Table 2 Main results.

From: Eliminating biasing signals in lung cancer images for prognosis predictions with deep learning

Model

Variables

\({\mathrm{MSE}}_{{\mathrm{y}}}\)

\({\mathrm{ATE}}\)

Regression

\(t\)

2.74

1.02

Regression

\(t,x^{\prime} ,z^{\prime}\)

1.39

0.65

Regression*

\(t,z^{\prime}\)

1.99

1.00

BiasedNet

\(t,image\)

1.83

0.66

CausalNet

\(t,{a}_{j}^{k}(j\,>\,1)\)

2.23

1.02

  1. Mean squared error for survival (MSEy) along with estimated average treatment effect (ATE). The linear regression metrics are the expected outcomes according to whether or not the model conditions on the collider \(x\). Regression* is the optimal value for our setup: (1) predicting the outcome based on relevant prognostic information from the image while (2) retaining a valid estimate of the treatment effect. All metrics were calculated on the validation set
Back to article page