RadEdit: stress-testing biomedical vision models via diffusion image editing

TL;DR

RadEdit uses diffusion-based image editing with multiple masks to simulate dataset shifts in biomedical images, helping diagnose model failures and robustness without extra data, thus improving deployment safety.

Contribution

The paper introduces RadEdit, a novel diffusion-based editing method that constrains changes using multiple masks to simulate realistic dataset shifts in biomedical imaging.

Findings

RadEdit effectively diagnoses model failures across different dataset shifts.

It quantifies model robustness without additional data collection.

The approach complements existing explainable AI tools.

Abstract

Biomedical imaging datasets are often small and biased, meaning that real-world performance of predictive models can be substantially lower than expected from internal testing. This work proposes using generative image editing to simulate dataset shifts and diagnose failure modes of biomedical vision models; this can be used in advance of deployment to assess readiness, potentially reducing cost and patient harm. Existing editing methods can produce undesirable changes, with spurious correlations learned due to the co-occurrence of disease and treatment interventions, limiting practical applicability. To address this, we train a text-to-image diffusion model on multiple chest X-ray datasets and introduce a new editing method RadEdit that uses multiple masks, if present, to constrain changes and ensure consistency in the edited images. We consider three types of dataset shifts: acquisition shift, manifestation shift, and population shift, and demonstrate that our approach can diagnose failures and quantify model robustness without additional data collection, complementing more qualitative tools for explainable AI.