The early detection of brain tumors is crucial for improving patient survival rates, presenting a bottleneck and potential technical improvements in medical imaging technologies. In this article, we propose a hybrid deep learning method for automatically detecting brain tumors from MRI images, integrating pixel intensity and texture-based features that might help increase performance accuracy. The methodology is organized into four stages: preprocessing, feature transformation, dual ResNet50-based feature extraction, and weighted feature fusion for classification. Preprocessing steps, including resizing, normalization, and Local Binary Pattern (LBP) transformation, enhance image data quality and emphasize texture patterns critical for distinguishing tumor characteristics. Two separate ResNet50 models are employed to capture critical features, one for the original image dataset and the other for the LBP-transformed dataset. These extracted features are fused through a weighted mechanism, enabling the model to learn from the pixel- and texture-based information. The final classification phase achieved an accuracy of 99.68%, significantly surpassing the traditional CNN architectures and benchmark models in the literature. The model’s high precision, recall, and ROC AUC scores underscore its potential reliability in clinical applications.