Fig. 1From: Residual refinement for interactive skin lesion segmentationThe workflow of our method. First, a box is drawn by the user who possesses domain knowledge to delineate the region of interest (ROI). The ROI is then cropped out to exclude most of the useless background and fed into a fully convolutional neural network to generate a segmentation. Most of the time, this segmentation will be very satisfactory. In addition, no further process is required. However, not all situations are ideal. Take the one in the figure as an example. There is a blot that we would like to remove. To address this type of situation, the user can provide another FCN with positive and negative clicks, which are transformed into Gaussian centred distance maps, to further refine the segmentationBack to article page