21. Coregister with AutoFocus

21.1. Coregistration basics

FireVoxel offers powerful tools for 3D and 4D image coregistration. These tools enable superimposing images of different dimensions, orientations, and even different imaging modalities (e.g., CT and MRI).

In the description below, the stationary image (that remains unchanged during coregistration) will be called the target image.

The source image is the image that is transformed (translated, rotated, scaled, etc.) to match the target image.

Coregistration with AutoFocus is a powerful and versatile method for 3D coregistration. Unlike the simpler coregistration method based on DICOM tags, AutoFocus can be used for coregistering images from different imaging sessions and different modalities.

FireVoxel also offers AutoFocus with motion correction, a AutoFocus-based method for 4D coregistration.

21.2. Introduction to AutoFocus based coregistration

AutoFocus is a scheme that iteratively optimizes a voxel-based similarity measure, such as mutual information.

AutoFocus registration may be done with the help of target ROI, a rough ROI around the organ of interest. The target ROI restricts coregistration to the organ of interest and speeds up the processing, since coregistration is a computationally intensive task.

The target ROI can be created on the target image using the Paintbrush tool to draw rough contours around the organ, followed by ROI > Morph > Fill 2D Contours and Morph Convex. The result will be a 3D ROI enclosing the organ.

Next, the user needs to activate the layer containing the target image. If the target ROI remains the active layer, Register with AutoFocus commands will warn the user of this and ask whether to proceed with coregistration.

Lastly, the user selects Register > [any of Coregister with AutoFocus] commands. These commands require two images. If no suitable image is found, FireVoxel will show an error message, “No suitable source volume found,” (as the active image is considered to be the target image).

The coregistration algorithm searches for a transformation that best matches the source and the target volumes. The command offers a choice of transformations ranging from simple translations to affine.

The matching of the volumes is based on optimizing a similarity measure.

The transformation is computed in two stages: the AutoFocus stage and the fine-tuning stage.

In the AutoFocus stage, the algorithm constructs a variety of transformations with combinations of parameters that span a multidimensional grid. The transformations include translation, scaling, rotation, and shear.

The transformations are ranked by how well they match the two volumes based on the similarity measure.

The number of the best transformations retained for the second, fine-tuning stage can be entered in the Power field.

In the fine-tuning stage, the algorithm performs iterative adjustment of the best transformation parameters until it finds a local optimum of the similarity measure.

Finally, the transformed source image is interpolated and saved as a new layer in the target image window. Let’s set interpolation to tri-linear. The other interpolation options include Nearest Neighbor and various W-sinc methods.

21.3. 3D Registration with AutoFocus Dialog

3D Registration with AutoFocus

Fig. 21.1 3D Registration with AutoFocus.

If a suitable source image are present, the command opens a dialog panel (3D Registration with AutoFocus, Fig. 21.1)

The panel contains the following parts:

21.3.1. Load Initial Transformation

(Browse for) Initial and final .VTF files. Click Dicom tags button to create the initial .vtf file based on matrix size and voxel size from the DICOM header. This .VTF file is created in the Temp directory. ADD DETAILS The final transformation is saved in user-specified .VTF file.

21.3.2. ROI (option)

Use Target ROI (checkbox) - If checked, target ROI is used and its name is displayed. If a visible ROI is present in the target window, by default its name will be listed next to the box labeled Use Target ROI. If the ROI is absent or invisible, this option will be grayed out.

Inflate/units - Text box and drop-down menu indicating grow distance (in voxels or millimeters) by which the target ROI should be inflated by the Grow command.

21.3.3. Measure

This part of the coregistration panel offers a selection of similarity measures in a dropdown menu and related settings.

These similarity measures range from simple to complex and more powerful, yet more computationally intense:

  • Signal Difference

  • Cross Correlation

  • Image Ratio Uniformity

  • Mutual Info

  • Mutual Info Normalized

  • URAL

  • URALTAU

Related settings include:

  • MI bin number - Mutual information bin number… ADD DETAILS

  • Source Noise - ADD DETAILS

  • Target Noise - ADD DETAILS

URAL Settings: (figure)

ADD DETAILS

21.3.4. AutoFocus

Subsample [1, 8] - Default 3

Translation max {X, Y, Z} - Default {30, 30, 30}

Scale Deformation matrix - Default {0, 0, 0} Grid 1

Uniform scale (checkbox) - Default - unchecked

Rotation angle max (deg) - Default {0, 0, 0} Grid 1

Shear Magnitude max [0, 10] - Default 0 Grid 1

21.3.5. Finetune

Power [0, 1000] - Default 20

Multipass (checkbox) - Default - unchecked

Transform (dropdown menu) - Rigid, Affine, Quadratic, Polynomial Multipass

21.3.6. Output Options

Interpolation (dropdown menu) - Nearest neighbor, Tri-linear (default), Wsinc2, Wsinc3, Wsinc4

Reslice Target to Source (checkbox) - Default - unchecked

21.4. Signal Difference with AutoFocus

Shortcut for coregistration with Similarity Measure: Signal Difference

21.5. Cross Correlation with AutoFocus

Shortcut for coregistration with Similarity Measure: Cross Correlation

21.6. Mutual Info with AutoFocus

Shortcut for coregistration with Similarity Measure: Mutual Information

21.7. URAL with AutoFocus

Shortcut for coregistration with Similarity Measure: URAL