The U-Net deep learning architecture is used in this study's ultrasound image-based breast cancer detection system to achieve accurate picture segmentation. The first step in the process is picture collecting and pre- processing, where methods like data augmentation, normalization, and noise reduction improve image quality and increase the generalizability of the model. The encoder-decoder structure and skip connections of the U-Net architecture allow for the precise and effective localization of malignant areas. The U-Net model provides high-precision segmentation by reducing false positives and collecting contextual information as well as fine characteristics. A dice loss function is used to optimize the system, guaranteeing strong model performance in locating and classifying breast cancer areas. The U-Net model's promise for clinical applications in breast cancer detection is highlighted by experimental results that demonstrate its ability to detect and segment malignant regions in ultrasound images with a decrease in training loss and an improvement in accuracy with time.