Mammography is the primary imaging modality used for early detection and diagnosis of breast cancer. Mammography analysis mainly refers to the extraction of regions of interest around tumors, followed by a segmentation step, which is essential to further classification of benign or malignant tumors. Breast masses are the most important findings among breast abnormalities. However, manual delineation of masses from native mammogram is a time consuming and error-prone task. An integrated computer-aided diagnosis system to assist radiologists in automatically detecting and segmenting breast masses is therefore in urgent need. We propose a fully-automated approach that guides accurate mass segmentation from full mammograms at high resolution through a detection stage. First, mass detection is performed by an efficient deep learning approach, You-Only-Look-Once, extended by integrating multi-scale predictions to improve automatic candidate selection. Second, a convolutional encoder-decoder network using nested and dense skip connections is employed to fine-delineate candidate masses. Unlike most previous studies based on segmentation from regions, our framework handles mass segmentation from native full mammograms without user intervention. Trained on INbreast and DDSM-CBIS public datasets, the pipeline achieves an overall average Dice of 80.44% on high-resolution INbreast test images, outperforming state-of-the-art methods. Our system shows promising accuracy as an automatic full-image mass segmentation system. The comprehensive evaluation provided for both detection and segmentation stages reveals strong robustness to the diversity of size, shape and appearance of breast masses, towards better computer-aided diagnosis.