MRI-Based Brain Tumor Detection and Classification Using a Hybrid MobileNet and Ensemble Learning Framework

Abstract

Diagnostic magnetic resonance imaging (MRI) imaging is an important tool in the diagnosis of brain tumors to aid timely clinical decisions, but is difficult to interpret due to inter- and intra-tumor heterogeneity, and similarities between tumor types. In this study, a lightweight hybrid computer-aided diagnosis (CAD) framework for automated brain tumor detection and classification based on MRI is proposed that combines deep feature extraction and ensemble machine learning. Preprocessing of MRI images involved intensity normalization, histogram equalization, resizing and region of interest (ROI) extraction using watershed transformation to improve the image quality and normalize input images. A MobileNet architecture is pre-trained and deep discriminative features are extracted from them and they are classified by a hybrid Support Vector Machine (SVM) and Random Forest (RF) ensemble architecture with soft voting strategy. As compared with conventional single-classifier model the proposed model improved the classification accuracy and reduced its computing complexity, achieving classification accuracy of 94.1% with a precision of 0.93, a recall of 0.94 and an F1 score of 0.935. The ensemble of classifiers, a hybrid model, and the lightweight MobileNet architecture provide a favorable trade-off to model performance and computational efficiency for neural networks, making it appropriate for resource-constrained clinical settings. The proposed technique will offer an efficient and scalable computer aided diagnosis solution that will support radiologists' efficient and accurate diagnosis. In the future, the following techniques of explainable AI and multi-center data sets and multimodal data integration will be incorporated into the model and it will further strengthen the model