This repository contains code for preprocessing the LIDC-IDRI dataset, a publicly available collection of annotated lung cancer screening CT scans.

Stages

The preprocessing code is broken up into two stages:

• stage1 extracts image data and annotations from the DICOM and XML files from the original dataset. The CT scans are represented as volumetric data after this stage.
• stage2 extracts axial, sagittal, and coronal slices from the stage1 output. This stage also produces augmented views of each example. The CT scans are represented as 2D images after this stage.

Citations

If you use this preprocessing code in your research, we would greatly appreciate you citing the following paper:

@article{nibali2017pulmonary,
  author="Nibali, Aiden and He, Zhen and Wollersheim, Dennis",
  title="Pulmonary nodule classification with deep residual networks",
  journal="International Journal of Computer Assisted Radiology and Surgery",
  year="2017",
  issn="1861-6429",
  doi="10.1007/s11548-017-1605-6",
}

And, of course, don't forget to cite the original sources of data and annotations:

• Smith K, Clark K, Bennett W, Nolan T, Kirby J, Wolfsberger M, Moulton J, Vendt B, Freymann J. Data from LIDC-IDRI. http://dx.doi.org/10.7937/K9/TCIA.2015.LO9QL9SX
• A. P. Reeves, A. M. Biancardi, "The Lung Image Database Consortium (LIDC) Nodule Size Report." Release: 2011-10-27-2. At: http://www.via.cornell.edu/lidc/, October 27, 2011