Breast cancer is emphasized as one of the significant contributing factors to mortality among women worldwide, and early diagnosis is a major factor in increasing the survival rate. The conventional processing of images is a complex task that requires a high level of expertise, and this may result in inconsistencies in the results. In an effort to overcome these issues, this paper proposes a hybrid deep-learning Approach for the screening process in breast cancer from images in BUSI dataset. The proposed model integrates medical-aware preprocessing methods such as CLAHE, tissue segmentation for background removal, dataset-aware normalization, and data augmentation methods for improving the robustness and quality related to the images. Deep feature extraction is performed using a MobileNetV2 backbone network After the model has been trained on the image dataset, the features were reduced using averaging and fed into the dense layers. to extract a compact 256-dimensional feature vector. The extracted features are analyzed using an ensemble method that combines the Kernel Extreme Learning Machine (KELM) and the Deep Belief Network (DBN). The final results are obtained through the probabilistic combination of the two classifiers. The ensemble method takes advantage of the fast learning capability of KELM and the superior ability of DBN in modeling deep hierarchical structures, thereby improving the robustness and stability of the classification task. Experimental evaluation confirms that the developed hybrid framework achieves a classification accuracy of 96.15% in differentiating benign and malignant breast tumors. The proposed framework is computationally efficient and demonstrates strong potential for improving early breast cancer detection and supporting modern computer-aided diagnostic systems.