Artificial intelligence (AI) and machine learning are transformative technologies that have sparked both excitement and concern. In radiation oncology, AI has been successfully applied to automate tasks, such as auto-contouring, geometric adaptation, calculation of radiation dose and quality assurance, and predicting outcomes to provide optimal patient care. The clinical deployment of AI-driven and machine learning-based tools, however, continues to lag behind their perceived potentials, owing to a range of technical, practical, ethical and legal concerns. The original predictive AI algorithms have been expanded with technologies such as generative AI and foundation models to enable new applications, such as automated treatment planning and synthetic computed tomography generation, as well as improved prediction of individual patient outcomes. The increasing availability of innovations such as agentic AI, digital twins and multimodal AI could further improve the delivery and, thus, the efficacy of radiotherapy. In this Review, we present examples of successful AI applications in radiation oncology, provide an overview of the current challenges for implementing such tools in this specialty and strategies for addressing them, and discuss the broader implications of these tools in optimizing treatment outcomes and the quality of patient care.

The extracellular matrix (ECM) is a dynamic meshwork of proteins and glycans that provides architectural support to cells. This meshwork also conveys mechanical and biochemical signals that orchestrate many cellular processes, including proliferation, survival and migration. ECM accumulation is a hallmark of cancers and correlates with tumour aggressiveness and survival across cancer types. Because of its abundance and accessibility, the ECM constitutes an appealing reservoir of potential biomarkers and therapeutic targets. Yet, early attempts to target the ECM in patients with cancer have failed to provide clinical benefit. Over the past few years, the development of novel proteomic and imaging methods, together with advances in our knowledge of ECM biology provided by fundamental research, have transformed our understanding of the roles of the ECM in tumour progression. In this Review, we describe novel technologies developed to define the composition, physical and mechanical properties, and signalling functions of the ECM and provide an overview of ECM-dependent mechanisms regulating tumour progression. We then illustrate how the ECM is being leveraged to devise prognostic or diagnostic clinical tests. Lastly, we discuss strategies to target the ECM for clinical benefit by using the ECM for the targeted delivery of therapeutic payloads, normalizing the tumour ECM through matritherapies or modulating ECM signalling pathways.

Oncolytic virotherapy has evolved from a platform predicated on lysis of cancer cells into a sophisticated system for intratumoural immune reprogramming; the prevailing 'cold-to-hot' paradigm captures only part of this potential. A fundamental limitation shared by immune checkpoint inhibitors (ICIs) and early-generation oncolytic viruses (OVs) is a reliance on pre-existing tumour-specific T cells (TSTs), which presents an immunological ceiling that constrains antitumour activity - given that these agents have a limited capacity to generate TSTs de novo. Next-generation OVs overcome this constraint by functioning as agents for antigen-agnostic in situ cancer vaccination: intratumoural infection triggers immunogenic cell death, releasing potentially the full cancer proteome under pathogen-associated and/or damage-associated molecular pattern adjuvant conditions, thereby driving T cell priming against patient-specific neoantigens. In this Perspective, we delineate four pillars for the development of next-generation OVs. First, intratumoural vaccination as immunological ignition, with initial clinical data demonstrating T cell clonotype broadening, abscopal tumour regression and survival benefit in patients with ICI-refractory disease. Second, optimized payload-driven immune priming to hyperactivate antigen-presenting cells. Third, revised efficacy evaluation and end points using response criteria as well as novel biological correlates that better capture delayed and abscopal immune-mediated tumour control. Fourth, OV as the foundational immuno-oncology platform: according to the 'triple-A' framework encompassing admission, availability and activation of TSTs, next-generation, payload-engineered OVs uniquely satisfy all prerequisites for antitumour immunity simultaneously, with evidence from clinical trials of such agents supporting an 'OV-prime, ICI-amplify' strategy. Notably, OVs are systemic immune-reprogramming platforms that are delivered locally, not local therapies with incidental systemic effects.

Glioblastomas remain the most lethal primary brain tumour in adults, with targeted therapies delivering only limited benefit despite deep molecular characterization. Several targeted drugs have received regulatory approval for low-grade gliomas, although progress in glioblastomas remains constrained by, among other aspects, extensive intratumoural heterogeneity, pathway redundancy, cellular plasticity and limited drug delivery to the central nervous system. Some of these challenges might be mitigated through strategies that enhance blood-brain penetration, including focused ultrasonography, convection-enhanced delivery, efflux avoidance and chemical modifications. Improved tumour profiling through multiregional sampling, prioritization of truncal dependencies and the development of novel therapeutic modalities, such as antibody-drug conjugates and theranostics, might also further improve outcomes. In this Review, we summarize the therapeutic landscape of targeted therapies in glioblastomas, spanning major target classes including receptor tyrosine kinases, intracellular signalling proteins, cell-cycle dysregulation and synthetic-lethal vulnerabilities. We also examine emerging strategies targeting genome integrity and telomeres, epigenetic modulators, and tumour-neural circuitry. Furthermore, we highlight tumour heterogeneity and extrachromosomal DNA dynamics as key drivers of oncogene amplification and therapeutic resistance as well as the roles of novel clinical trial designs and liquid biopsy-based monitoring strategies. Lastly, we discuss pathway-based glioblastoma classification and master kinase mapping as methods for aligning drugs with functional tumour states.