Raman spectroscopy is a well-established technique that can differentiate biological tissues by identifying features associated with established pathological conditions. Despite its significant potential, the adoption of Raman technology in routine medical diagnostics remains challenging, primarily because of difficulties in incorporating Raman data acquisition and analysis into existing clinical workflows. In this study, we show that Raman signals in the CH stretching region (2800–3050 cm −1 ) offer a robust approach for the detection and diagnosis of brain tumors. Our investigation focuses on medulloblastoma disease, which accounts for 20% of pediatric brain tumors. We investigated analytical approaches ranging from a simple single-channel Bayesian statistical model to more advanced machine learning techniques. Our results show that relying solely on CH band intensities within a Bayesian framework limits the predictive power of the spectra. In contrast, machine learning algorithms substantially enhance diagnostic performance, achieving an accuracy of 90% on the full dataset and exceeding 79% in external validation. These findings suggest that such algorithms are suitable for ex vivo analysis in both laboratory and surgical settings. • Raman CH stretching vibration is effective for tumor discrimination. • Ex-vivo murine model closely mimics human medulloblastoma. • Single Channel Bayesian analysis has limited diagnostic efficiency . • Machine learning significantly improves classification accuracy. • Random-Forest captures complex biochemical interactions.
Giordo et al. (Fri,) studied this question.