Hidden Stratification Causes Clinically Meaningful Failures in Machine Learning for Medical Imaging

TL;DR

This paper investigates hidden stratification in medical imaging ML models, revealing it causes significant performance gaps in important but overlooked subgroups, and emphasizes the need for better evaluation methods.

Contribution

It introduces methods to measure hidden stratification effects and demonstrates their impact across multiple datasets, highlighting critical evaluation gaps.

Findings

Hidden stratification occurs in low-prevalence and subtle feature subsets.

Performance differences over 20% are observed in important subgroups.

Evaluation of hidden stratification is essential for clinical deployment.

Abstract

Machine learning models for medical image analysis often suffer from poor performance on important subsets of a population that are not identified during training or testing. For example, overall performance of a cancer detection model may be high, but the model still consistently misses a rare but aggressive cancer subtype. We refer to this problem as hidden stratification, and observe that it results from incompletely describing the meaningful variation in a dataset. While hidden stratification can substantially reduce the clinical efficacy of machine learning models, its effects remain difficult to measure. In this work, we assess the utility of several possible techniques for measuring and describing hidden stratification effects, and characterize these effects on multiple medical imaging datasets. We find evidence that hidden stratification can occur in unidentified imaging subsets with low prevalence, low label quality, subtle distinguishing features, or spurious correlates, and that it can result in relative performance differences of over 20% on clinically important subsets. Finally, we explore the clinical implications of our findings, and suggest that evaluation of hidden stratification should be a critical component of any machine learning deployment in medical imaging.