Improved Glioma Detection and Classification through the EVGG19 Model

Gliomas, a diverse and complex category of brain tumors, present significant challenges in accurate classification due to their heterogeneous nature. Precise classification of gliomas into their respective subtypes and grades is crucial for effective clinical decision-making and personalized treatment planning. This study proposes an enhanced convolutional neural network architecture, EVGG19, designed specifically for the classification of gliomas using MRI data. This effort aims to incorporate domain-specific innovations and improve the glioma classification's accuracy and reliability. Our proposed workflow begins with the preprocessing and normalization of MRI images, followed by utilizing the DHA-ISSP model for accurate tumor segmentation. The segmented tumor regions are then fed into the EVGG19 model, which includes additional convolutional layers, increased model depth, dropout regularization, and a dedicated classification layer to refine the extraction and representation of features relevant to glioma classification. The performance of the EVGG19 model was rigorously evaluated using the TCGA dataset. Our model achieved an accuracy of 0.94, precision of 0.89, recall of 0.91, and F1 score of 0.9, significantly outperforming existing baseline models such as VGG Net-Based Deep Learning, UNet-VGG16 with transfer learning and VGG-UNET. Furthermore, EVGG19 demonstrated superior specificity (0.96), sensitivity (0.93), and AUC (0.97), along with the lowest MAE of 0.1 and MSE of 0.2. These findings demonstrate how well the EVGG19 model can distinguish glioma grades and subtypes, providing a robust tool for clinical application and furthering the potential for improved patient outcomes through more precise diagnostic capabilities.