Describir: Using weakly annotated data for medical image segmentation based on pre-trained deep neural networks