create pixel Label Datastore from categorical tensor

Is it possible to create a pixel Label Datastore from a categorical tensor?

Input: 160x160x25 categorical (labels for 25 images with 160x160 pixels each)

desired Output: pixelLabelDatastore

The function read(pxds) offers this functionality, but unfortunately only in the other direction.

I need the pixelLabelDatastore for semanticImageSegmentation, given the images as 160x160x5x25 double (25 images with 5 channels and 160x160 pixels each) and the labels in the format given above.