Fig. 2.

Accuracy of DL-based segmentation in comparison to manual annotation. (a) The manual annotations of three experimenters and the DL-based segmentation are shown for three representative animals as white overlays on the first time frame of each of the image stacks. Besides minor variations, all outlines are visually very close and there is no apparent difference between the automated and manual segmentations. (b) The pairwise Dice coefficients show a high concordance between all the segmentations, where the lowest agreement was observed between two of the three experimenters (n = 21 for each pairwise comparison).