Systems and methods for image segmentation

A method of training a machine learning model for performing image segmentation, comprising:

• obtaining at least one image comprising a plurality of pixels arranged in a plurality of dimensions;
• determining a ground truth segmentation of the at least one image, the ground truth segmentation dividing the at least one image into a background, at least one instance, and a basin corresponding to the at least one instance;
• assigning to each of the plurality of pixels of the at least one instance a plurality of amplitudes, each of the plurality of amplitudes for each of the plurality of pixels corresponding to exactly one of the plurality of dimensions, wherein
• those of the plurality of pixels within the at least one instance are assigned amplitudes indicating a direction to the basin.

A method comprising:

• receiving at least one image of a sample stained with a nuclear stain, wherein the at least one image comprises one or more nuclei;
• and
• providing the at least one image to a pretrained machine learning model, and receiving therefrom a pixel classification map,
• wherein the pixel classification map comprises:
• a plurality of stationary pixels,
• a plurality of flow pixels around the plurality of stationary pixels, wherein each flow pixel of the plurality of flow pixels is assigned a direction of a plurality of directions, and
• a plurality of background pixels.