Does this technique potentially allow training on smaller datasets? I am thinking in application with neuroimaging datasets, which are usually numbered in the h

Does this technique potentially allow training on smaller datasets? I am thinking in application with neuroimaging datasets, which are usually numbered in the hundreds.