Gliomas are the most common primary tumors of the central nervous system, and their pathological grading plays a critical role in guiding treatment decisions and prognostic evaluation. In this study, we retrospectively collected data from 53 patients who underwent glioma surgery at the Eastern Theater General Hospital between January 2023 and January 2024. Among these, 33 cases were high-grade gliomas, and 20 were low-grade gliomas. Raman spectral data of tumor tissue samples were obtained using the InVia laser confocal Raman spectrometer (UK), with 50 points collected for each sample. The spectral data were preprocessed using various methods, including the Savitzky-Golay (SG) algorithm, spectral curve smoothing, and min-max normalization. A convolutional neural network (CNN) was developed to classify gliomas into high- and low-grade categories, and its performance was compared with traditional machine learning models, including support vector machines (SVM), random forests (RF), and decision trees (DT). Each predictive model was evaluated using receiver operating characteristic (ROC) curves, and four key metrics- accuracy, precision, recall, and five-fold cross-validation- were employed to assess model performance. Experimental results demonstrated that the CNN model significantly outperformed the SVM, RF, and DT models in various classification tasks, achieving an area under the curve (AUC) of 0.983 9, compared to 0.915 7 for SVM, 0.903 1 for RF, and 0.780 9 for DT. These findings suggest that integrating Raman spectroscopy with deep learning techniques offers an innovative approach to the grading diagnosis of gliomas. This method improves diagnostic accuracy and efficiency and lays a solid foundation for the future development of automated cancer diagnostic systems.