How can we use a weighting map for deep learning pixel classification ?

I am training a UNet convolutional network in matlab. In the UNet original paper (https://arxiv.org/pdf/1505.04597.pdf), they manage to modify the crossentropy function in order to give to each pixel a certain weight (by introducing the so-called weighting map, bottom of page 4 and page 5 in the paper).

In matlab we have the possibility to weight classes when training a neural network. Once I have defined a certain weighting coefficient for each pixel in a matrix, is it possible to 'input' this matrix in the network architecture to get individual weighted pixels ?