ANTs Atropos Tissue Segmentation¶

Library: ANTs | Docker Image: antsx/ants

Function¶

Probabilistic tissue segmentation using expectation-maximization algorithm with Markov random field spatial prior.

Modality: Brain-extracted 3D NIfTI volume plus brain mask.

Typical Use: GMM-based brain tissue segmentation with spatial regularization.

-d (dimension), -a (input image), -x (mask), -i (initialization: KMeans[N] or PriorProbabilityImages), -c (convergence), -o (output)

Initialize with KMeans[3] for basic GM/WM/CSF or use prior probability images. MRF prior improves spatial coherence.

Name	Type	Required	Label	Flag
`dimensionality`	`int`	Yes	Image dimensionality (2, 3, or 4)	`-d`
`intensity_image`	`File`	Yes	Input intensity image for segmentation	`-a`
`mask_image`	`File`	Yes	Mask image defining segmentation region	`-x`
`output_prefix`	`string`	Yes	Output segmentation filename	`-o`
`initialization`	`string`	Yes	Initialization method (e.g., kmeans[3], otsu[3], priorProbabilityImages[...])	`-i`
`likelihood_model`	`enum`	No	Likelihood model for intensity estimation	`-k`
`mrf`	`string`	No	MRF parameters [smoothingFactor,radius] e.g., [0.3,1x1x1]	`-m`
`convergence`	`string`	No	Convergence parameters [iterations,threshold] e.g., [5,0.001]	`-c`
`prior_weighting`	`double`	No	Prior probability weight (0-1)	`-w`
`use_euclidean_distance`	`boolean`	No	Use Euclidean distance for label propagation	`-e`
`posterior_formulation`	`string`	No	Posterior formulation (e.g., Socrates[1])	`-p`
`winsorize_outliers`	`string`	No	Outlier handling method (e.g., BoxPlot[0.25,0.75,1.5])	`--winsorize-outliers`
`verbose`	`boolean`	No	Enable verbose output	`--verbose`

Name	Type	Glob Pattern
`segmentation`	`File`	`$(inputs.output_prefix)`
`posteriors`	`File[]`	`$(inputs.output_prefix)Posteriors.nii.gz`
`log`	`File`	`Atropos.log`

likelihood_model: Gaussian, HistogramParzenWindows, ManifoldParzenWindows

Available versions: latest, master, v2.6.5, 2.6.5, v2.6.4, 2.6.4, v2.6.3, 2.6.3, v2.6.2, 2.6.2 and 5 more