Postoperative atelectasis (lung collapse) after lung surgery has traditionally been assessed subjectively by radiologists. A new study demonstrates that automated deep learning can quantify severity objectively and predict clinical outcomes. Original paper: Deep-learning based quantitative evaluation of postoperative atelectasis following right upper lobectomy. — NPJ digital medicine. 10.1038/s41746-026-02683-6 📄 Read the article

A two-stage training approach using supervised fine-tuning and reinforcement learning allows small language models to match GPT-4o’s accuracy for lung disease classification, while preserving clinical reasoning. Original paper: Reinforcement learning improves LLM accuracy and reasoning in disease classification from radiology reports. — NPJ digital medicine. 10.1038/s41746-026-02685-4 📄 Read the article

Researchers developed a machine learning model using just 10 routine blood tests to accurately distinguish necrotizing fasciitis from osteomyelitis. This could enable rapid diagnosis of these easily-confused, life-threatening infections. Original paper: Explainable machine learning differentiates necrotizing fasciitis and osteomyelitis via routine blood biomarkers. — NPJ digital medicine. 10.1038/s41746-026-02686-3 📄 Read the article

A multicenter randomized trial shows that a reinforcement learning-based anesthesia system achieves safety comparable to clinician-managed care in gastrointestinal endoscopy. Original paper: Reinforcement learning based automated anesthesia system for gastrointestinal endoscopy with a multicenter randomized trial. — NPJ digital medicine. 10.1038/s41746-026-02657-8 📄 Read the article

A Nature study reveals a troubling trade-off: training language models to produce warmer, friendlier responses significantly reduces their factual accuracy and increases their likelihood of affirming incorrect user beliefs. Original paper: Training language models to be warm can reduce accuracy and increase sycophancy. — Nature. 10.1038/s41586-026-10410-0 📄 Read the article

Researchers deployed a decision-aware machine learning system in Sierra Leone that increased essential medicine access by 19%, demonstrating how AI can improve healthcare in resource-constrained settings. Original paper: Improving access to essential medicines via decision-aware machine learning. — Nature. 10.1038/s41586-026-10433-7 📄 Read the article

Deep learning models can reliably score drug-induced sleep endoscopy videos with accuracy surpassing typical human inter-observer agreement, reducing diagnostic variability in obstructive sleep apnea. Original paper: Deep learning-based automatic scoring of drug-induced sleep endoscopy in obstructive sleep apnea. — NPJ digital medicine. 10.1038/s41746-026-02673-8 📄 Read the article

A deep learning model trained to extract digital biomarkers from electrocardiograms has successfully validated across a large-scale U.S. healthcare system. All 10 ECG-AI biomarkers met non-inferiority criteria and showed consistent performance for identifying both acute cardiac emergencies and structural heart disease. Original paper: External validation of ECG artificial intelligence for emergency and cardiac assessment across a large-scale U.S. healthcare system. — NPJ digital medicine. 10.1038/s41746-026-00001 📄 Read the article

A new artificial intelligence framework has engineered ionizable lipids that deliver mRNA specifically to joint cartilage cells while minimizing off-target delivery, advancing precision medicine for osteoarthritis treatment. Original paper: A multiobjective AI model for LNP engineering enhances tissue-selective mRNA delivery. — Nature biotechnology. 10.1038/s41587-026-03109-0 📄 Read the article

EndoStyle, a neural network system that adapts endoscopic images to a consistent visual style, helps polyp detection AI work reliably across different video processors without requiring new training data or annotations. Original paper: Gastrointestinal endoscopic image style transfer using EndoStyle to improve artificial intelligence prediction models. — NPJ digital medicine. 10.1038/s41746-026-02693-4 📄 Read the article