Deep active learning for suggestive segmentation of biomedical image stacks via optimisation of Dice scores and traced boundary length

•A semi-automated deep learning segmentation method for image stacks (e.g., histology).•It reduces manual delineation time up to 60–70% without loss in accuracy.•It requests labels for 1 ROI on 1 image at the time rather than all ROIs for 1 image.•Traced boundary length is used as proxy for effort, accounting for shared boundaries.•It optimises the Dice score of the final result, rather than a proxy like the entropy.