Breast cancer remains a leading cause of mortality among women worldwide, highlighting the critical need for early and accurate diagnosis. This paper presents a deep learning-based model designed for breast cancer classification and tumor segmentation using ultrasound images. The proposed model combines convolutional layers, residual blocks, squeeze-and-excitation modules, and spatial attention mechanisms to enhance feature extraction and focus on important regions of interest. The model achieves a classification accuracy of 98.95%, significantly outperforming state-of-the-art methods despite having only 0.0716 million parameters, making it highly computationally efficient. In addition to classifying images into benign, malignant, and typical cases, the model accurately segments tumor regions, providing critical insights for clinical decision-making. The lightweight architecture of the model makes it particularly well-suited for real-time applications in medical diagnostics. Future work will aim to extend the dataset and evaluate the model’s generalizability across diverse clinical environments and imaging systems.