Data files and code for the MSDSlab Veronika Cheplygina will host on the 31st of May 2018.
short description of the data
The data contains additional annotations collected from the crowd, for some images from the ISIC 2017 challenge (https://challenge.kitware.com/#challenge/n/ISIC_2017%3A_Skin_Lesion_Analysis_Towards_Melanoma_Detection). One of the goals of the challenge is to predict melanoma from the images.
We collected several visual assessments of the images from the crowd, such as whether the image is asymmetric or not (this is something that experts look at). In total for this part of the data there are 5 visual characteristics (Asymmetry, Border, Color, Dermoscopic structures, Blue glow), each of which was assessed by 6 crowd annotators. For example column Asymmetry_7_1 is how the 1st annotator assessed the characteristic "Asymmetry".
We want to investigate whether such crowd assessments can help us to train better machine learning algorithms for predicting melanoma in the future. Relevant questions are whether there are relationships between the assessments and the Melanoma label (column AG), whether some characteristics are more informative than others, whether there is any information in disagreement between annotators, etc.