Confounding factors and biases abound when predicting molecular biomarkers from histological images - PubMed
4 hours ago
- #deep learning
- #biomarkers
- #histopathology
- Deep learning models for predicting molecular biomarkers from tissue images are being explored as alternatives to molecular testing.
- Datasets used for training these models have dependencies between biomarkers and clinicopathological features, leading to confounded signals.
- Prediction accuracy varies with the status of codependent biomarkers and clinicopathological variables.
- Current approaches are not yet suitable substitutes for molecular testing but can support triage or complementary decision-making.
- Unconfounded biomarker prediction requires models that learn causal rather than correlational relationships.