Cancer imaging articles from across Nature Portfolio

This study investigates technical noise in Xenium spatial transcriptomics data, including transcript spillover, and introduces SPLIT to resolve mixed signals and enhance cell-type specificity.

Chronological age is a known predictor of cancer survival, but it cannot fully capture variations in biological aging rates. Here, the authors introduce Face Aging Rate (FAR) as a non-invasive biomarker that predicts biological age from serial facial photographs using artificial intelligence, which can serve as a prognostic factor in cancer patients.

Zhu et al. have developed DeepFAN, a transformer-based deep learning model for diagnosis of incidental pulmonary nodules from computed tomography scans, and evaluated it for human–artificial intelligence collaboration in a multireader, multicase clinical trial.

In this Tools of the Trade article, Zhengyuan Pang describes the development and use of volumetric clearing-assisted tissue click chemistry (vCATCH), a pipeline that enables mapping of covalent drug binding at cellular resolution across a mouse body.

Applications of digital pathology in clinical oncology have largely depended on the requirement for labeled data and model re-training. A study now presents PRET, a training-free framework with robust performance for pan-cancer diagnosis that adapts pathology foundation models to diverse tasks at inference stage, from screening and subtyping tasks to segmentation and metastasis detection tasks.

Whether support from artificial intelligence (AI) models improves breast screening is a topic of research in national cancer screening programmes and clinical oncology. Three studies now show that AI tools can assist radiologists with evaluating mammograms, substantially reducing workloads and possibly improving screening performance.

In this Journal Club, Siliceo and Rodríguez discuss a study that presents histoCAT, an interactive toolbox for spatial biology that combines image-based visualization with unsupervised cellular neighborhood analysis to uncover tissue architecture and cell–cell interactions.

Foundation models hold transformative promise for oncology, yet their clinical implementation remains limited, largely owing to their current model design as narrow specialists optimized for static tasks, whereas clinical oncology requires generalist systems capable of integrating multimodal data, capturing disease evolution over time and considering patient perspectives. Design along these requirements is essential to integrating foundation models as trusted partners in cancer care.