Evaluation of Deep Convolutional Neural Network Architectures for Brain Tumor Classification
Tipo:
Articulos de Divulgación
Autor:
Dorcelus, S., Del Razo-López, F., Alejo-Eleuterio, R., Villanueva-Vásquez, D., Granda-Gutiérrez, E.E.*
Fecha:
2026-05-01
Descripción:
Artificial intelligence is increasingly helping doctors analyze medical images, such as magnetic resonance imaging (MRI) scans, to detect conditions such as brain tumors. However, not all AI models work equally well, and selecting an appropriate model is essential. In this study, we compare three widely used deep learning models (VGG16, ResNet50, and InceptionV3) to determine which performs best in classifying brain tumors from MRI images. Under identical evaluation conditions for all three models, we found that InceptionV3 clearly outperformed the others, achieving 94% accuracy and consistently higher scores across multiple performance metrics. Its unique design, which captures image details at multiple scales simultaneously, appears especially well-suited for the complex patterns seen in brain tumors. While VGG16 offers a simpler alternative with acceptable results when computing resources are limited, ResNet50, despite its popularity in other fields, showed weaker performance in this specific task. Our findings offer practical, evidence-based guidance for researchers and healthcare teams looking to apply AI tools in neuroimaging, highlighting InceptionV3 as the preferred choice for brain tumor classification.
Keywords: Artificial intelligence, Brain tumor classification, Convolutional networks, Deep neural networks.
;)
