Get Started with Medical Registration Estimator

Since R2024b

The Medical Registration Estimator app enables you to align 3-D medical image volumes, and visually and quantitatively compare the results of different registration techniques. You can apply intensity and surface-based rigid registration and deformable registration techniques. You can export aligned volume data, affine transformation objects or nonrigid displacement fields, and generate functions to apply outside the app.

This tutorial provides an overview of the capabilities of the Medical Registration Estimator and the registration techniques it supports.

Medical Registration Estimator app window

Open Medical Registration Estimator App

Open the Medical Registration Estimator app from the Apps tab on the MATLAB^® Toolstrip, under Image Processing and Computer Vision. You can also load the app by using the medicalRegistrationEstimator command.

Load Volumes

The app supports loading data from medicalVolume objects in the workspace, or from files in the NIfTI, NRRD, DICOM, and TIFF formats. You can load medicalVolume objects while opening the app. Alternatively, load medicalVolume objects or files from inside the app.

To load the medicalVolume objects movingVol and fixedVol while opening the app, use this syntax:
```
medicalRegistrationEstimator(movingVol,fixedVol)
```
The app opens and automatically loads movingVol as the moving volume and fixedVol as the fixed volume.
To load volumes after opening the app follow these steps:
1. From the app toolstrip, select Import.
2. In the dialog box that opens, under Fixed Volume and Moving Volume, select the data to import. The app supports importing volumes from a single file (NIfTI, NRRD, DICOM, or TIFF), multifile volumes from a folder of DICOM files, and loading medicalVolume objects from the workspace.
3. For each volume, under Spatial Referencing Information, you can optionally select a medicalref3d object in the workspace that defines spatial referencing for the volume. This is particularly useful when importing volumes from TIFF files. If you leave the default option selected, the app uses spatial referencing from the file or medicalVolume, if available. If the data does not specify spatial referencing, the app assumes a voxel spacing value of 1-by-1-by-1 and that the first voxel is located at (1, 1, 1).
4. Under Initial Transformation Object, you can optionally select an affinetform3d object from the workspace that defines an initial registration of the volumes, such as from a previous app session or calculated programmatically outside the app using a function like imregister.
5. When you are done specifying the data to import, click OK.

Import Volume dialog box, showing options for importing the fixed and moving volumes, spatial referencing information, and an initial transformation object

Adjust Display Settings

Once you import data, the app displays the volumes together in a four-panel view. You can optionally change these display settings in the Display tab of the app toolstrip.

Toolstrip Section	Description
Select Volume	Select whether to display only the fixed volume, the moving volume, or both.
2-D Slices	Select the overlay method used to display the 2-D slice overlays. By default, the app displays a green-magenta overlay, where the fixed volume is magenta and the moving volume is green. You can change this option to view the slices as a red-cyan overlay, as a difference image, as a checkerboard of the two images, or as a dynamic image that flickers between the fixed and moving volumes.
3-D Volumes	Change the isovalue used to define the 3-D surfaces, the display color of each volume, and the transparency of each volume. To change the isosurface value or the transparency values, drag the corresponding slider. To change the display color, click Fixed Color or Moving Color to open a color picker dialog box.
3-D Background	Turn off the 3-D display pane background gradient, change the background and background gradient colors, or restore the default background options.
Markers	Use the Display Markers setting to turn off the L-R and A-P 2-D display markers, or the orientation axes and scale bar in the 3-D display pane.
Inspect	Use the Crosshair Navigation setting to turn on crosshair indicators in the 2-D slice panes that show the relative positions of the slices in the other planes. To navigate slices, drag an indicator to a new position.
Layout	Change the layout of the display panes. By default, the app shows the 2-D slice panes and 3-D pane using equal sizes. Select one of the Focus layouts to view one pane larger than the others.

These images illustrate display settings related to displaying data and available display markers.

Display settings

Manage App Session Using Trials

The app manages your registration experiments using trials. You can apply multiple techniques in a series within one trial, or compare techniques applied independently in separate trials. When you apply a series of techniques within one trial, each technique is a step.

You can view details about each trial in the Registration Browser. Each trial is displayed, with the option to expand an view details of each step. The step lists the name of the technique applied, the accuracy (SSIM), and the runtime, in seconds, for that step. To learn more about the structural similarity (SSIM) index, see ssim. The display panes and Transformation pane show the results for that step, so you can click between steps to evaluate the incremental impact of each technique you apply. You can add or modify trials using the controls in the Trial section of the app toolstrip.

Registration browser

In each trial, you can apply a combination of manual and automated registration techniques to align the volumes. The techniques you use and the order in which you apply them depends on your data and application. In general, a typical workflow includes these steps:

Perform a rough alignment using manual registration or a faster automated, affine registration technique like Moment of Mass.
Apply one or more refined automated techniques to improve the registration. If your data includes local nonlinear deformations you want to estimate or correct, apply a deformable technique.
Optionally adjust the registration using manual registration. Note that manual registration cannot be applied after a deformable technique.

Apply Manual Registration

To apply manual registration in the app, follow these steps:

From the app toolstrip, select Manual Registration. An new step, Manual, is added to the trial in the Registration Browser.
In the Manual Parameters panel, use the controls to specify the rotation, translation, and scaling needed to achieve a rough alignment. For most volumes with specified spatial referencing information, scaling is not required. To change a parameter, drag the slider. The current value for the control is displayed next to the slider, and you can make fine adjustments using the up and down arrows. The sensitivity of the slider depends on the Slider Range, which you can optionally adjust.
As you apply the transformations, the 3-D view updates in real time. Pause to view the updated 2-D slice views. The Transformation pane also updates to display the 4-by-4 transformation matrix that describes the current registration.

Manual Registation panel

Apply Automated Registration

To apply an automated registration technique in the app, follow these steps:

In the app toolstrip, from the Automated gallery, select a technique. The techniques are grouped as Intensity, Surface, and Deformable techniques based on their underlying algorithm. For details about each technique, see the table below.
After you select a technique from the gallery, the Parameter Panel opens on the right side of the app window. The parameters for each technique correspond to adjustable settings of the underlying algorithm. For details about the parameters for each technique, see the table below.
In the app toolstrip, select Run.
When the technique completes, the 2-D and 3-D display panes update to show the registered volumes. The Transformation pane updates to show the updated 4-by-4 transformation matrix for affine techniques. If you apply a deformable technique, the transformation is cleared.
Optionally apply another technique. To apply multiple techniques in series within a trial, in the app toolstrip, select the new technique from the gallery. Adjust the parameter values for the technique, and select Run. The display panes and Transformation pane updates to show the new cumulative registration results. To assess the incremental impact of the step, click between the current and previous step in the Registration Browser. To undo the addition of the second technique, in the app toolstrip, select Delete Last Step.
To apply the second technique in a separate trial, in the app toolstrip, select Add Trial > Add Trial with Manual Registration. This action creates a new trial without the first automated technique, but with the initial manual transformation you created in the previous section. If you do not want to include the manual transformation, select Add Trial > Add Empty Trial. Select your parameter values, and select Run. You can compare the results of the two trials by clicking between them in the Registration Browser.

This table describes the available automated registration techniques. Expand the Parameters section to view details about each parameter for the technique.

Technique	Description
`Monomodal`	Intensity-based registration of volumes acquired using the same imaging modality. Equivalent to `imregister` with a `RegularStepGradientDescent` optimizer and `MeanSquares` metric. Parameters Normalize — Normalize volume data to the range [0, 1] before registration. Apply Gaussian Blur — Smooth data before registration. Sigma — Standard deviation of Gaussian blur filter. Transformation — Type of transformation to estimate (similarity, affine, rigid, translation). Pyramid Levels — Number of downsampled pyramid levels to use during registration. Gradient Magnitude Tolerance — Tolerance to determine whether optimizer has reached a plateau. Minimum Step Length — Minimum step size used by the optimizer. Decrease to increase the likelihood of convergence to an accurate result, at the cost of increased computation time. Maximum Step Length — Maximum step size used by the optimizer. Increase value to increase computation speed, at the cost of increased chance of failure to converge. Maximum iterations — Maximum number of iterations performed by the optimizer at each pyramid level. Relaxation Factor — Scalar in the range [0, 1] that defines the rate at which the step size changes. Display Optimization — Display optimization progress in the command window.
`Multimodal`	Intensity-based registration of volumes acquired using the same imaging modality. Equivalent to `imregister` with a `OnePlusOneEvolutionary` optimizer and `MattesMutualInformation` metric. Parameters Normalize — Normalize volume data to the range [0, 1] before registration. Apply Gaussian Blur — Smooth data before registration. Sigma — Standard deviation of Gaussian blur filter, in the range [0, 1]. Transformation — Type of transformation to estimate (similarity, affine, rigid, translation). Pyramid Levels — Number of downsampled pyramid levels to use during registration. Use All Voxels — Select to use intensities from all overlapping voxels when computing the mutual information metric, or clear to select a random subsample of voxels. Spatial Samples — Number of random samples to use to calculate the mutual information metric when Use All Voxels is selected. Histogram Bins — Number of histogram bins used to calculate the joint distribution histogram while computing the mutual information metric. Growth Factor — Rate at which search radius grows during optimization. Epsilon — Minimum step size for the search radius. Initial Radius — Initial search radius for the optimizer. A larger value decreases computation time, but increases the likelihood that the optimizer fails to converge. Maximum iterations — Maximum number of iterations performed by the optimizer at each pyramid level. Display Optimization — Display optimization progress in the command window.
`Moment of Mass`	Fast, similarity registration using the moment of mass method. Equivalent to `imregmoment`. Parameters Normalize — Normalize volume data to the range [0, 1] before registration. Apply Gaussian Blur — Smooth data before registration. Sigma — Standard deviation of Gaussian blur filter, in the range [0, 1]. Median Threshold Bitmap — Perform median threshold bitmap processing before registration, which can improve results if the volumes have different intensity ranges or have been acquired using different imaging modalities.
`Demons`	Estimate nonrigid displacement field using demons deformable registration. Equivalent to `imregdemons`. Parameters Pyramid Levels — Number of downsampled pyramid levels to use during registration. Iterations — Number of downsampled pyramid levels to use during registration. Smoothing — Smoothing applied at each iteration, typically in the range [0.5, 3]. Larger values create a smoother displacement field, and smaller values create a more localized displacement field.
`Deformable`	Estimate nonrigid displacement field using the total variation method. Equivalent to `imregdeform`. Parameters Grid Spacing — Grid spacing for the displacement field, where smaller values enable a finer grid resolution. Voxel Size — Size of voxels in the volumes, in world units such as millimeters. By default, the app uses the voxel spacing from the metadata of the fixed volume. Pyramid Levels — Number of downsampled pyramid levels to use during registration. Grid Regularization — Weight for the grid displacement regularization, where larger values create a smoother displacement field, and smaller values create a more localized displacement field. Display Progress — Display progress in the command window.
`Surface`	Register surfaces using iterative closest point method. Equivalent to `imregicp`. Parameters Normalize — Normalize volume data to the range [0, 1] before registration. Apply Gaussian Blur — Smooth data before registration. Sigma — Standard deviation of Gaussian blur filter. Isovalue — Isovalue used to define the surface for registration. Minimization Metric — Minimization metric. Use `PointToPoint` for all nonplanar surfaces. For planar surfaces, using `PointToPlane` can reduce the number of iterations and decrease computation time. Maximum iterations — Maximum number of iterations. Outlier Removal — Vector [n, r], where surface points with fewer than n neighbors within a radius r are removed before registration. Distance Threshold — Maximum distance for the algorithm to search for matching points between volumes. Higher values can help register sparse surfaces, while lower values can improve registration accuracy. Display Progress — Display progress in the command window.

Export Results

You can export the results of a specific trial to the workspace, or generate a function that recreates the registration steps of a trial that you can apply outside the app.

To export results to the workspace, follow these steps:

In the app toolstrip, select Export > To Workspace.
In the dialog box, specify the name of the variable to save. In the workspace, the app creates a structure with two fields.
- The tform field contains the transformation object for the transformation of the last step of the selected trial. The tform field is an affinetform3d object for affine transformations, and a 4-D displacement field for deformable registrations.
- The regVol field contains the registered moving volume as a medicalVolume object.

Export to workspace dialog box

To export a function, in the app toolstrip, select Export > Generate Function. Any manual registrations are included as hardcoded values, which might not generalize to other pairs of volumes.