Segmentation masks INbreast

This dataset provides manually created segmentation masks of the images in the INbreast dataset by I.C. Moreira et al[1]. The masks are saved as nrrd files with pixel-wise ground truth for background (0), breast (1), and pectoral muscle (2) (when present). This dataset is created for the development of a mammogram segmentation model[2]. Segmentation masks were created in three steps, first initialization of the breast boundary by Otsu thresholding[3], second a pectoral muscle initialization for MLO images, and lastly a manual adjustment of the mask, as show in Figure 1. For the MLO views, the already publicly-available annotations of the pectoral muscle were used as the initialization. Finally, each segmentation mask was checked visually and adjusted manually using ITK-SNAP 3.6.0[4] by one of four medical imaging scientists with experience in mammography. This also includes adding pectoral muscle annotation were it was visible in CC views. [1] I. C. Moreira et al., "INbreast: Toward a Full-field Digital Mammographic Database", Acad. Radiol. 19(2), 236–248 (2012) [2] S.D. Verboom et al., "Deep learning-based breast region segmentation in raw and processed digital mammograms: generalization across views and vendors", Journal of Medical Imaging, 11(1), 014001 (2023) [3] N. Otsu, "A Threshold Selection Method from Gray-Level Histograms", IEEE Trans. Syst. Man. Cybern. 9(1), 62–66 (1979) [4] P. A. Yushkevich et al., "User-guided 3D active contour segmentation of anatomical structures: Significantly improved efficiency and reliability", Neuroimage 31(3), 1116–1128 (2006)