MRI-Based Identification of Brain Tumors Using Deep Convolutional Neural Networks: A Case Study on Inception V3

The detection of brain tumors in MRI scans is a critical task in medical imaging, aiding timely diagnosis and treatment. This paper presents a comparative analysis of deep learning and traditional machine learning approaches for brain tumor detection. Specifically, convolutional neural networks (CNNs) and support vector machines (SVM) were evaluated to assess their performance. Five state-of-the-art CNN architectures—InceptionV3, VGG16, VGG19, MobileNetV2, and ResNet50—were employed using transfer learning and advanced feature extraction techniques. InceptionV3 demonstrated the highest recall, excelling in multi-scale feature extraction, while ResNet50 provided robust and consistent performance. VGG models balanced accuracy and computational efficiency, and MobileNetV2 proved suitable for resource-constrained environments. Additionally, the study explored the impact of dataset size and preprocessing techniques on model performance, revealing that data augmentation significantly improved CNN generalization. SVM achieved commendable precision on smaller datasets through texture-based feature extraction using the Gray-Level Co-Occurrence Matrix (GLCM), though it struggled with larger datasets due to limited scalability. Experimental results revealed that CNNs significantly outperformed SVM in scalability and adaptability, with InceptionV3 achieving a 94.12% accuracy, highlighting the superiority of deep learning in complex medical imaging tasks.