Harmonization and the Worst Scanner Syndrome

If a predictive machine is made invariant to a set of domains, the accuracy of the output predictions (as measured by mutual information) is limited by the domain with the least amount of information to begin with. If a real label value is highly informative about the source domain, it cannot be accurately predicted by an invariant predictor. These results are simple and intuitive, but we believe that it is beneficial to state them for medical imaging harmonization. All images in medical imaging have a scanner-bias. We receive imagesX from e.g. an MRI machine, and we would like to predict a label Y, which may be a disease state, or a tumor location, or a tissue label map. The images are in part dependent on these labels, but are also dependent on the equipment that collected them, the scanner/site variables S. En masse, the biases in equipment (or inhomogeneity in the populations put into one piece of equipment versus another) can be a predictive of the labelsY in a static dataset.