From the Journals
Commentary & Perspectives

Are AI Models Cheating in Biomarker Predictions?

Study finds image-based predictions often reflect confounding factors rather than true molecular signals

  • By

  • March 6, 2026

  • 3 min

Share

A multi-cohort analysis involving 8,221 cancer patients assessed the efficacy of deep learning models in predicting cancer biomarkers from histology images. The study highlighted that while AI systems exhibited high accuracy, their performance was often influenced by unrelated factors such as tumor grade and mutation burden, rather than the biomarkers themselves. Published in Nature Biomedical Engineering, the findings emphasize the need for stricter evaluation protocols in AI development to distinguish genuine predictive capabilities from simpler grade-based correlations. This underscores the importance of molecular testing in treatment decisions.

Original Source(s)

Are AI Models Cheating in Biomarker Predictions?

Study finds image-based predictions often reflect confounding factors rather than true molecular signals

  • March 6, 2026

  • 3 min

Related Content

Sequencing Therapies in Non-Hodgkin Lymphomas

Novel Therapies and Multidisciplinary Strategies in Metastatic Triple-Positive Breast Cancer

United in Care at JADPRO Live 2025