This study introduces an enhanced VGG-19 model designed to improve the accuracy of brain tumor detection from MR images. The model enhancement involves optimizing convolutional neural network (CNN) layers, fully connected layers, and layer weights to achieve superior performance. Additionally, preprocessing techniques are applied to the datasets to further refine the model’s predictive capabilities. A modified Stochastic Gradient Descent (SGD) optimizer is proposed to enhance training efficiency and model accuracy. This methodology is evaluated on two distinct datasets: the SARTAJ dataset Br35H and the Brain MRI Images from the Brain Tumor Detection NAVONEEL CHAKRABARTY dataset on Kaggle. The proposed enhanced VGG-19 model achieved remarkable classification accuracies of 98.41 and 90.20% on these datasets, respectively. These results demonstrate the effectiveness of the improved model and preprocessing techniques in accurately detecting brain tumors from MR images.