Fig. 1
From: The impact of pre-processing techniques on deep learning breast image segmentation
Results from the Domain-Specific (DS) and Domain Non-Specific (DNS) pipelines using the CBIS-DDSM dataset: (a) mean validation Dice Similarity Coefficient and (b) mean Hausdorff Distance across 5 folds for models with different resizing methods and pixel intensity normalization, shown with their respective standard deviations. DS and DNS models differ only in resizing and normalization parameters.
