EfficientNetB3-Based Deep Learning Framework for High-Precision Brain Tumor Classification from MRI Scans: A Comprehensive Study

Abstract

Brain tumors are among the most deadly and difficult neurological diseases, requiring prompt and precise diagnosis for proper treatment planning and improved patient outcomes. Magnetic Resonance Imaging (MRI) is the gold standard of brain imaging because of its excellent resolution and capacity to discriminate between soft tissues. Yet, manual reading of MRI scans is time-consuming as well as susceptible to radiologist subjective variability. To solve this, we suggest a deep learning model with the EfficientNetB3 architecture to classify brain tumors automatically. The model is trained to classify tumors into three types of clinical importance: glioma, meningioma, and pituitary tumors. A specially prepared dataset with 2,144 training, 458 validation, and 462 test images is employed in this research. We employed a detailed preprocessing and data augmentation pipeline to improve generalization as well as prevent overfitting. By exploiting transfer learning from ImageNet and fine-tuning the EfficientNetB3 model, we attained a superb classification accuracy of 99.69% on the test set. Besides accuracy, we present high precision, recall, as well as F1-scores, which further establish the reliability and robustness of the model. This article offers a comprehensive analysis of model design, training approach, evaluation measures, and comparison with other CNN-based models like VGG19 and ResNet50. Our results indicate that EfficientNetB3 is not only computationally sound but also highly efficacious for medical image classification tasks. We deduce that our framework has high prospects for real-world deployment in clinical settings, assisting radiologists with early diagnosis and decision-making