Multi-target stain normalization for histology slides

This research introduces a novel stain normalization method for histopathology slides that enhances robustness against staining variations by using multiple reference images. The parameter-free approach, MultiMacenkoNormalizer, improves the accuracy of tasks like nuclei segmentation by capturing diverse staining patterns. It compares four methods (stochastic, concatenation, avg-pre, avg-post) and finds that avg-post and concatenation with a moderate number of references (e.g., 8 or 10) yield the best results. The study, using the Lizard dataset and evaluating mean intersection-over-union (IoU), demonstrates improved performance when generalizing to external datasets. The method enhances the reliability and robustness of computational pathology tasks without requiring specific normalization knowledge or a fixed reference set, making it a promising tool for enhancing segmentation accuracy in histology images.