Although precise identification of breast cancer remains difficult, deep learning (DL) may help with more accurate picture interpretation. In this work, we create a highly robust DL model for categorising benign and malignant breast cancers using combined B-mode and strain elastography ultrasound (SE) images. This research comprised 85 individuals, 42 with benign lesions and 43 with cancers, all of which were verified by biopsy. AlexNet and ResNet, two deep neural network models, were trained independently using 205 B-mode and 205 SE pictures (80% for training and 20% for validation) from 67 patients with benign and malignant lesions. These two models were then built to function as an ensemble on a dataset of 56 photos from the other 18 patients, and tested on a dataset of 56 images from the remaining 18 patients. To distinguish benign from malignant tumours, the ensemble model collects the different properties found in the B-mode and SE pictures and also includes semantic information from the AlexNet and ResNet models. The experimental findings show that the proposed ensemble model has a 90% accuracy, which is higher than the individual models and the model trained just on B-mode or SE pictures. Furthermore, the suggested ensemble technique accurately categorised certain patients who were misclassified by existing methods. Because of improved classification accuracy for breast tumours in ultrasound (US) pictures, the proposed ensemble DL model would help radiologists to attain greater detection efficiency.