Performance comparison of different model configurations using the 3D ResNet18 backbone. Best BATE values are highlighted in bold
| Model | Predictive accuracy | Causal effect bias (BATE) () | |||
|---|---|---|---|---|---|
| MAE | MSE | ChemoRT versus RT | ChemoRT versus RT+EGFRI | RT versus RT+EGFRI | |
| Baseline (concatenation) | 1.81 0.09 | 5.35 0.28 | 1.89 0.17 | 0.65 0.10 | 1.70 0.20 |
| Baseline + Bi-AdaIN | 1.76 0.08 | 4.22 0.26 | 1.85 0.16 | 0.53 0.09 | 1.66 0.19 |
| Baseline + adversarial | 1.84 0.10 | 5.41 0.31 | 1.59 0.20 | 0.38 0.12 | 1.42 0.22 |
| Bi-AdaIN + adversarial | 1.69 0.09 | 4.30 0.29 | 1.15 0.17 | 0.28 0.10 | 0.95 0.18 |
| Baseline + MI | 1.86 0.10 | 5.44 0.32 | 0.48 0.15 | 0.23 0.09 | 0.46 0.16 |
| Bi-AdaIN + MI (proposed) | 1.68 0.08 | 4.25 0.27 | 0.210.07 | 0.140.06 | 0.180.08 |
| Model | Predictive accuracy | Causal effect bias ( | |||
|---|---|---|---|---|---|
| ChemoRT versus | ChemoRT versus RT+EGFRI | ||||
| Baseline (concatenation) | 1.81 | 5.35 | 1.89 | 0.65 | 1.70 |
| Baseline + Bi-AdaIN | 1.76 | 4.22 | 1.85 | 0.53 | 1.66 |
| Baseline + adversarial | 1.84 | 5.41 | 1.59 | 0.38 | 1.42 |
| Bi-AdaIN + adversarial | 1.69 | 4.30 | 1.15 | 0.28 | 0.95 |
| Baseline + | 1.86 | 5.44 | 0.48 | 0.23 | 0.46 |
| 1.68 | 4.25 | ||||
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