Table 3.

Performance comparison of different model configurations using the 3D ResNet18 backbone. Best BATE values are highlighted in bold

ModelPredictive accuracyCausal effect bias (BATE) ()
MAEMSEChemoRT versus RTChemoRT versus RT+EGFRIRT versus RT+EGFRI
Baseline (concatenation)1.81 ± 0.095.35 ± 0.281.89 ± 0.170.65 ± 0.101.70 ± 0.20
Baseline + Bi-AdaIN1.76 ± 0.084.22 ± 0.261.85 ± 0.160.53 ± 0.091.66 ± 0.19
Baseline + adversarial1.84 ± 0.105.41 ± 0.311.59 ± 0.200.38 ± 0.121.42 ± 0.22
Bi-AdaIN + adversarial1.69 ± 0.094.30 ± 0.291.15 ± 0.170.28 ± 0.100.95 ± 0.18
Baseline + MI1.86 ± 0.105.44 ± 0.320.48 ± 0.150.23 ± 0.090.46 ± 0.16
Bi-AdaIN + MI (proposed)1.68 ± 0.084.25 ± 0.270.21±0.070.14±0.060.18±0.08
Note(s):

The BATE values involving the RT+EGFRI treatment group should be interpreted with caution due to the smaller sample size (n = 72) in this cohort compared to the ChemoRT (n = 1,413) and RT (n = 1,861) groups

or Create an Account

Close Modal
Close Modal