Multi-Artifact Detection and Filtering in Digital Pathology Using Intrinsic Image Properties

Digital pathology images, particularly whole slide images (WSIs), often contain macro-scale artifacts such as tissue folds, out-of-focus regions, and pen markings that can introduce bias and compromise the performance of machine learning models. The variability of tissue types, artifact appearances, and sample preparation and imaging conditions makes reliable artifact detection challenging. Here we introduce WSI Spectral Thresholding for Artifact Removal (WSI-STAR), a reference-free approach for detecting and removing macro-scale artifacts from WSIs that leverages the relative scale, spatial distribution, and spectral characteristics of macro-scale artifacts compared with tissue. Our pipeline combines spectral analysis and superpixel segmentation to effectively group and merge adjacent regions based on their frequency information, followed by an adaptive thresholding process. The result is an improved tissue mask that serves as a higher-quality input to downstream analysis.