If you are asking if the algorithm can be generalized from 1D to 2D or even higher, then the answer is yes. For example, one could take the BWT of all rows and then all columns (using a different terminal symbol for each dimension), and that would be reversible.
Question is if that is useful for compression.