Surface enhanced Raman spectroscopy and machine learning as an accurate and rapid diagnostic tool for hydrocephalus

Hydrocephalus is a severe neurological disorder marked by cerebrospinal fluid (CSF) accumulation in the brain's ventricles, causing elevated intracranial pressure, neurological deficits, and substantial societal and medical costs. Rapid detection in neonates and children remains challenging, as primary diagnostic indicators such as increased head size often manifest late, delaying intervention. Hydrocephalus stems from diverse causes, including prematurity, intracranial bleeds, CNS infections, and brain tumours. CSF collected for diagnostic purposes provides an opportunity to explore innovative rapid detection methods. In this paper, we show that integrating Surface-Enhanced Raman Spectroscopy (SERS) and machine learning offers a novel molecular diagnostic tool for hydrocephalus. Using silver nanoparticle-layered-cellulose strips as SERS substrates and a portable Raman spectrometer, we analysed CSF samples from 117 patients (70 controls, 47 hydrocephalus cases). Within 5 min of sample placement in strip, the optimized Random Forest algorithm achieved 97% accuracy in blind testing, with 100% specificity and 95% sensitivity. The miniaturized Raman spectrometer and standarised strips enable portability and support clinical use, particularly in resource-limited settings. Our grid search-based ML workflow and scoring system enable the prediction of overfitting, as well as assessment of feature importance within the spectra, improving model interpretability, which could prove helpful in other vibrational spectroscopy applications. Overall, this study provides evidence of an accurate, rapid, portable, and interpretable diagnostic technique for hydrocephalus. Further validation with larger cohorts will refine predictive models and expand clinical utility, advancing diagnostic precision for hydrocephalus and related neurological conditions.