<p>Currently, brain tumors are among the deadliest diseases. After heart disease, cancer ranks as the second leading cause of death worldwide. The field of artificial intelligence focuses on various problems, especially those related to the health of humans, animals, and plants. These techniques are crucial for accurately diagnosing different diseases and play a significant role in medical diagnosis. In this work, we address brain tumor detection and classification using an advanced artificial intelligence technique called deep transfer learning. Brain tumors are some of the most lethal diseases globally, posing significant challenges in diagnosis because traditional methods are time-consuming and prone to errors. This underscores the need for improved detection methods to enhance patient outcomes. After reviewing several papers, we found that many models underperform in terms of accuracy and efficiency. Additionally, these models often struggle to detect small lesion patterns in the brain, affecting their ability to predict and classify accurately. This research introduces a model that combines the best features of EfficientNetV2B3 and ResNetV2 to optimize performance. The addition of custom layers further improves the model’s capabilities with appropriate labels. The dataset used combines three datasets, creating a larger dataset with three classes of brain tumors—glioma, meningioma, and pituitary—along with MR images of healthy individuals without tumors. By applying data augmentation, normalization, and cross-validation, the model reduces the risk of overfitting. It achieved a remarkable accuracy of over 99% during training and nearly 100% during testing, demonstrating its robustness and effectiveness. The model was also tested on an external dataset and performed well. We evaluated its performance using various metrics and found it suitable across almost all parameters.</p>