Hasty Briefsbeta

Bilingual

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.