Developing an image-domain transformation technique for adapting deep learning algorithms: preliminary work using simulated tomosynthesis of breast patches