The hierarchical setting in LapIRN is configured with 3 levels, which is essentially an "unfolding" operation. The authors only describe this in a retrospective manner without explaining why a 3-level deep Laplacian network was chosen. Does this ensure convergence and generalizability? In other words: Is a 3-level deep Laplacian network sufficient to handle all types of patient conditions or body parts? Can theoretical derivations and/or experimental evidence be provided? In short, the authors need to further elaborate on why the current 3-level deep Laplacian workflow was chosen. Merely describing it in a retrospective manner is far from sufficient. I suspect that considerations were made regarding the trade-off between runtime, accuracy, and efficiency. Could you elaborate on this? #37