Authors: Jyoti Gahora, Bhanu Pratap Singh
Abstract: Brain tumors are among the most critical neurological disorders that require early and accurate diagnosis for effective treatment and improved patient survival. Magnetic Resonance Imaging (MRI) is widely used for brain tumor diagnosis because of its superior soft tissue visualization capability. However, manual tumor detection and classification are time-consuming and highly dependent on radiologists’ expertise. To overcome these limitations, this research proposes an intelligent MRI-based brain tumor detection and classification system using deep learning techniques. The proposed framework integrates preprocessing, segmentation, feature extraction, deep learning classification, and performance evaluation into a unified automated system. Initially, MRI images undergo preprocessing steps such as artifact removal, noise reduction, intensity normalization, and bias field correction to improve image quality. Segmentation techniques including thresholding, region growing, and watershed algorithms are then applied to isolate tumor regions from healthy brain tissues. Histogram-based, texture-based, and shape-based features are extracted to improve discriminative learning. The EfficientNetB3 deep learning model is employed for tumor and non-tumor classification due to its efficient feature learning and lightweight architecture. Hyperparameter tuning techniques such as optimized learning rate, batch size, dropout regularization, and data augmentation are used to improve classification performance and reduce overfitting. The proposed model achieves high performance with improved accuracy, precision, recall, and F1-score compared to existing approaches. Experimental results demonstrate that the proposed framework provides accurate and reliable brain tumor detection with enhanced segmentation and classification capability. The system also supports intelligent clinical decision-making and has the potential for future real-time healthcare applications.
