This repository contains the following data elements
The study itself is currently under review. Once a final version has been published, we will update this repository with a link to the manuscript.
To download the images from the ISIC Archive, please install the
isic-cli python package with
pip install isic-cli
and then download the collection (images and metadata file) with
isic image download -l 0 -c 166 ./images
The metadata.csv file contains the exemplar feature for which each of the 248 images was submitted, a benign/malignant indicator, and the diagnosis. It also contains in which of the 4 sub-studies (with 5 expert readers each) the image appeared
The masks subfolder contains 4,703 individual feature annotations across the 248 images by 5 readers from among 31 self-selected features.
As an alternative to the masks, the annotation data is also available as a pair of superpixel images (per original lesion image) and the indices into those superpixel images.
A Jupyter (python) notebook for how to access/process these images will be made available shortly (in this repository).