ANTS(1)

📦
version 2.6.4 · 2 versions
, in ubuntu resolute

💖 ANTS: User Commands Bestie 💖

Okay so this is the ANTS manual page 🦋 It lives in the User Commands section, and it's all about the ANTS registration suite. Let's get into it ✨

✨ NAME ✨

ANTS, part of the ANTS registration suite 🌸

✨ DESCRIPTION ✨

The DESCRIPTION is split into two parts, a COMMAND bit and an OPTIONS bit. Main character energy 💅

💅 COMMAND 💅

ANTS

🎀 OPTIONS 🎀

  • -x, --mask-image maskFileName — okay so this mask is defined in the 'fixed' image space and marks the region of interest the registration gets computed over. Above 0.1 means inside the mask. Continuous values in the range [0.1,1.0] act kinda like a probability 🔮 and anything > 1 gets treated as 1.0 💀
  • -m, --image-metric — the metric weights are relative to the weights on the N other metrics you pass to ANTS (and N is unlimited, no cap!) 🔥 So the weight on the i^{th} metric becomes w_i=w_i/ ( sum_i w_i ). Here's the metric squad:

Intensity-Based Metrics:

CC/cross-correlation/CrossCorrelation[fixedImage,movingImage,weight,radius/OrForMI-#histogramBins]
MI/mutual-information/MutualInformation[fixedImage,movingImage,weight,radius/OrForMI-#histogramBins]
SMI/spatial-mutual-information/SpatialMutualInformation[fixedImage,movingImage,weight,radius/OrForMI-#histogramBins]
PR/probabilistic/Probabilistic[fixedImage,movingImage,weight,radius/OrForMI-#histogramBins]
SSD
--- standard intensity difference.[fixedImage,movingImage,weight,radius/OrForMI-#histogramBins]
MSQ/mean-squares/MeanSquares
-- demons-like, radius > 0 uses moving image gradient in metric
deriv.[fixedImage,movingImage,weight,radius/OrForMI-#histogramBins]

Point-Set-Based Metrics:

PSE/point-set-expectation/PointSetExpectation[fixedImage,movingImage,fixedPoints,movingPoints,weight,pointSetPercentage,pointSetSigma,boundaryPointsOnly,kNeighborhood,PartialMatchingIterations=100000]

The partial matching option assumes the complete labeling is in the first set of label parameters ... more iterations leads to more symmetry in the matching. 0 iterations means full asymmetry 🙈

[fixedImage,movingImage,fixedPoints,movingPoints,weight,pointSetPercentage,pointSetSigma,boundaryPointsOnly,kNeighborhood,alpha,meshResolution,splineOrder,numberOfLevels,useAnisotropicCovariances]
  • -o, --output-naming — the name for your output, either a prefix or a name+type. Examples: -o OUT, -o OUT.nii, or -o OUT.mha 🍓
  • --R — TODO/FIXME 🙈 this --R sets an ROI option. It passes a vector of parameters that sets the center and bounding box of the region of interest for a sub-field registration. For example, in 3D the option setting -r 10x12x15x50x50x25 sets up a bounding box of size 50,50,25 with origin at 10,12,15 in voxel (should this be physical?) coordinates. Values: 0
  • -i, --number-of-iterations — the number of iterations per level, as a vector. For example 100x100x20. Values: 10x10x5
  • --Restrict-Deformation — this one restricts the gradient that drives the deformation by scalar factors along the dimensions you pick. It's a TReal vector of length ImageDimension that gets multiplied against the similarity metric's gradient values. In 3D, 0.1x1x0 sets the z gradient to zero, scales x by 0.1, and keeps y as is (1, so no change). You end up with a 2.5-Dimensional registration, since there's still 3D continuity in the mapping ✨ Values: 1x1
  • -v, --verbose — verbose output, so you get all the tea 🫖
  • --use-all-metrics-for-convergence — enables using a weighted sum of all metric terms for the convergence computation. By default only the first metric is used 🔮 Values: 0
  • -h — prints the help menu (the short version) 🧚 Values: 0
  • --help — prints the help menu, the full version 💜 Values: 1, 0
  • -t, --transformation-model — pick a transformation with the format TRANSFORMATION[gradient-step-length,number-of-time-steps,DeltaTime,symmetry-type]. Then choose one of these TRANSFORMATIONS:
    • Diff = diffeomorphic
    • Elast = Elastic
    • Exp = exponential diff
    • Greedy Exp = greedy exponential diff, like diffeomorphic demons. Same parameters.
    • SyN = symmetric normalization

DeltaTime is the integration time-discretization step (sub-voxel), and the n-time steps are currently fixed at 2. Values: SyN[0.5]

  • -r, --regularization — the format is REGULARIZATION[gradient-field-sigma,def-field-sigma,truncation]. Choose one of these REGULARIZATIONS:
    • Gauss = gaussian
    • DMFFD = directly manipulated free form deformation

Values: Gauss[3,0.5]

  • -a, --initial-affine — uses the input file as the initial affine parameter 🦋
  • -F, --fixed-image-initial-affine — uses the input file as the initial affine parameter for the fixed image 🌈
  • --fixed-image-initial-affine-ref-image — the reference space for using the input file as the initial affine parameter for the fixed image 🎀
  • -T, --geodesic — takes 0, 1, or 2. 0 = not time-dependent, 1 = asymmetric, 2 = symmetric 🌸
  • -G, --go-faster — takes true or false. If true, SyN is faster but loses some accuracy with respect to the inverse-identity constraint (see Avants MIA 2008). > Heads up bestie, speed has a cost here ⚠️ Values: false
  • --continue-affine — true (default) or false. This decides whether or not to perform affine, given the initial affine parameters 💅 Values: true
  • --number-of-affine-iterations — the number of iterations per level, as a vector. For example 100x100x20. Values: 10000x10000x10000
  • --use-NN — uses nearest neighbor interpolation 🍓 Values: 0
  • --use-Histogram-Matching — uses histogram matching of moving to fixed image ✨ Values: 0
  • --affine-metric-type — picks the affine metric:
    • MI: mutual information (default)
    • MSQ: mean square error
    • SSD
    • CC: Normalized correlation
    • CCH: Histogram-based correlation coefficient (not recommended)
    • GD: gradient difference (not recommended)

Values: MI

  • --MI-option — the option of mutual information, written as MI_bins x MI_samples. The default is 32x32000 🧮 Values: 32x5000
  • --rigid-affine — uses a rigid transformation: true or false (default) 🪨 Values: false
  • --do-rigid — uses a rigid transformation: true or false (default) 🪨 Values: false
  • --affine-gradient-descent-option — the option of gradient descent in the affine transformation, in the order maximum_step_length x relaxation_factor x minimum_step_length x translation_scales 🎢 Values: 0.1x0.5x1.e-4x1.e-4
  • --use-rotation-header — uses the rotation matrix in the image headers: true (default) or false 🧭 Values: false
  • --ignore-void-origin — ignores the apparently unmatched origins (when use-rotation-header is false and the rotation matrix is identity). Takes true (default) or false 🫥 Values: false
  • --gaussian-smoothing-sigmas — at each resolution level the image gets subsampled and smoothed by Gaussian convolution. This option lets you override the default smoothing by specifying sigma values (in mm) for smoothing both the fixed and moving images at each resolution level 💖 Values: (blank)
  • --subsampling-factors — at each resolution level the image gets subsampled and smoothed by Gaussian convolution. This option lets you override the default subsampling by specifying the subsampling factor for both the fixed and moving images at each resolution level ✨ Values: (blank)

That's the whole ANTS page, bestie 🫶 It's from ANTS 2.6.4+dfsg, December 2025, and it's the ANTS(1) manual. Slay the registration 💅🌈

📦 where this page comes from