Format for both provided and submitted data: All of the data is provided in a {binary raw + ASCII header} format. We chose to use the header format MHD from ITK, since it can be read and written by open-source software, such as (at least) ITK and ImageJ with the plugin '3D IO' The MHD header text file contains all the necessary relevant geometry information to open correctly the volume data: dimensions, coding type, voxel spacing, raw data filename. The gray-level data of every CT dataset is stored in a .raw binary file, e.g. MICCAI09_HN01.raw. As outlined in the associated same-name .mhd header text file, e.g. MICCAI09_HN01.mhd, the gray-level values are coded with 16 bits per voxel (MET_USHORT, to accommodate the 12 bits usually significant in CT). Each of the two expert segmented structures is stored as a volumetric binary mask in the same file format, e.g. MICCAI09_HN01_MANDIBLE.raw + MICCAI09_HN01_MANDIBLE.mhd, and MICCAI09_HN01_BRAINSTEM.raw + MICCAI09_HN01_BRAINSTEM.mhd. In the volume, the region of interest is labeled with 1 values, the background is labeled with 0 values. The content of the .mhd header for each segmented structure is the same as the header for the associated CT dataset, except values are coded with 8 bits per voxel (MET_UCHAR) which leads to a .raw file size of exactly half the CT .raw file size, and the raw data filename indicated is different. In order to compare all segmentation methods at an equal level, whether they use voxel-level segmentations, or sub-voxel surface based segmentations, all methods need to submit a voxel level segmentation. This is also due to the fact that the expert manual segmentations, which are used as a quasi gold-standard to compare against, are on a voxel-level (and were delineated in axial slices just like in clinical practice). Thus, you have to submit a binary mask volume of the same size and spacing as the corresponding CT dataset. The file has to be stored in RAW format (e.g. MHD from ITK) with 8 bits per voxel (MET_UCHAR). A value of zero is interpreted as background label, all other values are treated as object labels. In order to automate the analysis on our end, we need you to store one file for each segmented structure, i.e. the brainstem and mandible structure need to be saved in different files. In case you use a surface-based segmentation method, you could resort to ITK's TriangleMeshToBinaryImageFilter for conversion to a volume. To sum up in other words, we need you to submit your segmentation results in the same format as the expert segmentation files provided in the training data. Downloads and Submissions: To limit space occupied by this website, no data will be hosted here, neither provided CT datasets and expert segmentations, nor the teams' submitted results. Instead, once you have successfully registered, we will contact you per email with details on how and where to download data and submit results. A file portal will be used. Note that any data exchanged this way remains available for download only for 21 days after upload onto the file portal, after which the data must be uploaded again if necessary. The content of this website is copyrighted © 2010 Vladimir Pekar, John Kim, Stéphane Allaire, Arish Qazi, David A. Jaffray