Terahertz-based prognostic framework for glioma: from spectral features to patient-level prognostic assessment

In glioma, accurate prediction of recurrence and survival is essential for clinical decision-making and individualized management. However, current prognostic tools do not fully capture outcome heterogeneity, and convenient approaches for early postoperative prognostic evaluation are lacking. We investigated whether terahertz (THz) spectroscopy could provide label-free prognostic information. In this single-center retrospective study, 56 patients were assigned to training (n = 33) and held-out internal testing (n = 23) cohorts. A total of 443 frozen sections were measured using THz time-domain spectroscopy, and patient-level features were derived from six spectral parameters across 0.2-1.4 THz. Separate progression-free survival (PFS) and overall survival (OS) signatures were developed using univariable Cox screening, LASSO-Cox selection, and multivariable Cox modeling and validated in the testing cohort. The THz-based risk score (THz-RS) significantly stratified PFS and OS in both cohorts, showed good time-dependent discrimination, and retained prognostic value after multivariable adjustment for clinicomolecular variables. Model interpretation identified the 1.2-1.3 THz band (B11) as a major contributor, while targeted LC-MS/MS revealed associations between tissue glutamate abundance and B11-derived THz features, providing preliminary biochemical support for this spectral region. These findings support THz spectroscopy as a label-free complementary tool for perioperative prognostic assessment in glioma, pending validation in larger multicenter cohorts.