Artificial intelligence-based biomarkers for the diagnosis and treatment of neurological conditions: a narrative review

Abstract

Artificial intelligence (AI) is transforming biomarker discovery in neurology by overcoming key limitations of conventional approaches that are often slow, reductionist, and unable to integrate complex multimodal data. In this narrative review, we searched the data bases; PubMed, Scopus, IEEE Xplore, CINAHL, Embase and the Cochrane Library from inception to 2025 to evaluate how AI supports biomarker identification, diagnosis, prognostication, and treatment stratification across neurovascular, neurodegenerative, neuro-oncological and seizure disorders. Evidence demonstrates that AI-driven imaging and multi-omics biomarkers can detect disease earlier, improve prediction accuracy, and support personalised care. For example, AI models improve stroke outcome prediction beyond conventional scores, identify intracranial aneurysms with sensitivities exceeding 90%, predict conversion from mild cognitive impairment to Alzheimer's disease with accuracies approaching 85-90%, and extract radiogenomic biomarkers in gliomas that outperform traditional diagnostic strategies. However, real-world translation remains constrained by dataset bias, limited external validation, interpretability challenges, and gaps in generalisability, particularly in underrepresented populations. Overall, AI-driven biomarker discovery offers a powerful pathway toward precision neurology, with the greatest impact expected when technical innovation is paired with robust clinical validation, regulatory integration, and equitable data representation.