Deep Learning Technique for MRI-Based Brain Tumor Classification

Brain tumors represent a severe threat to human health due to their potential to disrupt vital functions controlled by the brain, such as movement, sensation, and cognition. This study aims to classify a comprehensive dataset of 7,023 magnetic resonance imaging (MRI) images from Figshare, SARTAJ, and BR35H into four distinct classes (Glioma, Meningioma, Pituitary, and No-Tumor). After evaluating the state-of-the-art models including VGG16, DenseNet201, ResNet50, and V3 Inception, the study proposes a novel Convolutional Neural Network (CNN) model with multiple inception layers. As a result, the model achieved an accuracy of 98.71% and high AUC scores for all four categories. These results suggest the potential of this model in enhancing diagnostic accuracy and reducing radiologist workload in clinical applications. Further improvements could focus on advanced feature extraction and integration of multi-parametric MRI data. The model's robustness and effectiveness underscore its potential utility in real-world clinical settings for accurate tumor classification.