Breast cancer is considered one of the most common cancer in women worldwide, reducing the risk of mortality is by early detection of cancer and providing the appropriate treatment. Working on mammography using machine learning can help detecting and diagnosing cancer. Convolutional neural network (CNN) is a deep learning technique which is one of the best for detection and classification of abnormalities in medical images. Training CNN models need large datasets so training it on medical datasets from scratch is difficult job because of their small size and the variation of the abnormality shapes in mammogram images, instead it is possible to use pre-trained network because the generic features from such model can be used for new different task like mammogram image dataset that has limited number of images. This research employed transfer learning by enhancing Local Binary Convolutional Neural Network (LBCNN) model which is deep learning technique to suit our new task and help classifying breast abnormalities in mammogram images by applying it on MIAS database. This includes image preprocessing, feature extraction by optimizing and fine-tuning number of convolutional layers and Finally image classification using new classifier with three classes normal, malignant and benign. The results proved that the enhanced model successfully predicted the presence of normality or not and its type with accuracy reached 90% in 15 minutes using about 1.2 million parameters which saves memory and time compared to 88.7% accuracy in 28 minutes using about 11.3 million parameters on standard CNN, this lead to conclude that the system capable of detecting and classifying mammogram images saving a lot of parameters.