Transfer Learning and Advanced CNN Models for Detecting Brain Tumors using MRI

Abstract

This paper explores the efficacy of three advanced deep convolutional neural network (CNN) models; DenseNet, MobileNet, and Xception in classifying brain tumors from MRI scans. Accurate detection and classification of brain tumors are critical for timely medical intervention, and recent advancements in deep learning offer promising tools for this task. We apply each model to a publicly available brain tumor dataset, evaluating their performance in terms of accuracy, sensitivity, specificity, and computational efficiency. The experiments utilize the well-known Brain Tumor Classification dataset, consisting of 3264 MRI images categorized into glioma, meningioma, pituitary tumor, and non-tumor classes. The results demonstrate that each model has unique strengths, with DenseNet showing superior accuracy, MobileNet excelling in computational efficiency, and Xception achieving a balance of both but better than others. The Xception achieved the most suitable performance with an accuracy of 97.6%, sensitivity of 97.9%, precision of 98.5%, specificity of 97.2%, and an F1-score of 97.9%. These results show that Xception excels over other architectures, making it highly effective in classifying abnormal and normal tumors from brain MRI images. Our findings highlight the importance of model selection based on specific clinical requirements and computational constraints and suggest pathways for further research and optimization in medical image analysis.