Figure 4 | Scientific Reports

Figure 4

From: Training calibration-based counterfactual explainers for deep learning models in medical image analysis

Figure 4

Using TraCE to infer relationships between a patient attribute (e.g., age) and disease states. For this analysis, we construct two independent predictive models, i.e., age and diagnosis state, and synthesize counterfactuals based on hypothesis on each of the predictions (e.g., patient age should be predicted as 70 while the diagnosis state should be abnormal. Finally, we combine the changes induced in the two counterfactuals, \(\Delta _A({\text {x}})\) and \(\Delta _D({\text {x}})\) respectively, and check if incorporating age-specific patterns strengthens the evidence for the abnormal class.

Back to article page