1. Two lung cancer cases: (ftp download of ZIP files) CT scans at multiple time points during treatment. Nodule markings are contained in .xml files accompanying the scans. More datasets will be presented in the future. 2. This first dataset (ftp download) provided by LIDC contains 23 nodules and a boundary definition of the nodules from a screening and diagnostic caseload. This preliminary database may prove to be useful for the evaluation of image segmentation methods. LIDC has plans to provide additional data sets later in 2005 and the final database of 400 cases in summer of 2006.

NOTES re download: DICOM Works can be downloaded, unzipped, and used to read the images. The file is tarfile. It seems to download without the .gz extension that it includes on the web site. WinRAR can be downloaded to unzip tarfile. It creates 2 folders, CT and PMAP, each of which is unzipped separately. After this is done, DICOM Works can be used to open the individual case folders (labeled 01, 02, etc) and p-map folders and to step through the images one at a time. Note that the p-map file does not seem to have a lot of information, with brightness only in the nodule(s).

DICOM files (part 10) for 23 lung nodules are enclosed. Both the CT data sets and their matching probability map (pmap) files are included as dual directories having matching nodule numbers. The CT data sets should be viewed using a normal lung window/level, e.g. 1500/-500 HU, while the pmaps should be viewed using a window/level something similar to 1500/750.

File organization after unzipping and extracting the tar file

ct                                      pmap
 |_ ./01                                 |_ ./01  
 |     |_./1.2....0.200501271.1.1        |     |_1.2....0.200501272.1.1
 |     |_./1.2....0.200501271.1.2        |     |_1.2....0.200501272.1.2
 |     .                                 |     .
 |     .                                 |     .
 |     .                                 |     .
 |     |_./1.2....0.200501271.1.36       |     |_1.2....0.200501272.1.36
 |_ ./02                                 |_ ./02
 |     |_ (etc)                          |     |_ (etc)
 .                                       .
 .                                       .
 .                                       .
 |_ ./23                                 |_ ./23
       |_ (etc)                                |_ (etc)

For each of the 23 nodules, 6 expert lung radiologists provided annotations, i.e. segmentations, of all nodules using 3 different methods, one manual and two auto-matic, for a total of 18 possible radiologist/method combinations. The pmap values for each voxel provide the fraction of radiologist/method combinations that included that voxel within their annotation scaled by 1000. For example, if all of the radiologist/method combinations included a voxel withing their outlines, the value of that pmap voxel would be 1000; if only 50% of the combinations included that voxel, its pmap value would be 500.

For the experiment all radiologists were instructed to draw the edge of nodules one voxel outside the nodule. Here the pmap files have the median edge map impressed on the pmap such that all voxels inside the median edge have prob- abilities of 50% or greater. The edge voxels in the pmap have had their pmap values replaced with the maximum possible signed short integer value, i.e. 32767, for convenient viewing at all window/level settings.

The data is a small data set of nodules which may be used for validation of nodule characterization/segmentation software. Further analysis of these data will be published by the LIDC in the near future.