The straight edges are not present in the input data, so it must be from internal tiling. Intensity shown is HH3 + panCK channel. In our benchmarking project, we find a greater number of cell masks with long straight edges than what competitors have.
Vertical yellow line shows where I think stitching boundary must have been. Also, other cells that are missed are surprising.