Glioma Classification Using Raman Spectroscopy and Machine Learning Models on Fresh Tissue Samples.

Identifying tumor cells infiltrating normal-appearing brain tissue is critical to achieve a total glioma resection. Raman spectroscopy (RS) is an optical technique with potential for real-time glioma detection. Most RS reports are based on formalin-fixed or frozen samples, with only a few studies deployed on fresh untreated tissue. We aimed to probe RS on untreated brain biopsies exploring novel Raman bands useful in distinguishing glioma and normal brain tissue. Sixty-three fresh tissue biopsies were analyzed within few minutes after resection. A total of 3450 spectra were collected, with 1377 labelled as Healthy and 2073 as Tumor. Machine learning methods were used to classify spectra compared to the histo-pathological standard. The algorithms extracted information from 60 different Raman peaks identified as the most representative among 135 peaks screened. We were able to distinguish between tumor and healthy brain tissue with accuracy and precision of 83% and 82%, respectively. We identified 19 new Raman shifts with known biological significance. Raman spectroscopy was effective and accurate in discriminating glioma tissue from healthy brain ex-vivo in fresh samples. This study added new spectroscopic data that can contribute to further develop Raman Spectroscopy as an intraoperative tool for in-vivo glioma detection.