Breast cancer is a major global health burden and remains one of the leading causes of mortality among women. Accurate and early detection significantly improves clinical outcomes, especially in settings where ultrasound is the primary diagnostic tool. This paper presents a deep learning approach for the automatic segmentation of breast tumors in ultrasound images based on a U-Net++ architecture. The proposed model integrates multiple convolutional neural networks to improve feature representation and segmentation accuracy. A dataset consisting of 780 ultrasound images from 600 patients was used for training and evaluation, following consistent preprocessing and normalization procedures. The performance of the model was benchmarked against established architectures such as UNet, SegNet, and ResNet. The results indicate that the proposed method outperforms the baseline models in the delineation and generalization of tumor boundaries, offering a promising solution for computer-assisted diagnosis of breast cancer in ultrasound images.