BenchX: Benchmarking AI Models for Cancer Detection and Localization with Demographic and Protocol Biases

2026-06-23

AI Summary

Current AI models for tumor detection exhibit inconsistent performance across variations in patient demographics and imaging protocols, with notably degraded accuracy in underrepresented subgroups such as younger individuals, women, and African Americans. To address this, this study introduces a large-scale open benchmark comprising 85,355 CT scans and systematically quantifies performance disparities across twelve state-of-the-art models along multiple dimensions—including tumor size and location, patient subgroups, and imaging protocols. The work innovatively leverages large language models to automatically extract demographic and clinical subgroup information from unstructured clinical text, enabling scalable and reproducible evaluation. By establishing an evaluation paradigm that accounts for real-world diversity, this research provides foundational data and methods to foster the development of fairer, more robust clinical AI systems.

Abstract

Artificial intelligence (AI) has achieved remarkable success in medical imaging, but it is widely recognized that these models often perform inconsistently across real-world clinical settings. Such inconsistencies occur when patient demographics and imaging protocols vary, for example, in detecting small tumors, analyzing scans from different contrast phases, or evaluating patients of different ages or sexes. To quantify these inconsistencies, we develop a large-scale, open benchmark of 85,355 CT scans that systematically evaluates 12 tumor-detection AI models across tumor size, location, patient subgroup, and imaging protocol. We leverage large language models (LLMs) to extract and organize subgroup information from clinical data, which makes the analysis both scalable and reproducible. Our benchmark reveals that current state-of-the-art AI models, optimized for average accuracy, perform poorly in rare or underrepresented subgroups, such as young, female African Americans. However, collecting sufficient annotated data for these rare cases is often impractical. The benchmark provides a foundation for building more reliable and robust AI models for tumor detection and highlighting the need for rigorous, subgroup-level evaluation in medical imaging and computer vision. Datasets, code

Problem

• AI bias
• cancer detection
• demographic disparity
• medical imaging
• subgroup evaluation

Innovation

• benchmarking
• demographic bias
• large language models
• medical imaging
• subgroup evaluation