Comparative Study of CNN Architectures for Brain Tumor Classification Using MRI: Exploring GradCAM for Visualizing CNN Focus †

Chinga A.; Bendezu W.; Angulo A. · Engineering Proceedings (2025) · DOI: 10.3390/engproc2025083022

Resumen

Brain cancer, with its varied nature, demands early detection for timely treatment. This study aims to refine the diagnosis of brain tumors using convolutional neural network algorithms. Currently, diagnostic accuracy is limited, therefore, our approach uses five different CNN architectures to accurately identify and classify affected brain regions, specifically glioma, meningioma, or pituitary tumors. The AlexNet architecture remarkably achieved training accuracy (99.84%) and validation accuracy (95.19%). By employing GradCAM, heat maps visually clarify the results. This research aims to improve the diagnosis of brain tumors using advanced CNN algorithms. In particular, the success of AlexNet indicates greater diagnostic and treatment potential, promising better outcomes for patients.

Instituciones: Universidad San Ignacio de Loyola