Comparative Analysis of Deep Learning Approaches for Brain Tumor Detection: CNN, Improved CNN, and Transfer Learning with VGG-16

Abstract

Brain tumor detection using deep learning has gained significant attention due to its potential for accurate and automated diagnosis. This study explores three advanced deep learning approaches namely Conventional CNN, Improved CNN architecture, and Transfer Learning with VGG-16 to evaluate their effectiveness in classifying brain tumors from MRI images. The models were trained (80%) and tested (20%) using a well-structured dataset specifically MRI dataset from Kaggle comprising 1755 images, and their performances were assessed based on accuracy and loss metrics. The results indicate that while the Conventional CNN provided a baseline accuracy of 85%, the Improved CNN demonstrated enhanced feature extraction capabilities, achieving 90.0% accuracy. However, Transfer Learning with VGG-16 outperformed both, attaining an impressive accuracy of 97.4% with minimal loss, highlighting its superior generalization and robustness. The confusion matrix further confirmed VGG-16’s reliability, showing significantly reduced misclassification errors. These findings underscore the effectiveness of deep learning in brain tumor detection, with VGG-16 proving to be the most efficient model. Despite these promising results, challenges such as dataset limitations and computational complexity remain. Future research should focus on optimizing deep learning architectures and leveraging larger datasets to further enhance accuracy and clinical applicability. This study contributes to the growing field of AI-driven medical diagnostics, offering insights into selecting the most effective model for brain tumor detection.