Brain Tumor Classification using Pre-Trained Deep Convolutional Neural Networks

Abstract

The classification of brain diseases is an extremely challenging task owing to their intricate and sensitive nature. Brain tumors are significant and life-threatening, requiring accurate and prompt diagnosis for effective treatment planning. MRI is a crucial medical imaging tool, providing detailed, non-invasive brain imaging. MRI is crucial in brain tumor detection, influencing the diagnosis and management of these conditions. The approach initiates with dataset preprocessing, encompassing MRI scans and clinical data from individuals with various brain conditions, both tumor and non-tumor cases. Dataset is categorized as training and testing sets. Identifying tumors in MRI scans encompasses various stages, including image preprocessing, feature extraction, and classification. The system uses Convolutional Neural Networks with VGG-16 and Xception models for brain image classification. VGG-16, known for its deep architecture and strong feature extraction, is used alongside Xception, a highly efficient model for image classification. The results indicate significant advantages of Xception's transfer learning models over VGG-16. Transfer learning models automatically extract hierarchical features from raw image data, eliminating the need for manual feature engineering. The feature extraction capability empowers CNNs to capture nuanced and complex patterns in brain images, elevating their diagnostic accuracy. The dataset combines figshare, SARTAJ, and Br35H sources, comprising 7,023 MRI images of human brain is divided into four classes: meningioma, no tumor, pituitary and glioma. This study results affirmed the method's effectiveness, achieving an impressive 94% accuracy in brain tumor detection. Users can input brain MRI images to predict tumor types and receive detailed diagnosis information based on accuracy of the model. Experiment results show that the proposed system surpasses existing systems in disease prediction efficiency.