Resumo

PETRLIK AZABACHE, Iván Carlo; COVENAS LALUPU, José  e  ESPARZA SILVA, Milciades Roberto. COMPARATIVE EVALUATION OF CNN CLASSIFICATION MODELS FOR BRAIN TUMORS IN MRI IMAGES. Aula Virtual [online]. 2025, vol.6, n.13, e563.  Epub 25-Out-2025. ISSN 2665-0398.  https://doi.org/10.5281/zenodo.17369525.

The research conducts a comparative evaluation of convolutional neural network (CNN) models for the classification of brain tumors in MRI images, in response to the need for accurate and efficient automated diagnostic systems in the medical field. A total of 1,683 images obtained from public repositories were used, corresponding to adenomas (403), gliomas (459), meningiomas (416), and healthy tissue (405). The methodology used in this research included five phases: acquisition, preprocessing with interpolation and outlier correction, class balancing using SMOTE, data augmentation, transfer learning, and evaluation. Eighteen CNN architectures were trained, of which only nine achieved an accuracy level of ≥80%. BrainNet-7 achieved the highest accuracy (99.2%), followed by EfficientNet-B2 (98.7%) and Swin Transformer Tiny (98.0%). On the other hand, validation using standard metrics and Grad-CAM showed that BrainNet-7 and EfficientNet-B2 are the models with clinical relevance.

Palavras-chave : CNN; classification; tumors; MRI.

        · resumo em Espanhol     · texto em Espanhol     · Espanhol ( pdf )