Fig. 6

Visual assessment of the nuclei segmentation of the BayesNuSeg with uncertainty and the baseline models. For the visualization purposes, we combined all the nuclei cell types and showed the nuclei boundaries as a contour. The first column is the original images, and the next columns represent the predicted nuclei segmentation overlaid on the original image by the models. The red contours represent the ground truth annotations provided by the expert pathologist of the PanNuke dataset, whereas the blue contours indicate the nuclei segmentation as predicted by the proposed and other baseline approaches. We annotated the predictions of the baseline approaches using green, orange, and cyan circles (with thick contours). The green circle indicates that FCN8 and U-Net have failed to accurately estimate the nuclei boundaries. The orange circle highlights the noise present throughout the image due to the predictions of the SegNet approach. The cyan circle indicates the overestimation of predicted nuclei boundaries by the Hovernet approach