Intraoperative Tumor-Tissue Discrimination and Histology-Defined Glioma Grading Using Combined Fluorescence and Diffuse Reflectance Spectroscopy
Fernando S Chiwo, Alankar Kotwal, Melisa Herrero Bocco, Alejandro Villasante-Tezanos, Eric Suero Molina, Brent T Harris, Joshua D Bernstock, Keith D Paulsen, David W Roberts, Ashok Veeraraghavan, Pablo A ValdesAbstract
Background
Maximal safe glioma resection requires distinguishing infiltrative tumor from surrounding brain. Visible 5-aminolevulinic acid (5-ALA)-induced protoporphyrin IX (PpIX) fluorescence guides surgery, but low or heterogeneous PpIX accumulation and optical attenuation limit sensitivity. We tested whether combined fluorescence spectroscopy (FS) and diffuse reflectance spectroscopy (DRS) could improve tumor-tissue discrimination beyond visible fluorescence.
Methods
Thirty-three patients with historically classified low-grade (LGG), high-grade (HGG), or recurrent glioma underwent 5-ALA-guided surgery. A handheld bimodal probe acquired in vivo FS and DRS from biopsy-paired and navigation-defined presumed non-tumor sites. A light-transport model estimated optical properties, corrected fluorescence, and quantified biomarkers. Five predictor representations were evaluated using supervised machine learning for non-tumor versus tumor discrimination and sequential LGG versus HGG classification with patient-grouped leave-one-patient-out cross-validation. Patient-clustered bootstrapping compared FS+DRS with visible fluorescence.
Results
Across 212 sites, FS+DRS achieved 85.82% sensitivity, 74.36% specificity, 80.09% balanced accuracy, and 0.81 AUROC for tumor discrimination. FS and DRS alone achieved 73.16% and 66.17% balanced accuracy, respectively. Compared with visible fluorescence, FS+DRS increased sensitivity by 34.3 percentage points (95% CI, 19.3–47.5; p < 0.001), while specificity differed by − 11.5 points (95% CI, −27.8–4.0; p = 0.2). FS+DRS detected 55/65 nonfluorescent tumor sites, including 19/23 historical LGG sites. For grade discrimination, biomarkers achieved 71.98% balanced accuracy and 0.71 AUROC.
Conclusion
Combined FS+DRS improved tumor detection and identified histology-positive tumor not visible with standard fluorescence. Prospective multicenter evaluation should use a locked model, molecularly classified cohorts, prespecified clinical endpoints, and spatial registration to determine whether the added signal improves intraoperative decisions and patient outcomes.