Comparative Analysis of CNN, ResNet50, and Vision Transformer Architectures for Brain Tumor Classification from MRI Images

Matthieu Kayembe (Université de Kinshasa)
Franklin Mwamba (Institut de Recherche en Sciences de la Santé)
Pierre Kafunda (Université de Kinshasa)
Fiston Oshasha (General Commissariat for Atomic Energy, Regional Center for Nuclear Studies of Kinshasa)
John Poma (Université de Kinshasa)

JOURNAL OF APPLIED INFORMATICS AND COMPUTING
Vol. 10 No. 3 (2026): June 2026

The classification of brain tumors from Magnetic Resonance Imaging (MRI) is a crucial task in computer-aided medical diagnosis. Recent advances in deep learning have significantly improved performance in this domain. In this work, a comparative analysis of three architectures is conducted: a Convolutional Neural Network (CNN) trained from scratch, a transfer learning-based model using ResNet50, and a Vision Transformer (ViT). The models are evaluated on a multi-class dataset containing four categories: glioma, meningioma, pituitary tumor, and no tumor. Experimental results show that the CNN achieves limited performance with moderate generalization capability. The ResNet50 model reaches high accuracy during training but suffers from severe overfitting, leading to a significant drop in performance on the test set. In contrast, the Vision Transformer achieves the best overall performance, with a test accuracy of 0.76 and a good balance between precision and recall. These results highlight the effectiveness of Transformer-based architectures for complex medical image classification tasks.