deep learning reconstruction.
Read the original →A new knowledge distillation framework compresses three large neural networks into a 6.97 MB model. The model classifies rice diseases, localizes infections, and quantifies severity on a budget smartphone.
Read the original →A new study finds that AI systems and large language models achieve at best 25 percent accuracy in diagnosing psoriasis in skin of color. Several presentations were unrecognized by any model tested.
Read the original →AI tools may help clinicians recognize sepsis earlier. False alarms, limited external validation, and uncertain effects on patient outcomes complicate their use in adult and neonatal intensive care.
Read the original →In the latest episode of her Breast Imaging in Focus series, Manisha Bahl, MD, MPH, FSBI, discusses new research. The research compares mammography-based AI models for breast cancer detection and short-term risk prediction.
Read the original →An XGBoost artificial intelligence model trained on Swedish neurointensive care data predicted dangerous intracranial pressure surges in traumatic brain injury patients. The prediction was successful across multiple centres.
Read the original →Technology that predicts when someone might experience intimate partner violence can undermine a survivor’s control over whether, when and how they share their abuse.
Read the original →Nonprofit hospitals are keeping pace with AI investments at PE-backed hospitals. Independently owned, for-profit facilities trail significantly.
Read the original →Mary-Brenda Akoda plans to launch GenMRI in Nigeria and the UK in 2027. The software uses AI to speed up MRI scans and reduce diagnostic delays.
Read the original →The tie-up marks the latest pharma company to enlist the AI-powered drug discovery services of Generate.
Read the original →The MHRA has published its 2030 Strategy. It outlines plans to modernise medicines and medical device regulation.
Read the original →Artificial intelligence could help shift healthcare systems from treatment-focused approaches toward prevention by analyzing medical data and identifying people at risk of developing diseases at an early stage. Realizing this potential requires appropriate healthcare funding models, rigorous safety assessments, and effective regulatory oversight.
Read the original →A new safety audit shows that machine unlearning in medical AI can silently degrade performance on rare diagnoses. It proposes a framework to detect and mitigate this spillover.
Read the original →The description contrasts past medicine, in which government bureaucracy gave doctors 18 minutes to solve every problem and conversations were interrupted, with what AI may offer. It suggests AI may allow patients uninterrupted conversation.
Read the original →Gunma University Hospital announced on the 13th that it has started operating a generative AI system to support medical document creation. The system helps find, organize and summarize needed information from electronic medical records. It aims to reduce the burden on doctors of writing documents.
Read the original →