This work presents a proposal for localization and delimitation of suspicious masses in breast ultrasound images by selecting regions of interest. Since the presence of speckle noise, which is a characteristic type of ultrasound images, to delimit a lesion becomes a difficult task. Some techniques of digital processing were applied in 50 clinical images of breast ultrasound, in order to minimize the noise allowing posterior extraction of their contour. The segmentation technique used to extract the contour was performed by means of the artificial neural network Self-Organizing Map. Then three evaluation metrics were used to compare the proximity of the automatically obtained area with the manual outline of a radiologist, resulting in percent values of 91% to accuracy, 67% to sensitivity, and 97% to positive predictive value, approximately.