Automated Brain Tumor Classification Using NASNet-Large and AutoGluon

Abstract

Brain tumors pose a crucial medical challenge, requiring prompt and accurate diagnosis for effective treatment. Traditional methods, like manual MRI analysis, can be time-consuming and susceptible to error. Recent progress in deep learning has led to automated MRI-based tumor classification systems. This study explores the use of NASNet-Large, coupled with AutoGluon's ImagePredictor and a stacked ensemble learning strategy to classify brain tumors into four categories: Glioma, Meningioma, Pituitary tumor, and No Tumor. The model, trained on a labeled MRI dataset, achieves an overall accuracy of 98.40% and F1-scores exceeding 96% for all tumor types, demonstrating its effectiveness in distinguishing between classes. These findings highlight the possibilities of deep learning tools in enhancing medical imaging, reducing diagnostic delays, and improving accuracy. However, challenges remain, such as the need for larger datasets and ensuring model generalizability across imaging modalities. Future research should focus on enhancing the model and integrating it into clinical workflows.