UWarp: A Whole Slide Image Registration Pipeline to Characterize Scanner-Induced Local Domain Shift

Histopathology slide digitization introduces scanner-induced domain shift that can significantly impact computational pathology models based on deep learning methods. In the state-of-the-art, this shift is often characterized at a broad scale (slide-level or dataset-level), but not at the patch level, which limits our comprehension of the impact of localized tissue characteristics on model accuracy. To address this challenge, we present a domain shift analysis framework based on UWarp, a novel registration tool designed to accurately align histological slides scanned under varying conditions. UWarp employs a hierarchical registration approach, combining global affine transformations with fine-grained local corrections to achieve robust tissue patch alignment. We evaluate UWarp using two private datasets, CypathLung and one from the BosomShield project, containing whole slide images scanned by multiple devices. Our experiments show that UWarp outperforms existing open-source registration methods, achieving a median target registration error (TRE) below 4 pixels (< 1μ m at 40× magnification), compared to 21 pixels (∼ 5μ m at 40×) for the best state-of-the-art algorithm, while significantly reducing computational time. Additionally, we perform a localized characterization of scanner-induced domain shift by analyzing the predictions of a deep learning model for breast cancer pathological response. Accurate patch-level alignment between scanners is ensured using UWarp. Our analysis reveals that prediction variability correlates with tissue versus background ratio and nuclei density at patch level, highlighting the importance of accounting for localized domain shift to improve model robustness and adaptation strategies in computational pathology.