A new deep learning approach predicts tuberculosis drug resistance directly from genetic sequences with diagnostic accuracy, and validates clinical relevance by identifying patients at risk of treatment failure. Original paper: Convolutional neural networks quantify antibiotic resistance in Mycobacterium tuberculosis with diagnostic grade accuracy and predict treatment response. — Nature communications. 10.1038/s41467-026-72225-x 📄 Read the article

The format of AI explanations fundamentally shapes physician decision-making. A randomized controlled trial demonstrates that chain-of-thought reasoning from GPT-4 improves radiologist diagnostic accuracy by 12.2% compared to standard AI output. Original paper: The effect of medical explanations from large language models on diagnostic accuracy in radiology. — NPJ digital medicine. 10.1038/s41746-026-00001-1 📄 Read the article

Researchers have developed DxDirector, an agentic large language model system that partners with physicians to drive the full clinical diagnosis process through iterative reasoning and interaction. Original paper: DxDirector: an agentic large language model driving the full-process clinical diagnosis. — Nature communications. 10.1038/s41467-026-71928-5 📄 Read the article

A machine learning model successfully separated brain changes specific to mild cognitive impairment from normal aging, achieving superior prediction of dementia risk. Original paper: Decoupling MCI-specific signatures from shared neurobiological substrates of cognitive aging via deep learning. — NPJ digital medicine. 10.1038/s41746-026-02597-3 📄 Read the article

A 3D foundation model trained on self-supervised learning demonstrates powerful generalization for detecting multiple neurological diseases in head CT, achieving 16-20% accuracy improvement over conventional approaches. Original paper: 3D foundation model for generalizable disease detection in head computed tomography. — Nature biomedical engineering. 10.1038/s41551-026-01668-w 📄 Read the article

An interpretable machine learning model using multi-parameter ultrasound achieves 84% accuracy in discriminating early- versus late-stage ALS, offering an objective, non-invasive clinical tool. Original paper: Predicting amyotrophic lateral sclerosis stage based on multi-parameter ultrasound: development and validation of an interpretable machine learning model. — BMC medicine. 10.1186/s12916-026-04871-3 📄 Read the article

A machine learning model called OsteoSC-M3 can identify women at high risk of osteoporosis approximately 29 months before standard diagnostic confirmation, enabling preventive intervention. Original paper: A multiparameter model (OsteoSC-M3) for early detection and risk stratification of osteoporosis in women: a multicentre cohort study in China. — EClinicalMedicine. 10.1016/j.eclinm.2026.103902 📄 Read the article

A direct comparison of study designs for deep learning models reveals that cohort-based approaches better preserve real-world risk estimates, while case-control designs—despite higher apparent accuracy—systematically overestimate patient risk. Original paper: Cohort vs case-control design for transformer-based prediction of asthma exacerbations in mild asthma. — NPJ digital medicine. 10.1038/s41746-026-02624-3 📄 Read the article

Researchers have identified azole compounds, especially talarozole, as promising drug candidates for Leigh syndrome by combining deep learning computational screening with human brain organoid models derived from patient cells. Original paper: Accelerating Leigh syndrome drug discovery through deep learning screening in brain organoids. — Nature Communications. 10.1038/s41467-026-71391-2 📄 Read the article

A novel two-stage deep learning model significantly improves histological diagnosis of early pregnancy loss, achieving 84.3% accuracy while reducing diagnostic time and substantially aiding pathologist decision-making. Original paper: Patch-to-slide fusion deep learning model for histological diagnosis of early pregnancy loss including hydatidiform mole. — NPJ digital medicine. 10.1038/s41746-026-02628-z 📄 Read the article