Image Registration

The ANTs provide two set of algorithms : Linear and Non-linear Transforms.
Linear Registration: is mainly for rough alignment of dataset and offers following algorithms.

1. Translation.
2. Rigid.
3. Similarity.
4. Affine.

Non-Linear Registration: is for fine alignment of dataset and it uses SyN( Symmetric Normalization ) algorithm for non-linear transform. It is based on diffeomorphic mapping. It generally takes quadratic time to do the registration. Algorithm spent most of the time in this step.

1. BSpline SyN.
2. Geodesic SyN.
3. Greedy SyN.

There is no standard algorithm that can solve all registration task. The choice of algorithm is totally dependent on dataset.

Similarity Metrics: It is the critical step in registration process which matches how well the target dataset is matched with referenced or fixed dataset. There is no fixed rule to select the metric, it totally depend on conditions like algorithm, dataset etc.
Generally, Mutal Information(MI) is preferred for linear transform and Cross-correlation(CC) is useful for fine scale non-linear registration step.

1. Mutal Information(MI)
2. Cross-correlation(CC)
3. Probabilistic Matching (PB)
4. Mean Square Distance(MSQ)
5. Point-set Expectation(PSE)

Following is the bash script for the image registration process:

                      
                      #!/bin/bash
                      #
                      dim=3 # image dimensionality
                      AP="" # /home/yourself/code/ANTS/bin/bin/  # path to ANTs binaries
                      ITK_GLOBAL_DEFAULT_NUMBER_OF_THREADS=8  # controlling multi-threading
                      export ITK_GLOBAL_DEFAULT_NUMBER_OF_THREADS
                      f=$1 ; m=$2    # f: fixed or referenced image file  m: moving image file or unregistered volume
                      mysetting=$3   # current settings like testing or forproduction
                      D=$4           #/home/image_registration/ # path to output directory
                      if [[ ! -s $f ]] ; then echo no fixed $f ; exit; fi
                      if [[ ! -s $m ]] ; then echo no moving $m ;exit; fi
                      if [[ ${#mysetting} -eq 0 ]] ; then
                      echo usage is
                      echo $0 fixed.nii.gz moving.nii.gz mysetting
                      echo  where mysetting is either forproduction or testing
                      exit
                      fi
                      nm1=` basename $f | cut -d '.' -f 1 `
                      nm2=` basename $m | cut -d '.' -f 1 `
                      reg=${AP}antsRegistration           # path to antsRegistration
                      if [[ $mysetting == "testing" ]] ; then
                        its=200x200x200
                        percentage=0.2
                        syn="20x20x0,1.e-8,5"
                      else
                        its=100x100x100
                        percentage=0.25
                        syn="100x100x50,1.e-8,5"
                        mysetting=forproduction
                      fi
                      echo affine $m $f outname is $nm am using setting $mysetting
                      nm=${D}${nm1}_fixed_${nm2}_moving_setting_is_${mysetting}   # construct output prefix
                      $reg -d $dim -r [ $f, $m ,1]  \
                                              -m mattes[  $f, $m , 1 , 32, regular, $percentage ] \
                                               -t translation[ 0.1 ] \
                                               -c [$its,1.e-8,20]  \
                                              -s 4x2x1vox  \
                                              -f 6x4x2 -l 1 \
                                              -m mattes[  $f, $m , 1 , 32, regular, $percentage ] \
                                               -t rigid[ 0.1 ] \
                                               -c [$its,1.e-8,20]  \
                                              -s 4x2x1vox  \
                                              -f 3x2x1 -l 1 \
                                              -m mattes[  $f, $m , 1 , 32, regular, $percentage ] \
                                               -t affine[ 0.1 ] \
                                               -c [$its,1.e-8,20]  \
                                              -s 4x2x1vox  \
                                              -f 3x2x1 -l 1 \
                                              -m mattes[  $f, $m , 0.5 , 32 ] \
                                              -m cc[  $f, $m , 0.5 , 4 ] \
                                               -t SyN[ .1, 3, 0 ] \
                                               -c [ $syn ]  \
                                              -s 4x2x1vox  \
                                              -f 4x2x1 -l 1 -u 1 -z 1 \
                                             -o [${nm},${nm}_diff.nii.gz,${nm}_inv.nii.gz]
                      
                      ${AP}antsApplyTransforms -d $dim -i $m -r $f -n linear -t ${nm}1Warp.nii.gz -t ${nm}0GenericAffine.mat -o ${nm}_warped.nii.gz
                      
                      

Imagine now that you want to do the inverse mapping i.e. you will take any mapped stack from the reference brain, or an empty stack with the Owen - masks, and transform that into your stack e.g. 2P stack and overlay. This allows you to immediately visualize neuro transmitter identity and brain regions together with planer neural activity. Here is an example stack, where we used the inverted mapping to map the zbrain vglut stack as well as the zbrain gadb1 stack into 2P stack. This registration reveals the pattern of excitatory and inhibitory cells within each imaged plane as shown below.
Finally, following are the slices of the registered image stack.