Breast cancer detection and classification are significantly loaded by variation of imaging data and features. The combined methodology presented in this work combines Multiscale Image Analysis (MIA) with advanced Machine Learning (ML) methods to increase diagnostic accuracy. To reduce noise and increase image quality, a Recursive Least Squares-based Wiener Filter (RLS-WF) is applied in the pre-processing phase. For segmentation, the Attention U-Net (AU–N) technique is employed to precisely designate cancer regions and concentrate on appropriate areas. Texture-based properties are extracted using the Gray Level Co-occurrence Matrix (GLCM) to captures the critical textural information and supports in tissue type discrimination. To recognize the important features for classification, feature selection is optimized using the Grey Wolf Optimizer (GWO). Lastly, Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) with Long Short-Term Memory (LSTM) are used to detection and classification using both spatial and temporal patterns in the data. Thus the results illustrate the proposed model performs better than traditional methods in both detection and classification. This proposed models shows how the MIA and ML models can be used to improve breast cancer diagnosis.