Recent advances in artificial intelligence, especially in the field of deep learning, have revolutionized the automatic and accurate analysis of image data in the medical field. Among them, the detection and classification of brain tumors using magnetic resonance images is one of the most interesting research areas due to its high sensitivity and its critical role in timely treatment. This research designs, implements, and deeply evaluates a model based on a convolutional neural network architecture that uses a differential layer as its central core. This model is used to perform two independent and important classifications, binary classification (initial diagnosis and assessment of the presence or absence of a tumor) and multiclass classification (distinguishing and distinguishing between four different types of brain tumors). The main goal of this research is not only to present a new model, but also to validate the claim of the basic paper [1] about the positive effect of differential filters on classification accuracy, as well as to investigate the efficiency of this approach at different levels of problem complexity. The differential layer, as an internal and trainable preprocessing stage, is placed at the beginning of the network architecture to provide a better and richer understanding of the input data to the subsequent layers, which are the RI of the network, by effectively extracting edge features and high-contrast regions. The proposed model is trained, validated, and tested on a large and dedicated dataset consisting of more than 3000 images including cases with and without tumors as well as examples from four tumor subtypes, and is implemented using deep learning frameworks such as Google Ⅽoⅼlab. All the model development and evaluation processes are performed in the context. The key findings of this research are reported quantitatively and statistically. The results clearly show that the integration of the differential layer leads to the highest accur