Detecting and Monitoring Bias for Subgroups in Breast Cancer Detection AI

TL;DR

This paper evaluates the performance and bias of AI models in breast cancer detection across subgroups, highlighting underperforming groups and proposing a monitoring method to detect performance drifts over time.

Contribution

It introduces a systematic analysis of subgroup biases in mammography AI models and presents a monitoring approach for detecting performance drifts in real-world deployment.

Findings

Certain subgroups show notable underperformance

The monitoring method effectively detects performance drifts

Ongoing subgroup performance monitoring is essential

Abstract

Automated mammography screening plays an important role in early breast cancer detection. However, current machine learning models, developed on some training datasets, may exhibit performance degradation and bias when deployed in real-world settings. In this paper, we analyze the performance of high-performing AI models on two mammography datasets-the Emory Breast Imaging Dataset (EMBED) and the RSNA 2022 challenge dataset. Specifically, we evaluate how these models perform across different subgroups, defined by six attributes, to detect potential biases using a range of classification metrics. Our analysis identifies certain subgroups that demonstrate notable underperformance, highlighting the need for ongoing monitoring of these subgroups' performance. To address this, we adopt a monitoring method designed to detect performance drifts over time. Upon identifying a drift, this method issues an alert, which can enable timely interventions. This approach not only provides a tool for tracking the performance but also helps ensure that AI models continue to perform effectively across diverse populations.