This paper introduced the adaptive multi-scale entropy (AME) measures, in which the scales are adaptively derived from the data by virtue of recently developed empirical mode decomposition. By removing the low or high frequency components from the raw data, the AME can be estimated at either coarse-to-fine or fine-to-coarse scales, over which the sample entropy is performed. Simulations illustrate its effectiveness and promising application in brain death diagnosis to discern the states of the coma and the brain death.