Description: This dataset consists of 143 mammograms and associated lesion segmentations saved as NumPy (.npy) files. The dataset is split across three directories by the lesion shape (Round Oval Irregular). Each individual file is a cropped mammogram with 2 channels each with dimension 224x224. The first channel contains the grayscale mammogram with pixel values from 0-1. The second channel is a binary segmentation of the lesion identified by 1's where the lesion exists and 0's elsewhere.
All validation images were segmented manually by a non-expert annotator (VS) using GIMP. These initial image segmentations were presented to a board certified radiologist with 10 post-graduation years of general radiology experience (FRS) for correction using PowerPoint. Corrections were then incorporated
into the binary segmentation mask labels by non-expert annotators (NH VS AW) using GIMP.
For more information refer to the publication titled "Improving Annotation Efficiency for Fully Labelling a Breast Mass Segmentation Dataset".
All validation images were segmented manually by a non-expert annotator (VS) using GIMP. These initial image segmentations were presented to a board certified radiologist with 10 post-graduation years of general radiology experience (FRS) for correction using PowerPoint. Corrections were then incorporated
into the binary segmentation mask labels by non-expert annotators (NH VS AW) using GIMP.
For more information refer to the publication titled "Improving Annotation Efficiency for Fully Labelling a Breast Mass Segmentation Dataset".