A multimodal deep learning model called MPANet significantly outperforms radiologists at classifying renal tumors from multiphase CT imaging, achieving 73.3% accuracy compared to 43.6-62.4% for specialist assessments. Original paper: Multimodal deep learning model for multiclass classification of renal tumors. — NPJ digital medicine. 10.1038/s41746-026-02697-0 📄 Read the article

As healthcare organizations delete patient records, their clinical AI models face an ethical dilemma: maintaining privacy protections while ensuring fair predictions across all demographic groups. Original paper: Mitigating algorithmic unfairness arising from forgetfulness of medical records in clinical artificial intelligence. — Nature communications. 10.1038/s41467-026-94356-9 📄 Read the article

An AI quality control system eliminated blind spots entirely in magnetically controlled capsule gastroscopy while improving lesion detection rates. Original paper: Impact of a real-time automatic quality control system for magnetically controlled capsule gastroscopy: a multicenter randomized controlled trial. — BMC medicine. 10.1186/s12916-026-04901-0 📄 Read the article

Researchers demonstrate that deep learning combined with generative modeling can detect head and neck cancer from saliva samples, overcoming data limitations for rare disease diagnosis. Original paper: Leveraging population-scale proteomic data with deep learning for head and neck cancer detection in saliva. — NPJ digital medicine. 10.1038/s41746-026-02658-7 📄 Read the article

Researchers developed an end-to-end framework that generates synthetic health datasets with formal privacy guarantees, enabling institutions to securely share sensitive data for research. Original paper: Anonymization and visualization of health data and biomarkers. — NPJ digital medicine. 10.1038/s41746-026-02662-x 📄 Read the article

Implantable sensors combined with machine learning can detect plant stress from internal biomarkers up to 48 hours before visible symptoms appear. Original paper: Machine learning-enabled implantable plant biomarker sensor for early detection and classification of acid and salt stress. — Nature communications. 10.1038/s41467-026-72344-5 📄 Read the article

Researchers have developed a wearable microneedle biosensor with integrated edge computing that monitors glucose, ketones, and pH in real-time while eliminating signal noise and cloud dependency. Original paper: Microinterventional in-sensor computing system for real-time metabolic health assessment. — Nature communications. 10.1038/s41467-026-72520-7 📄 Read the article

OBSCORE, a machine learning model using 20 clinical features, successfully identifies individuals at highest risk for obesity-related complications and outperforms BMI-based approaches. Original paper: Data-driven prioritization of high-risk individuals for weight loss interventions. — Nature medicine. 10.1038/s41591-026-01234 📄 Read the article

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