Multi-class Classification of Brain Tumors from MRI Image using Hybrid CNN

Abstract

Brain tumors are a significant health concern, requiring timely and accurate diagnosis to improve treatment outcomes. Magnetic Resonance Imaging (MRI) is a widely used approach for detecting brain tumors, and the traditional method of MRI interpretation is manual, time-consuming and liable to errors. In this study, a hybrid deep neural network is proposed utilizing the architectural features of VGG16, ResNet50 and EfficientNetB3 for multi-class brain tumors classification. The aim is to develop a model that is not only efficient and high performing, but also capable of classifying four tumorous and non tumorous brain MRI scans: glioma, meningioma, and pituitary tumors. The proposed model was trained and evaluated on a publicly available dataset from Kaggle, containing 7,023 MRI images. A robust preprocessing pipeline was used which consisted of normalization, resizing, and augmentation to adapt the images to be more consistent and to reinforce generalization for the training. The model was trained using a 5-fold cross validation strategy to ensure unbiased performance and robustness across different subsets of the data. The hybrid model achieved an accuracy of 98.77%, with weighted precision, recall, and F1-scores all also recorded at 98.77%. Additionally, the model’s compact size of just 2.20 MB ensures computational efficiency and adaptability, making it suitable for resource-constrained environments. Compared to existing classification methods, which require higher storage and computational resources, the proposed model performs effectively while requiring fewer resources. This research demonstrates the potential of hybrid deep learning models for medical image classification, providing a promising solution for faster and more accurate brain tumor diagnosis. The findings suggest that the model’s compact size and high accuracy could significantly enhance clinical workflows and decision-making processes in medical imaging.