Multimodal artificial intelligence in glioma management: integrating neuroimaging and hematologic biomarkers for precision oncology

Background – Gliomas are biologically heterogeneous primary brain tumors that remain challenging to diagnose, prognosticate, and monitor noninvasively, owing to marked intratumoral heterogeneity, treatment-related imaging changes, and limited accessibility of tissue biomarkers. Despite advances in molecular classification, clinical decision-making still relies heavily on neuroimaging, highlighting the need for integrative, data-driven approaches. Objective – This narrative review examines how artificial intelligence (AI) can integrate multimodal neuroimaging with hematologic and other liquid biomarkers to support clinical decision-making in glioma management. Content – We synthesize recent advances in machine learning (ML) and deep learning (DL) applied to MRI and PET for glioma detection, segmentation, molecular phenotype inference, and outcome prediction. We review both segmentation-based and segmentation-free modeling paradigms, highlighting their respective assumptions, advantages, and limitations. Advanced imaging techniques, including diffusion (DWI, DTI) and perfusion imaging, MR spectroscopy, and metabolic and amino acid PET, are discussed as sources of biologically specific signals that extend beyond conventional structural imaging. We further examine blood-derived biomarkers, such as inflammatory and immune mediators, circulating nucleic acids, and extracellular vesicle cargo, which provide complementary insights into tumor–host interactions and enable longitudinal assessment. Emerging generative and systems-level modeling approaches are also reviewed in the context of multimodal data integration and clinical application. Conclusion – Multimodal AI has the potential to integrate spatial imaging phenotypes with systemic biological signals to improve noninvasive diagnosis, molecular risk stratification, and treatment monitoring in gliomas. Translation to clinical practice will depend on appropriate methodological design choices, standardized workflows, rigorous external validation, uncertainty-aware decision support, and continuous performance monitoring in real-world settings.