Hyperspectral imaging and machine learning in brain tumor surgery: Current evidence, clinical applications, and future directions

Background: Brain tumor surgery aims to achieve maximal safe resection while preserving neurological function; however, accurate intraoperative discrimination between tumor tissue and normal brain remains a major challenge. Hyperspectral imaging (HSI) has emerged as a promising label-free optical imaging modality that provides quantitative information on tissue biochemical and physiological characteristics. The integration of machine learning (ML) has further enhanced its potential for automated tissue classification and real-time intraoperative decision support. This narrative review summarizes the current evidence regarding HSI and ML in brain tumor surgery, highlighting technological principles, clinical applications, current limitations, and future directions. Methods: This narrative review was based on a comprehensive literature search of PubMed, Scopus, and Google Scholar to identify studies evaluating the application of HSI and artificial intelligence (AI) in brain tumor surgery. The search strategy included combinations of the keywords “ hyperspectral imaging ,” “ brain tumor ,” “ glioma ,” “ machine learning ,” “ deep learning ,” “ artificial intelligence ,” and “ neurosurgery .” Priority was given to peer-reviewed original research articles, narrative and systematic reviews, and clinically relevant investigations describing the technical development, intraoperative application, and translational potential of HSI in neurooncology. Landmark studies that established the foundations of HSI, together with recent publications reflecting advances in ML and clinical implementation, were preferentially included to provide both historical context and contemporary perspectives. Manual screening of the reference lists of key publications was performed to identify additional relevant studies. As a narrative review, this manuscript was intended to provide a critical overview of current evidence rather than a systematic assessment of the literature. Consequently, no formal study selection algorithm, quality assessment tool, or quantitative meta-analysis was performed. Results: Current evidence indicates that HSI combined with ML enables accurate intraoperative tissue classification, improved tumor margin delineation, and emerging molecular characterization, including prediction of isocitrate dehydrogenase mutation status. Deep learning approaches generally outperform conventional ML techniques by integrating spectral and spatial information. Although early clinical studies have demonstrated encouraging diagnostic performance, widespread implementation remains limited by heterogeneous imaging protocols, relatively small patient cohorts, limited multicenter validation, and computational challenges. Conclusion: HSI represents a promising adjunct for precision-guided brain tumor surgery by providing real-time biochemical tissue characterization without the need for exogenous contrast agents. Continued advances in AI, standardized imaging protocols, multicenter clinical validation, and multimodal intraoperative integration will be essential to facilitate routine clinical adoption and establish its impact on surgical and patient outcomes.