plastimatch

Image segmentation (MABS) command file reference

The mabs parameter file can have seven sections: PREALIGNMENT, ATLAS-SELECTION, TRAINING, REGISTRATION, STRUCTURES, LABELING and OPTIMIZATION_RESULT. Each section contatins specific entries that can not be mixed. All the possible parameters are listed below:

Option Section Default value Possible values Description
mode PREALIGNMENT disabled disabled, default, custom Set the method for prealign the images
reference PREALIGNMENT not set   Set the reference image
spacing PREALIGNMENT not set   Set the reference spacing
registration_config PREALIGNMENT not set plastimatch registration file without the GLOABL section Set the registration parmameters for prealign if “custom” mode is chosen
prealign_struct PREALIGNMENT not set name of an atlas structure Use the specified structure to execute center of gravity initialization
         
enable_atlas_selection ATLAS-SELECTION false true, false Enable atlas selection process
atlas_selection_criteria ATLAS-SELECTION nmi nmi, nmi-post, nmi-ratio, mse, mse-post, mse-ratio, random, precomputed Set the criterion to use for ranking the atlases (see at the bottom of this page for further explanations)
similarity_percent_threshold ATLAS-SELECTION 0.40 Ranging from 0.0 to 1.0 Percentage threshold used to pick the atlases from the ranking obtained on the basis of a nmi/mse similarity value. All the atlases having a similarity value greater equal to the following value will be selected. minimum_similarity_percentage_value = similarity_min + (threshold * (similarity_max - similarity_min))
atlases_from_ranking ATLAS-SELECTION -1 (means disabled) ranging from 0 to the whole number of atlases Number of atlases to pick from the ranking (nmi/mse based or precomputed).
mi_histogram_bins ATLAS-SELECTION 100 as usual for mi histogram bins setting Set the number of histogram bins to use for nmi computation
percentage_nmi_random_sample ATLAS-SELECTION not set by plastimatch, itk default value is used between 0 (not included) and 1 (included) Set the number of random voxels to use for NMI computation. The final value is equal to the number of voxels of fixe images time the set value.
roi_mask ATLAS-SELECTION not set image file name (.mha/.mhd/.nii.gz/.nrrd/ecc) Set the mask to reduce the volume of the subject/atlas images where the nmi/mse will be computed
selection_reg_parms ATLAS-SELECTION not set plastimatch registration file without the GLOABL section File where are stored the registration parameters (without GLOBAL stage) that will be used for the nmi-post, nmi-ratio mse-post and mse-ratio selection
lower_mi_value_subject ATLAS-SELECTION not set in the range of the intesities of the subject image. Anyway lower than the upper_mi_value_subject Lower intensity bound on the histogram of the subject image. Only the values greater than this threshold will be used for nmi computation
upper_mi_value_subject ATLAS-SELECTION not set in the range of the intesities of the subject image. Anyway higher than the lower_mi_value_subject Upper intensity bound on the histogram of the subject image. Only the values lower than this threshold will be used for nmi computation
lower_mi_value_atlas ATLAS-SELECTION not set in the range of the intesities of the atlas image. Anyway lower than the upper_mi_value_atlas Lower intensity bound on the histogram of the atlas image. Only the values greater than this threshold will be used for nmi computation
upper_mi_value_atlas ATLAS-SELECTION not set in the range of the intesities of the atlas image. Anyway higher than the lower_mi_value_atlas Upper intensity bound on the histogram of the atlas image. Only the values lower than this threshold will be used for nmi computation
min_random_atlases ATLAS-SELECTION 6 ranging from 1 to the whole number of atlases. Anyway equal/lower than max_random_atlases Minimum number on atlases to extract when random selection is choosen
max_random_atlases ATLAS-SELECTION 14 ranging from 1 to the whole number of atlases. Anyway equal/greater than min_random_atlases Maximim number on atlases to extract when random selection is choosen
precomputed_ranking ATLAS-SELECTION note set text file where a precomputed ranking was stored Text file containing the precomputed ranking. For each line there is a ranking for an patient. The style is: patient1: atl1 atl2 atl3 atl4
         
training_dir TRAINING not set folder path Folder that contains the data for the training
fusion_criteria TRAINING gaussian “gaussian”, “staple”, “gaussian,staple” and “none” Labels fusion criterion
distance_map_algorithm TRAINING not set   Implementation for distance map computation
minsim_values TRAINING L 0.0001:1:0.0001   Minimum similarity values for gaussian labels fusion
rho_values TRAINING 1:1:1   Rho values for gaussian labels fusion
sigma_values TRAINING L 1.7:1:1.7   Sigma values for gaussian labels fusion
threshold_values TRAINING 0.5   Threshold values for gaussian labels fusion
write_distance_map_files TRAINING 1 (true) 0 (false) or 1 (true) Write distance map files on the disk (only for gaussian fusion)
write_thresholded_files TRAINING 1 (true) 0 (false) or 1 (true) Write thresholded files on the disk (only for gaussian fusion)
write_weight_files TRAINING 1 (true) 0 (false) or 1 (true) Write weight files on the disk (only for gaussian fusion)
write_warped_images TRAINING 1 (true) 0 (false) or 1 (true) Write warped images on the disk
write_warped_structures TRAINING 1 (true) 0 (false) or 1 (true) Write warped structures on the disk
confidence_weight TRAINING 1.0   Multiplicative factor for the prior probability that any pixel would be classified as inside the structure (only for staple fusion)
         
registration_config REGISTRATION not set plastimatch registration file without the GLOABL section Set the registration parmameters for the deformable registration
         
no predefined entries, just the list of the structures to segment (without extension file) STRUCTURES not set    
         
input LABELING not set    
output LABELING not set    
         
registration OPTIMIZATION_RESULT not set    
gaussian_weighting_voting_rho OPTIMIZATION_RESULT 1.0    
gaussian_weighting_voting_sigma OPTIMIZATION_RESULT 30.0    
gaussian_weighting_voting_minsim OPTIMIZATION_RESULT 0.3    
optimization_result_confidence_weight OPTIMIZATION_RESULT 1.0    
gaussian_weighting_voting_thresh OPTIMIZATION_RESULT 0.4    

The possible atlas selection criteria are nmi, nmi-post, nmi-ratio, mse, mse-post, mse-ratio, random, precomputed.

Criterion Description
nmi Normalized mutual information computed between prealigned (ridgly) images
nmi-post Normalized mutual information computed between images after a deformable registration
nmi-ratio Score calculated using the normalized mutual information computed before and after a deformable registration
mse Root mean square error computed between prealigned (ridgly) images
mse-post Root mean square error computed between images after a deformable registration
mse-ratio Score calculated using the root mean square error computed before and after a deformable registration
random Random selection of a random number of atlases
precomputed Ranking read from a text file containing a precomputed list of atlases

Structure name mapping

It is required to put the name of each structure to be segmented into the STRUCTURES section. If structure names are uniformly defined, then you simply need to list them:

[STRUCTURES]
Prostate
Bladder
Rectum

If the structure names are not uniform, you can use this section to map multiple names (within an atlas) to the same target name. To do so, put the target name to the left of the equal sign, and the original name to the right of the equal sign. For example, the following would map three different spellings to the same target name:

[STRUCTURES]
Prostate
Prostate=prostate
Prostate=Pros_GTV