Breast cancer is one of the most life-threatening diseases for women around the world. In 2022, breast cancer caused 670,000 deaths globally in 2022. Early detection of this disease is extremely important to avoid death. For diagnosing this image ultrasound is a safe approach since ultrasound doesn’t have any risk of radiation. That is why in this study we have decided to use ultrasound image dataset. In this study, we will utilize different deep learning models to classify breast cancer from ultrasound images. There are 3 classes in the dataset that have been used; benign, malignant, and Normal. The dataset has 780 images along with their mask. DenseNet201 and ResNet50 these two deep learning models were employed in this research. The models were trained using two different methods and for both methods dataset was split into train 80%, validation 10% and test 10%. At first, the models were trained using only the images. In second method, those masks that were in the dataset were overlaid with images. The models were then trained with those overlaid images. Hyperparameter tuning was done for each method. Data augmentation was done when training the models for both techniques as well. Models’ performance when training with only raw images was not up to expectations and the accuracy was around only 80%. But when trained with overlaid images their performance becomes a lot better. Both models achieved an accuracy of 97% accuracy when trained with overlayed images. The research perfectly portrays the effectiveness of manually created masks.