The early detection of breast cancer is one of the top critical areas of the health care of Moroccan women. The following steps were undertaken to perform this screening in the study. First, a database of mammographic images of Moroccan women, comprised of various breast cancer cases, was collected. Then, the data preparation and cleaning procedures were executed to arrange the interoperability and quality of the collected datasets, this involved meticulously labeling each image with corresponding annotations of breast lesions. Afterwards, the deep learning-based approach was applied to the classification of breast lesions as per the BIRADS scoring system. To this end, the Convolutional Neural Network model was built, which was specifically adapted to our dataset. The model was trained on the pre-processed data based on the annotations of the lesions; the data augmentation was utilized to improve the robustness and transfer ability of the model. Finally, we evaluated the performance of our model on a validation set, where we achieved an accuracy of 87.50%, measuring its ability to accurately classify breast lesions based on BIRADS categories. The results obtained demonstrate the feasibility and effectiveness of our approach in breast cancer screening among Moroccan women, despite the challenges posed by a small dataset, paving the way for future research and clinical applications in this vital area of public health.