To leverage artificial intelligence for developing diagnostic tools based on multiparametric analyses of cerebrospinal fluid (CSF) inflammatory markers across neurological disorders.

CSF provides a unique window into central nervous system pathology and remains routinely used despite advances in neuroimaging. It is a valuable but underexploited source of biomarkers, as current analyses typically assess only a few markers. Recent multiplex technologies now enable comprehensive profiling of inflammatory, glial, and neuronal signals, which require advanced computational approaches for interpretation.

We retrospectively analyzed 750 patients who underwent lumbar puncture at Lausanne University Hospital (2010–2025) and received a definite neurological diagnosis established by a neurologist across >40 etiologies (including multiple sclerosis, Alzheimer’s disease, CNS lymphoma, and inflammatory neuropathies). For each patient, 57 cytokines, chemokines, and growth factors were quantified by Luminex, in addition to routine CSF parameters (cell count, protein levels, oligoclonal bands, etc.). Unsupervised clustering and supervised machine learning models, including XGBoost, were applied to classfy patients according to diagnosis and to identify key discriminatory biomarkers.

High-dimensional CSF profiling distinguished major diagnostic categories and revealed disease-specific immune signatures. Integrating multiparametric data into machine learning models increase diagnostic performance compared with conventional CSF parameters alone. For instance, the classifier successfully differentiated clinically overlapping entities such as multiple sclerosis vs. neurosarcoidosis (AUC:0.82) and CIDP vs. Guillain–Barré syndrome (AUC:0.71). We further developed an interactive visualization tool to facilitate clinical interpretation and support diagnostic decision-making.

Combining multiparametric CSF analyses with artificial intelligence enables identification of disease-specific inflammatory signatures and improves diagnostic accuracy across neurological disorders. This integrative approach holds promise for precision diagnostics and for guiding future biomarker-driven clinical applications but require validation in a prospective manner.