Deep learning no reconhecimento de câncer cerebral em ressonâncias magnéticas

Abstract

Machine learning has proven to be a promising tool for the early diagnosis and effective treatment of diseases through the analysis of medical images. In this study, we explored the application of the Xception model to the classification of brain tumors into three categories: meningioma, glioma and pituitary gland tumor. Using weights pre-trained in ImageNet, the model achieved a validation accuracy of 97%, demonstrating consistency across different classes. Complementary metrics such as recall, F1-score and precision were used to evaluate performance. We also tried to implement the Grad-CAM technique to interpret the model's decisions. However, technical difficulties prevented the consistent generation of heat maps, highlighting the need for adjustments in future work. The results reinforce the potential of deep learning-based models to support the clinical analysis of brain images.