Skip to main content
. Author manuscript; available in PMC: 2023 Nov 1.
Published in final edited form as: Med Image Anal. 2022 Sep 14;82:102615. doi: 10.1016/j.media.2022.102615

Table 6:

System-level comparison of various TransMorph designs and the customized VoxelMorph on the validation and test datasets of inter-patient MRI, atlas-to-patient MRI, and XCAT-to-CT registration tasks. ”Val. DSC” denotes the Dice scores on the validation dataset; “Test DSC” denotes the system-level comparison of the Dice scores on the test dataset. The bolded numbers denote the highest scores, while the italicized ones indicate the second highest.

Inter-patient MRI Atlas-to-patient MRI XCAT-to-CT
Model Val. DSC Test DSC Val. DSC Test DSC Val. DSC Test DSC
w/o conv. skip. 0.753±0.119 0.743±0.124 0.752±0.129 0.754±0.125 0.591±0.319 0.586±0.314
w/o Trans. skip. 0.740±0.124 0.727±0.130 0.734±0.127 0.736±0.125 0.578±0.315 0.588±0.314
w/ shuffling 0.755±0.119 0.744±0.125 0.751±0.127 0.754±0.123 0.588±0.314 0.597±0.310
w/ rel. positional bias 0.755±0.120 0.742±0.125 0.751 ±0.131 0.753±0.127 0.593±0.315 0.592±0.319
w/ lrn. positional embedding 0.755±0.120 0.744±0.125 0.749±0.131 0.751±0.129 0.594±0.315 0.586±0.315
w/ sin. positional embedding 0.755±0.120 0.744±0.125 0.752±0.126 0.754±0.123 0.583±0.320 0.572±0.317
TransMorph 0.756±0.119 0.745±0.125 0.753±0.127 0.754±0.124 0.600±0.317 0.604±0.314
TransMorph-tiny 0.710±0.132 0.696±0.140 0.545±0.180 0.543±0.180 0.502±0.311 0.501±0.312
TransMorph-small 0.751±0.121 0.740±0.126 0.746±0.128 0.747±0.125 0.572±0.320 0.570±0.318
TransMorph-large 0.757±0.119 0.746±0.124 0.753±0.130 0.754±0.128 0.608±0.305 0.611±0.311
VoxelMorph-huge 0.755±0.119 0.744±0.124 0.750±0.133 0.751±0.130 0.543±0.320 0.550±0.319