Breakouts are collapsed zones on wellbore walls caused by compressive failure. Their identification is fundamental for estimating the borehole s stability and obtaining the direction and magnitude of the maximum horizontal stress in the rock formation. Traditionally, professional interpreters identify and characterize breakouts manually in image logs, which can be considered a very laborious and time-consuming task due to the massive size of the wellbore data. Other aspects that make the interpretation difficult are the complexity of the structures of interest and several noisy artifacts in the image log data. Therefore, more than traditional image processing methods are required to solve this detection task. In recent years, solutions based on deep learning have become increasingly promising for computer vision problems, such as object detection and image segmentation. This work explores using a convolutional neural network to create a pixel-by-pixel classification of the breakout regions in the image log data. The architecture model used in this work for the supervised training was the DC-UNet. This architecture is a variation of the classical U-Net, an acknowledged network for medical image segmentation. The proposed method reached an average F-Score of 72.3 percent and qualitative results with some prediction cases even better than ground truth. After evaluating the results, the work can be considered promising for automatically characterizing and segmenting borehole structures in well image logs.