Welcome to the CT segmentation challenge
Moist Run
Last update: 3/28/2013 05:30 PM EST
The goal of the CT segmentation challenge is to compare the bias (where possible) and repeatability of automatic, semi-automatic and manual segmentations for lung CT studies. In this phase, investigators from Columbia, MGH, Moffitt and Stanford have identified 50 lung CT nodules and made available the data in DICOM format. Algorithm developers and users are requested to submit at least 4 repetitions of their algorithm for each nodule. We are quite flexible in accepting a variety of image formats for the segmentation volumes. We have experience with NIFTI, NRRD, JPG, PNG, DICOM-SEG, DICOM-RT, AIM, and LIDC-XML. If you would like to submit segmentations in a format other than one of these, please contact Jayashree at kalpathy@nmr.mgh.harvard.edu.
Data description:
We have 5 different collections that will be used for the "moist run".
The data (one zipped file per CT volume containing the DICOM slices) and locations of the nodules to be segmented (where available, in excel) are given below.
CUMC_FDA Phantom:
The CUMC-FDA phantom is a single CT volume with 10 lesions that have been identified here. Please submit your results for each nodule separately. We expect to receive 10n files where n is the number of repeat runs
Moffitt:
There are 10 volumes available.-these were the originally posted volumes.
For the moist run, please use QIN-LUNG-01-0007 and QIN-LUNG-01-0013 as well as the 8 new ones posted here. The locations of the nodules of interest are available here
Rider
10 of the studies from the RIDER dataset will be used for this challenge.
The initially posted studies are here
The locations of the nodules of interest are available here
The other set of studies are posted below
Stanford
10 volumes are available.The locations of the nodules are available here
LIDC
10 volumes are available.The locations of the nodules are available here