We derived the Cramer Rau bound for estimating volume changes from deformable image registration, which is a lower bound on the variance of an estimator. We demonstrated its implications for image registration performance in asymmetric methods versus symmetric methods, the former generally performing better.