Optical express-biopsy of gliomas using macroscopic fluorescence lifetime imaging.

In glioma surgery, the quality of tumor resection largely determines patient prognosis. Accurate intraoperative discrimination of glial tumors from normal brain tissue and delineation of tumor margins is a major challenge. Given the inherent biochemical differences between tumor and normal tissue, imaging techniques based on cellular autofluorescence represent a promising approach to address this challenge. The aim of this study was to evaluate the ability of macroscopic fluorescence lifetime imaging, macro-FLIM, to discriminate between different classes of glioma (glioblastoma, astrocytoma, oligodendroglioma) and normal brain tissue and to identify glioma cells in the peritumoral region. The study was performed on 110 freshly excised tissue samples from 53 patients. Macro-FLIM images were acquired in the NAD(P)H spectral channel (ex. 375 nm, em. 435-485 nm) using a confocal laser macroscanner. In human performance, the sensitivity of macro-FLIM in discriminating glioblastoma from normal tissue was 92.3% (AUC 0.905), astrocytoma and oligodendroglioma - 62.5% (AUC 0.796 and 0.687). To automatically classify the macro-FLIM images, the Random Forests machine learning algorithm was developed, which reliably discriminated glioblastoma from all normal (82.4% sensitivity, AUC 0.86), astrocytoma from white matter (80.3% sensitivity, AUC 0.857), and oligodendroglioma from gray matter (89.2% sensitivity, AUC 0.875). In addition, the classification model demonstrated the ability to detect areas of tumor infiltration within the peritumoral white matter. The current results demonstrate the potential of NAD(P)H-based macro-FLIM combined with machine learning as a surgical guidance tool to improve glioma resection.