When normalization hallucinates: unseen risks in AI-powered whole slide image processing
This work addresses a critical safety issue in computational pathology for clinical deployment, highlighting an underappreciated risk that could lead to misdiagnosis.
The paper tackles the problem of hallucinated content in AI-powered whole slide image normalization, which can mask important features and introduce realistic-looking artifacts not present in the original tissue, and demonstrates that this risk is significant in real-world clinical data, revealing inconsistencies and failures in existing methods through a novel detection measure.
Whole slide image (WSI) normalization remains a vital preprocessing step in computational pathology. Increasingly driven by deep learning, these models learn to approximate data distributions from training examples. This often results in outputs that gravitate toward the average, potentially masking diagnostically important features. More critically, they can introduce hallucinated content, artifacts that appear realistic but are not present in the original tissue, posing a serious threat to downstream analysis. These hallucinations are nearly impossible to detect visually, and current evaluation practices often overlook them. In this work, we demonstrate that the risk of hallucinations is real and underappreciated. While many methods perform adequately on public datasets, we observe a concerning frequency of hallucinations when these same models are retrained and evaluated on real-world clinical data. To address this, we propose a novel image comparison measure designed to automatically detect hallucinations in normalized outputs. Using this measure, we systematically evaluate several well-cited normalization methods retrained on real-world data, revealing significant inconsistencies and failures that are not captured by conventional metrics. Our findings underscore the need for more robust, interpretable normalization techniques and stricter validation protocols in clinical deployment.