Ultrasound (US) images are the images which are mostly used by the radiologist to detect and diagnose the breast cancer. In this research work, we are proposing the deep learning approach for the classification problem in the United States. In medical image analysis, transfer learning and Convolution Neural Network (CNN) are widely used for the object identification. The fine-tuning techniques are aimed to update the weights of the pretrained networks but is difficult when the trainable parameters are large and the medical data are scarce. In this research work, we are proposing the concept where the deep representation scaling layers are added between the pretrained CNN blocks to process the images and that parameters are only updated during the training of the model to enable the good flow of information in the network. In the proposed research, the pretrained model used is Inception V3. The proposed method combined with the fine-tuning technique achieves excellent classification performance among the two classes benign and malignant achieving an accuracy of 92%.