Deep learning has shown great promise in medical image analysis, particularly for automating diagnostic support. Breast cancer remains one of the most prevalent cancers among women, and the early detection of calcifications in mammography images is essential. A key challenge is that calcifications are very small objects, frequently overlooked when images are resized for deep learning models. This raises the research question: How can mammography images be preprocessed to preserve calcification details for accurate use in detection models? We propose a methodology that includes image division into overlapping patches, patch classification to isolate breast tissue, image reconstruction, breast edge dilation, and segmentation. This preprocessing step ensures the preservation of diagnostic details while generating optimized inputs for deep learning models. Experimental evaluation with the CBIS-DDSM dataset shows strong performance in patch classification, with the VGG16 model achieving precision of 0.979, recall of 0.915, and accuracy of 0.942. The results demonstrate that the proposed preprocessing pipeline effectively prepares mammography images for subsequent small-object detection tasks such as calcification identification.