Negative controls reveal volume-driven confounding in radiomics and imaging foundation model features

2026-07-30

AI Summary

This study addresses the susceptibility of radiomic and foundation model–derived features to confounding factors such as tumor volume or acquisition artifacts, which often obscure true biological structure. To mitigate this, the authors propose the READII-2-ROQC framework, which introduces—for the first time—a volume-preserving negative control mechanism via controlled voxel perturbations that disrupt spatial organization without altering tumor volume, thereby enabling systematic assessment of feature dependence on genuine image patterns. Integrating PyRadiomics, foundation models, and multi-region perturbation strategies, the method was validated across three public cancer cohorts comprising 3,552 tumors. Results revealed that multiple published models exhibited no significant performance drop after structural disruption, indicating their reliance on volume or contextual confounders rather than biologically meaningful signals. This work substantially enhances the interpretability and reproducibility of imaging biomarkers.

Abstract

Radiomics and imaging foundation models promise non-invasive biomarkers of tumour biology, yet predictive signatures may reflect tumour volume or acquisition artifacts rather than meaningful image structure. We introduce READII-2-ROQC, an open-source framework that uses volume-preserving negative controls to assess whether radiomic and deep imaging features capture independent spatial signals. READII-2-ROQC generates voxel-perturbed images across tumour, background and whole-image regions using configurable randomization strategies, then compares feature behaviour and model performance between original and control images. Applied to three public cancer imaging cohorts, the framework processed 3,552 tumour volumes and extracted PyRadiomics and foundation-model features from original images and nine matched controls. Reproducing published survival and HPV-status signatures, we show that multiple models retain performance after spatial structure is destroyed, revealing volume-driven or contextual confounding, whereas others show perturbation-sensitive signal. READII-2-ROQC provides a scalable quality-control strategy for developing interpretable, biologically grounded imaging biomarkers and reproducible radiomics workflows.

Problem

• radiomics
• imaging foundation models
• confounding
• tumour volume
• biomarkers

Innovation

• negative controls
• radiomics
• imaging foundation models
• volume confounding
• feature robustness