HARP: Unsupervised Histopathology Artifact Restoration

Abstract

Histopathological analysis, vital for medical diagnostics, is often challenged by artifacts insample preparation and imaging, such as staining inconsistencies and physical obstructions.Addressing this, our work introduces a novel, fully unsupervised histopathological artifactrestoration pipeline (HARP). HARP integrates artifact detection, localization, and restorationinto one pipeline. The first step to make artifact restoration applicable is an analysisof anomaly detection algorithms. Then, HARP leverages the power of unsupervised segmentationtechniques to propose localizations for potential artifacts, for which we select thebest localization based on our novel inpainting denoising diffusion model. Finally, HARPemploys an inpainting model for artifact restoration while conditioning it on the artifact localizations.We evaluate the artifact detection quality along with the image reconstructionquality, surpassing the state-of-the-art artifact restoration. Furthermore, we demonstratethat HARP improves the robustness and reliability of downstream models and show thatpathologists can not tell the difference between clean images and images restored throughHARP. This demonstrates that HARP significantly improves image quality and diagnosticreliability, enhancing histopathological examination accuracy for AI systems.