The Multispectral Semantic Segmentation example has not fully explained how to train UNet with RGB image for one class for example Skin Cancer or leaves

Is there any way to train and test the UNet on RGB images.' If yes, how to create the dataset that will work to this model. I have the images with their masks. I was able to generate the image data set which merely uses 'imageDatastore' but how to turn the mask images to labels? Any help would be greatly appreciated!