Breast cancer, as the most commonly diagnosed type of cancer, poses a significant global health challenge. Early detection, in which mammography plays a crucial role, is essential for reducing mortality rates. However, distinguishing between benign and malignant lesions, which visually differ in contours on mammographic images, presents a substantial challenge. To standardize this process, the BI-RADS system for categorizing lesions was introduced. Despite the longstanding use of computer-aided diagnostics (CAD) in the analysis of mammographic images, the results achieved are still not at the expected level. This is where the application of deep learning, a segment of machine learning based on artificial neural networks, comes into play. This technology has the potential to significantly ease and enhance the work of radiologists. In the scope of this thesis, a deep learning model intended for the segmentation of breast lesions was developed. This model analyzes each individual pixel in the image and classifies it into one of three categories: background, benign lesion, or malignant lesion. Although the model was trained on a relatively limited data set, the results achieved are promising - it successfully segments lesions regardless of their size and achieves a significant result according to the Dice coefficient. Modern technologies, such as deep learning, are opening new doors in the field of mammography. However, it is necessary to continue with further clinical research to fully uncover their potential. This thesis contributes to this goal, exploring the possibilities offered by the intersection of healthcare and technology.