“…For pixel-level boundaries, semantic segmentation approaches are ideal 4 , 6 , 7 but require laborious pixel-wise annotations and a larger amount of training data. 21 , 22 , 23 An alternative approach for providing this specificity may be to combine superpixels with deep learning 24 , 25 hoping to benefit from superpixels’ more nuanced detection of organ/tissue boundaries.…”