AdverX-Ray: Ensuring X-Ray Integrity Through Frequency-Sensitive Adversarial VAEs

TL;DR

AdverX-Ray addresses covariate shift in X-ray imaging by providing a lightweight, real-time image quality assessment layer for CAD pipelines. It uses an adversarial VAE where the discriminator learns a boundary between ID and OOD by training on reconstructions and adversarially generated patches, leveraging a patch-based strategy and BN statistics to capture multi-frequency information. The method achieves high AUROC on Philips X-ray data (near 100% across modes) and strong performance on BIMCV-COVID19+ data (average in the mid-90s), outperforming VAE and GLOW baselines while maintaining a small, fast model. The work extends DisCoPatch to X-ray, provides new covariate-shift datasets, and delivers public code and pretrained models for real-time quality assurance in clinical imaging.

Abstract

Ensuring the quality and integrity of medical images is crucial for maintaining diagnostic accuracy in deep learning-based Computer-Aided Diagnosis and Computer-Aided Detection (CAD) systems. Covariate shifts are subtle variations in the data distribution caused by different imaging devices or settings and can severely degrade model performance, similar to the effects of adversarial attacks. Therefore, it is vital to have a lightweight and fast method to assess the quality of these images prior to using CAD models. AdverX-Ray addresses this need by serving as an image-quality assessment layer, designed to detect covariate shifts effectively. This Adversarial Variational Autoencoder prioritizes the discriminator's role, using the suboptimal outputs of the generator as negative samples to fine-tune the discriminator's ability to identify high-frequency artifacts. Images generated by adversarial networks often exhibit severe high-frequency artifacts, guiding the discriminator to focus excessively on these components. This makes the discriminator ideal for this approach. Trained on patches from X-ray images of specific machine models, AdverX-Ray can evaluate whether a scan matches the training distribution, or if a scan from the same machine is captured under different settings. Extensive comparisons with various OOD detection methods show that AdverX-Ray significantly outperforms existing techniques, achieving a 96.2% average AUROC using only 64 random patches from an X-ray. Its lightweight and fast architecture makes it suitable for real-time applications, enhancing the reliability of medical imaging systems. The code and pretrained models are publicly available.

Figures (4)

• Figure 1: Sample images of the applied datasets. At the left, patches from the Philips X-ray dataset, showcasing different fluoroscopy and exposure doses. At the right, example chest X-ray images and patches from the BIMCV-COVID19+ data, each from a different machine (bottom right shows magnified views).
• Figure 2: Overview of AdverX-Ray's architecture. During inference, only the discriminator is used.
• Figure 3: Examples of OOD scores generated by the AdverX-Ray model trained on the Philips X-ray scans.
• Figure 4: AUROC and FPR scores for AdverX-Ray on the BIMCV-COVID19+ dataset for varying patch counts per image. GMM scans are not included due to the low degree of variability for different patch counts.