Coregistration options in mdreg#

Setup#

import numpy as np
import mdreg

Load test data

data = mdreg.fetch('MOLLI')
array = data['array'][:,:,0,:]

# Throughout this example we use the same signal model:
molli = {
    'func': mdreg.fit_abs_exp_recovery_2p,
    'TI': np.array(data['TI'])/1000,
}

# Visualise the data
anim = mdreg.plot.animation(array, vmin=0, vmax=1e4)

Once Loop Reflect

Coregistration with skimage#

If we don’t provide any detail on the coregistration method, mdreg uses the optical flow method optical_flow_tvl1() from the package skimage for coregistration:

coreg, fit, _, _ = mdreg.fit(
    array,
    fit_image=molli,
)

Visualise the results

anim = mdreg.plot.series(array, fit, coreg, vmin=0, vmax=1e4)

Once Loop Reflect

This has clearly had some effect, but the motion correction is not perfect and the coregistered image is somewhat blurred. We could try to improve the result by optimizing the parameters.

One parameter in optical_flow_tvl1() is the attachment, which by default is set to 15, but can be set to smaller values for smoother results. In this case we want the opposite, so let’s see what happens when we set it to 30.

We can do this by providing the fit_coreg argument to mdreg.fit:

coreg, fit, _, _ = mdreg.fit(
    array,
    fit_image=molli,
    fit_coreg={'attachment': 30},
)

# Visualise the results
anim = mdreg.plot.series(array, fit, coreg, vmin=0, vmax=1e4)

Once Loop Reflect

This appears to have had the desired effect and has resulted in a better motion correction.

Coregistration with elastix#

An other approach to optimization is to use another coregistration engine. mdreg also wraps elastix, so let’s try that.

By default mdreg applies elastix with bspline deformations, and default settings for all parameters. The result can be further customized by providing any valid elastix parameter in the dictionary. For this example we will change the default grid spacing from 16mm to 50mm. Since elastix is not default, we need to provide the package name as well. It also needs the pixel spacing so that dimensions in mm are well defined:

coreg, fit, _, _ = mdreg.fit(
    array,
    fit_image=molli,
    fit_coreg={
        'package': 'elastix',
        'spacing': data['pixel_spacing'],
        'FinalGridSpacingInPhysicalUnits': 50.0,
    },
)

# Visualise the results
anim = mdreg.plot.series(array, fit, coreg, vmin=0, vmax=1e4)

Once Loop Reflect

Coregistration with ants#

Apart from skimage and elastix, mdreg also wraps ants as a registration engine.

It can be called like elastix, by specifying the package and, if needed, any of the parameters to be set to different values than the default. We illustrate this here by coregistering with a transform type different from the default:

coreg, fit, _, _ = mdreg.fit(
    array,
    fit_image=molli,
    fit_coreg={
        'package': 'ants',
        'type_of_transform': 'SyNOnly',
    },
)

invalid value encountered in cast

Looking at the results, this is not a great solution for these data.

anim = mdreg.plot.series(array, fit, coreg, vmin=0, vmax=1e4)

Once Loop Reflect

Total running time of the script: (55 minutes 9.884 seconds)

Gallery generated by Sphinx-Gallery