Fig. 1
From: Nuclei instance segmentation from histopathology images using Bayesian dropout based deep learning
The BayesNuSeg model consists of one encoder and two independent decoders. The seed branch decoder predicts k seed maps for each class label. The instance branch decoder computes the offset vectors in both x and y dimensions, which are further added to the coordinate maps along the corresponding axes to obtain pixel embedding and mean of the instances (sigma). The seed maps, sigma, and pixels embedding are clustered using sequential clustering approach, which involves grouping similar pixels together based on their feature representation, to segment the nuclei by sampling the pixel embedding with the highest seed margin and using that coordinate location as instance center \({\textbf {C}}_k\). The output is the predicted nuclei segmentation and the model uncertainty quantification