US EditionStudy Warns AI Cancer Tools Rely on Shortcuts, Not Biology
University of Warwick study finds AI cancer pathology models often rely on correlated features, with accuracy dropping when confounding factors are controlled, based on over 8,000 patient samples.
- This past week, University of Warwick researchers warned many deep-learning pathology systems rely on visual shortcuts rather than true biological signals in a study published in Nature Biomedical Engineering.
- Analysis showed that correlated tissue features cause models to rely on biomarker correlations instead of isolating causal signals, making statistical shortcuts collapse when confounders are controlled.
- Using more than 8,000 patient samples across four cancer types, the team found AI achieved just over 80% accuracy versus around 75% using tumour grade alone.
- Researchers warn that premature adoption risks inappropriate therapies as AI tools confuse correlated signals; current models should not replace molecular testing and need stronger evaluation before routine deployment.
- The authors propose moving beyond correlation-based learning toward causal, biology-aware models and call for rigorous, bias-aware evaluation, Professor Nasir Rajpoot said.
9 Articles
9 Articles
Confounding factors and biases abound when predicting molecular biomarkers from histological images - Nature Biomedical Engineering
Deep learning models that infer clinically relevant biomarker status from tissue images are being explored as rapid and low-cost alternatives to molecular testing. Here we show, through statistical analysis across multiple cancer types, datasets and modelling approaches, that the datasets used to train these models contain strong dependencies between biomarkers and clinicopathological features, which prevent models from isolating the effect of a…
Faster cancer screening? New AI system offers a better way to detect abnormal cells
One way cancer specialists detect the disease is by examining cells and bodily fluids under a microscope, a time-consuming and labor-intensive process called cytology. It involves visually inspecting tens of thousands to one million cells per slide for subtle 3D morphological changes that might signal the onset of cancer. But AI offers an approach that is potentially faster and more accurate.
Artificial intelligence in cancer pathology: Applications, challenges, and future directions
The application of artificial intelligence (AI) in cancer pathology has shown significant potential to enhance diagnostic accuracy, streamline workflows, and support precision oncology. This review examines the current applications of AI across ...
Coverage Details
Bias Distribution
- 100% of the sources are Center
Factuality
To view factuality data please Upgrade to Premium
