Difference between revisions of "3D Reslicing using COMKAT image tool (basic)"

From COMKAT wiki
Jump to navigation Jump to search
 
(8 intermediate revisions by one other user not shown)
Line 9: Line 9:
 
sliceVolume() is a mex-file written in c with an interface to MATLAB that makes the operation particularly efficient.  COMKATImageTool uses sliceVolume() and you can use it too.
 
sliceVolume() is a mex-file written in c with an interface to MATLAB that makes the operation particularly efficient.  COMKATImageTool uses sliceVolume() and you can use it too.
  
 +
[[Image:DICOM_coordinate_system.png]]
  
 
----
 
----
Line 18: Line 19:
  
  
Create lists of indices of all pixels in the 2D slice (rectangular grid) that we are creating
+
Create lists of indices of all pixels in the 2D slice (rectangular grid) that we are creating.
  
 
   [i, j] = meshgrid(0 : Nc-1, 0 : Nr-1); % i and j will be 2D arrays, meshgrid is a function built into MATLAB
 
   [i, j] = meshgrid(0 : Nc-1, 0 : Nr-1); % i and j will be 2D arrays, meshgrid is a function built into MATLAB
Line 25: Line 26:
 
The first and second rows of ij are the indices corresponding to column and row indices of all voxels in the desired slice. So the dimension of ij is 3 x ( # of desired voxels ).
 
The first and second rows of ij are the indices corresponding to column and row indices of all voxels in the desired slice. So the dimension of ij is 3 x ( # of desired voxels ).
  
For example, the first column of ij could be [ 0 ; 0 ] ; the second column could be [ 0 ; 1 ] , etc.
+
*Note: the first column of ij could be [ 0 ; 0 ] ; the second column could be [ 0 ; 1 ] , etc.  
  
  
  
Compute the physical x,y,z locations, in mm, from pixel indices according to the DICOM coordinate system ref [http://medical.nema.org/dicom/2004/04_03PU.PDF] p. 275.
+
Then, we compute the physical x,y,z locations, in mm, from desired voxel indices according to the DICOM coordinate system ref [http://medical.nema.org/dicom/2004/04_03PU.PDF] p. 275.
 
+
According the ref, the coordinate transformation matrix, M, is listed as follows:
This is a two-step process. The first step is to generate the coordinate transformation matrix, M.  Note that pixel spacing/zoom, orientation, and position for the slice are specified in the transformation matrix.
 
 
[[Image:Eq_M.png]]
 
[[Image:Eq_M.png]]
  
  
The transformation matrix, M, can be further separated into three parts as follows:
+
The transformation matrix can be further separated into three parts.
  
 
[[Image:Eq_M2.png]]
 
[[Image:Eq_M2.png]]
Line 47: Line 47:
 
   *  xyz is a matrix consisted of three rows the physical x, y, and z location (m) of all the desired voxels. The dimension of xyz is 3 x ( # of desired voxels ).
 
   *  xyz is a matrix consisted of three rows the physical x, y, and z location (m) of all the desired voxels. The dimension of xyz is 3 x ( # of desired voxels ).
  
These xyz locations are the same as those in the image volume that is being sliced to make the 2D image.  From these xyz locations, we find the corresponding 3D indices, (u,v,w), into the volume.  This uses the transformation matrix for the volume, Mhat, that relates the indices to the xyz physical location.  This is analogous to M used for the desired slice but here the pixel spacing, orientation, and position indicate how the volume data are stored.
+
These xyz locations are the same as those in the image volume being sliced to make the 2D image.  From these xyz locations, we find the corresponding 3D indices, (u,v,w), in the volume.  This uses the transformation matrix for the volume, Mhat, that relates the indices to the xyz physical location.  This is analogous to M used for the desired slice but here the pixel spacing, orientation, and position indicate how the volume data are stored.
  
 
The spatial mapping between the indices in the original image volume (uvw) and xyz can be described as
 
The spatial mapping between the indices in the original image volume (uvw) and xyz can be described as
  
   xyz = Mhat * uvw.
+
   xyz = Mhat * uvw;
  
Therefore, Mhat is the matrix mapping  uvw  to its physical location (mm) and can be separated into three parts, M'<sub>orientation</sub>, M'<sub>spacing</sub>, and M'<sub>position</sub>, as the transformation M.
+
Therefore, Mhat is the matrix mapping  uvw  to its physical location (mm) and can be separated into three parts, M'<sub>orientation</sub>, M'<sub>spacing</sub>, and M'<sub>position</sub>, as done for the transformation M.
 
* Note: M'<sub>orientation</sub>, M'<sub>spacing</sub>, and M'<sub>position</sub> are the mapping matrices between  uvw  and  xyz  which are different from the mapping matrices between  ij  and  xyz.  
 
* Note: M'<sub>orientation</sub>, M'<sub>spacing</sub>, and M'<sub>position</sub> are the mapping matrices between  uvw  and  xyz  which are different from the mapping matrices between  ij  and  xyz.  
  
Line 67: Line 67:
  
  
Separate uvw into the components
+
In order to use sliceVolume() to do the interpolation, we need to separate uvw into the components
 
     u = ( row indices in the domain of original of volume data );
 
     u = ( row indices in the domain of original of volume data );
 
     v = ( column indices in the domain of original of volume data );
 
     v = ( column indices in the domain of original of volume data );
 
     w = ( plane indices in the domain of original of volume data );
 
     w = ( plane indices in the domain of original of volume data );
Therefore, the u, v, w are the first, second, and third rows of uvw.
+
 
 
+
    *Note: the u, v, w are the first, second, and third rows of uvw.
  
Use sliceVolume() to interpolate the desired slice
 
  
  slice = sliceVolume(idv, v, u, w, rawbackgroundPixelValue , 'linear'); 
+
Before doing the final step, interpolation, we have to set a unscaled raw background value of images, rawbackgroundPixelValue. The value is used for the interpolation when it is nothing there for calculating the interpolation value.  
rawbackgroundPixelValue is a value that used for the interpolation when it is nothing there for calculating the interpolation value.
 
The value is a unscaled raw background value of images.
 
  
 
In gerneral, rawbackgroundPixelValue = (scaled_pixel_value -  rescale_intercept) / (rescale_slope) ;
 
In gerneral, rawbackgroundPixelValue = (scaled_pixel_value -  rescale_intercept) / (rescale_slope) ;
  
e.g. For PET, the scaled background value is usually zero.  
+
For PET images, the scaled background value is usually zero.  
  
 
  ==>    rawbackgroundPixelValue  = ( 0 - rescale_intercept) / rescale_slope ;
 
  ==>    rawbackgroundPixelValue  = ( 0 - rescale_intercept) / rescale_slope ;
Line 92: Line 89:
 
    
 
    
 
   rawbackgroundPixelValue  = - o / s;
 
   rawbackgroundPixelValue  = - o / s;
 +
 +
 +
Finally, we use sliceVolume() to interpolate the desired slice
 +
 +
  slice = sliceVolume(idv, v, u, w, rawbackgroundPixelValue , 'linear'); 
 +
  
  
Line 108: Line 111:
  
 
   [u, v, w] = coordinateGen(ivd, Nc, Nr, pixelSpacing, planePos, orientation); % Input the desired pixelSpacing, planePos and orientation matrices
 
   [u, v, w] = coordinateGen(ivd, Nc, Nr, pixelSpacing, planePos, orientation); % Input the desired pixelSpacing, planePos and orientation matrices
 
+
 
 +
  *Note: pixelSpacing, planPos, and orientation are the M<sub>spacing</sub>, M<sub>position</sub>, and M<sub>orientation</sub>, respectively.
  
 
Use sliceVolume() to interpolate
 
Use sliceVolume() to interpolate
Line 120: Line 124:
 
----
 
----
 
====Examples (Cardiac-PET)====
 
====Examples (Cardiac-PET)====
We're going to describe four examples. We'll use the approach I to generate the new slice in the first example and use the approach II to accomplish the others.  
+
We're going to describe four examples. The approach I is going to be used to generate the new slice in the first example, and we'll use the approach II to accomplish the others.  
A set of DICOM cardiac-PET data is going to be used as an example.
+
A set of DICOM cardiac-PET data is going to be used in those examples.
 +
 
  
 
=====Obtain a new slice=====
 
=====Obtain a new slice=====
Line 135: Line 140:
  
  
Create the indices of each voxel.
+
Create the indices of each voxel in the desired slice.
 
   [i, j] = meshgrid(0 : Nc-1, 0 : Nr-1);
 
   [i, j] = meshgrid(0 : Nc-1, 0 : Nr-1);
 
       ij = [i(:)’ ; j(:)’];  
 
       ij = [i(:)’ ; j(:)’];  
Line 147: Line 152:
  
  
Set the orientation and pixelSpacing matrix.
+
Set the orientation and pixel spacing matrix.
 
   orientationInput = [1, 0;  
 
   orientationInput = [1, 0;  
 
                                   0, 1;
 
                                   0, 1;
Line 154: Line 159:
 
   pixelSpacingOrg = diag([pixelSize(2), pixelSize(1)]);
 
   pixelSpacingOrg = diag([pixelSize(2), pixelSize(1)]);
  
As we mentioned in the approach I, we need to map the indies from  ij  to  xyz  by using the transformation matrix, M. Now, we're going to calculate physical locations (mm) of voxels in the desired image.  
+
 
 +
As mentioned in the approach I, we need to map the indies from  ij  to  xyz  by using the transformation matrix, M.  
 
   xyz = orientationInput * pixelSpacingOrg* ij + repmat(planePosOrg ,1 , Nr * Nc);
 
   xyz = orientationInput * pixelSpacingOrg* ij + repmat(planePosOrg ,1 , Nr * Nc);
  
  
  
After obtaining the physical locations of each index, we need to calculate the indices of each voxel in the original volume space to perform inverse mapping approach. As we mentioned in the approach I, the matrices used for inverse mapping are different from the matrices mapping  ij  to  xyz.
+
After obtaining the physical locations of each index, we need to calculate the indices of each voxel in the original volume space to perform inverse mapping approach.
  
We need to obtain the original pixel size, orientation, and image position matrices for the inverse mapping. You can specify your original volume information. Here we use function get() to obtain the information we need from the IVD.
+
As we mentioned in the approach I, the matrices used for inverse mapping are different from the matrices mapping from  ij  to  xyz. For this reason, we need to obtain the original pixel size, orientation, and image position matrices for the inverse mapping (you can specify your original volume information. Here we use function get() to obtain the information we need from the IVD).
 
   [ny, nx, nz, nf] = get(ivd, 'VolumeDimension');  
 
   [ny, nx, nz, nf] = get(ivd, 'VolumeDimension');  
 
   pixDim = get(ivd, 'PixelSpacing');  % get pixelSpacing
 
   pixDim = get(ivd, 'PixelSpacing');  % get pixelSpacing
Line 178: Line 184:
  
  
In order to use sliceVolume() to do the interpolation, we need to separate uvw in three separate matrices.
+
In order to use sliceVolume() to do the interpolation, we need to separate uvw in three matrices.
 
   u =  ( reshape(uvw(1,:), [Nr, Nc]) ) + 1;
 
   u =  ( reshape(uvw(1,:), [Nr, Nc]) ) + 1;
 
   v =  ( reshape(uvw(2,:), [Nr, Nc]) ) + 1;
 
   v =  ( reshape(uvw(2,:), [Nr, Nc]) ) + 1;
 
   w =  ( reshape(uvw(3,:), [Nr, Nc]) ) + 1;
 
   w =  ( reshape(uvw(3,:), [Nr, Nc]) ) + 1;
 
    
 
    
   * Note: the indices in original image volume are defined to start from 1. After performing the inverse mapping, the indices need to be add by 1.
+
   * Note: the indices in original image volume are defined to start from 1. After performing the inverse mapping, the indices need to be added by 1.
  
  
  
As we described in the approach I, the raw background value of PET can be calculated as follows:
+
As we described before, the raw background value of PET can be calculated as follows:
  
 
   s = get(ivd, 'VolumeFrameBufferScaleFactor');
 
   s = get(ivd, 'VolumeFrameBufferScaleFactor');
Line 195: Line 201:
  
  
Finally, we can interpolate the new slice using the sliceVolume().   
+
Finally, we can interpolate the new slice using sliceVolume().   
 
   slice = sliceVolume(ivd, v , u , w, rawBackgroundPixelValue, 'linear');
 
   slice = sliceVolume(ivd, v , u , w, rawBackgroundPixelValue, 'linear');
  
  
Show the new slice (the new image could be looked like the first image in the figure).
+
Show the new slice (the new image can be seen the first image in the figure).
 
   figure, imagesc(slice), axis image, colormap(hot) % show org image
 
   figure, imagesc(slice), axis image, colormap(hot) % show org image
  
Line 272: Line 278:
  
  
Firstly, we create the rotation matrix. You can find more detail in [http://en.wikipedia.org/wiki/Rotation_matrix].
+
Firstly, we create the rotation matrix. You can find more detail about the rotation matrix in [http://en.wikipedia.org/wiki/Rotation_matrix].
 
Let's try to rotate the image counter-clockwisely about z-axis.
 
Let's try to rotate the image counter-clockwisely about z-axis.
 
   rotMat= [ cos(rotAngle), -sin(rotAngle),      0 ;
 
   rotMat= [ cos(rotAngle), -sin(rotAngle),      0 ;
 
                   sin(rotAngle),  cos(rotAngle),      0 ;
 
                   sin(rotAngle),  cos(rotAngle),      0 ;
 
                                     0,                    0,      1 ];
 
                                     0,                    0,      1 ];
 +
  
 
By combining the rotation matrix and the original orientation marix, we will get a new orientation matrix with rotation.
 
By combining the rotation matrix and the original orientation marix, we will get a new orientation matrix with rotation.
 
   orientationInputRot = rotMat * orientationInput;
 
   orientationInputRot = rotMat * orientationInput;
 +
  
 
After rotation, we adjust the center of the rotated image.
 
After rotation, we adjust the center of the rotated image.
Line 300: Line 308:
 
----
 
----
  
The slicing images may look like this:
+
The results of the slicing images:
  
 
[[Image:Fig_eg3Dslicing.png]]
 
[[Image:Fig_eg3Dslicing.png]]
  
 
----
 
----

Latest revision as of 16:09, 16 August 2018

Reslicing 3D image volume using COMKAT image tool (basic)

Overview

Reslicing a 3D (or 3D vs time) image dataset can be accomplished using various components of COMKAT including the function sliceVolume() which is a method of ImageVolumeData. This example explains how to create a 2D image by sampling (slicing) from a volume at a position, plane orientation, and magnification specified by the user. The approach is to load the image volume dataset into an instance of an ImageVolumeData (abbreviated IVD) object and to use the sliceVolume() method.

Background

sliceVolume() is a mex-file written in c with an interface to MATLAB that makes the operation particularly efficient. COMKATImageTool uses sliceVolume() and you can use it too.

DICOM coordinate system.png


Approach I. Demonstrate the method for coordinate transformations

Create an instance of an IVD object and load it with an image volume.

  ivd = ImageVolumeData(); % create an instance of an ImageVolumeData object
  ivd = read_DICOM(ivd, pathName, fileName); % load volume into an instance of IVD object;


Create lists of indices of all pixels in the 2D slice (rectangular grid) that we are creating.

  [i, j] = meshgrid(0 : Nc-1, 0 : Nr-1); % i and j will be 2D arrays, meshgrid is a function built into MATLAB
  ij = [i(:)' ; j(:)'];  % make matrix, each column corresponding to a single pixel in the slice we are creating

The first and second rows of ij are the indices corresponding to column and row indices of all voxels in the desired slice. So the dimension of ij is 3 x ( # of desired voxels ).

  • Note: the first column of ij could be [ 0 ; 0 ] ; the second column could be [ 0 ; 1 ] , etc.


Then, we compute the physical x,y,z locations, in mm, from desired voxel indices according to the DICOM coordinate system ref [1] p. 275. According the ref, the coordinate transformation matrix, M, is listed as follows: Eq M.png


The transformation matrix can be further separated into three parts.

Eq M2.png


Once we set the transformation matrix (or the orientation, pixel spacing, and image position matrices), we can use the transformation matrix to calculate the physical (mm) locations of each voxel in the desired slice.

  xyz = M * ij; 
      or
  xyz = Morientation * Mspacing * ij + Mposition  ; 
  
  *  xyz is a matrix consisted of three rows the physical x, y, and z location (m) of all the desired voxels. The dimension of xyz is 3 x ( # of desired voxels ).

These xyz locations are the same as those in the image volume being sliced to make the 2D image. From these xyz locations, we find the corresponding 3D indices, (u,v,w), in the volume. This uses the transformation matrix for the volume, Mhat, that relates the indices to the xyz physical location. This is analogous to M used for the desired slice but here the pixel spacing, orientation, and position indicate how the volume data are stored.

The spatial mapping between the indices in the original image volume (uvw) and xyz can be described as

  xyz = Mhat * uvw;

Therefore, Mhat is the matrix mapping uvw to its physical location (mm) and can be separated into three parts, M'orientation, M'spacing, and M'position, as done for the transformation M.

  • Note: M'orientation, M'spacing, and M'position are the mapping matrices between uvw and xyz which are different from the mapping matrices between ij and xyz.


The indices of the desired slice can be calculated by performing the inverse mapping.

  uvw =  inv( Mhat ) * xyz;    
      or
  uvw = inv( M'spacing ) * inv( M'orientation ) * (xyz - M'position ) ; 
  
  * uvw is the matrices consisted of the indices in the domain of the original volume and is consisted of three rows corresponding to row, column, and plane indices of the original volume data.
     The dimension of uvw is identical of that of xyz.


In order to use sliceVolume() to do the interpolation, we need to separate uvw into the components

    u = ( row indices in the domain of original of volume data );
    v = ( column indices in the domain of original of volume data );
   w = ( plane indices in the domain of original of volume data );
  
   *Note: the u, v, w are the first, second, and third rows of uvw.


Before doing the final step, interpolation, we have to set a unscaled raw background value of images, rawbackgroundPixelValue. The value is used for the interpolation when it is nothing there for calculating the interpolation value.

In gerneral, rawbackgroundPixelValue = (scaled_pixel_value - rescale_intercept) / (rescale_slope) ;

For PET images, the scaled background value is usually zero.

==>    rawbackgroundPixelValue  = ( 0 - rescale_intercept) / rescale_slope ;

So rawbackgroundPixelValue can be calculated as follows:

  s = get( ivd, 'VolumeFrameBufferScaleFactor');           % rescale_slope
  o = get( ivd, 'VolumeFrameBufferRescaleIntercept');  % rescale_intercept
  
  rawbackgroundPixelValue  = - o / s;


Finally, we use sliceVolume() to interpolate the desired slice

  slice = sliceVolume(idv, v, u, w, rawbackgroundPixelValue , 'linear');  




Approach II. Use coordinateGen() to do the coordinate transformation

  • This should create same result as approach I but require fewer lines of coding since coordinateGen() does most things that are needed.


Read the image volume into an ImageVolumeData object

  ivd = ( read data into IVD);


Use coordinateGen() to generate uvw

  [u, v, w] = coordinateGen(ivd, Nc, Nr, pixelSpacing, planePos, orientation); % Input the desired pixelSpacing, planePos and orientation matrices
  
  *Note: pixelSpacing, planPos, and orientation are the Mspacing, Mposition, and Morientation, respectively.

Use sliceVolume() to interpolate

  slice = sliceVolume(idv, v, u, w, backgroundPixelValue, 'linear');





Examples (Cardiac-PET)

We're going to describe four examples. The approach I is going to be used to generate the new slice in the first example, and we'll use the approach II to accomplish the others. A set of DICOM cardiac-PET data is going to be used in those examples.


Obtain a new slice

First, let's use the approach I to obtain a new slice.

  ivd = ImageVolumeData(); % create an instance of an ImageVolumeData object
  ivd = read_DICOM(ivd, pathName, fileName); % load DICOM volume into an instance of IVD object


To obtain a new slice, we need to specific the number of rows and columns in the new image.

  Nc = 370;  % # of columns
  Nr = 370;  % # of rows


Create the indices of each voxel in the desired slice.

  [i, j] = meshgrid(0 : Nc-1, 0 : Nr-1);
      ij = [i(:)’ ; j(:)’]; 


Specify the sampling pixel size and the position of sampling plane.

  pixelSize = [1.5501; 1.5501];   % Set pixel size (mm)
  
  planePosOrg = [-0.7751; -0.7751; 88.0000];   % Set the reslicing plane location


Set the orientation and pixel spacing matrix.

  orientationInput = [1, 0; 
                                 0, 1;
                                 0, 0];     
  
  pixelSpacingOrg = diag([pixelSize(2), pixelSize(1)]);


As mentioned in the approach I, we need to map the indies from ij to xyz by using the transformation matrix, M.

  xyz = orientationInput * pixelSpacingOrg* ij + repmat(planePosOrg ,1 , Nr * Nc);


After obtaining the physical locations of each index, we need to calculate the indices of each voxel in the original volume space to perform inverse mapping approach.

As we mentioned in the approach I, the matrices used for inverse mapping are different from the matrices mapping from ij to xyz. For this reason, we need to obtain the original pixel size, orientation, and image position matrices for the inverse mapping (you can specify your original volume information. Here we use function get() to obtain the information we need from the IVD).

  [ny, nx, nz, nf] = get(ivd, 'VolumeDimension'); 
  pixDim = get(ivd, 'PixelSpacing');  % get pixelSpacing
  vol_Uspacing = pixDim(2);
  vol_Vspacing = pixDim(1);
  vol_Wspacing = pixDim(3);
  
  vol_pos  = get(ivd, 'ImagePositionPatient');  % get patient position
  vol_orient  = get(ivd, 'ImageOrientationPatient');  % get orientation


Therefore, the inverse mapping from the matrix xyz to uvw can be done as it has been described in the approach I.

  uvw =  diag([1./vol_Uspacing, 1./vol_Vspacing, 1./vol_Wspacing]) * inv(vol_orient) * ( xyz - repmat(vol_pos, 1, Nr * Nc) );
  
  * Note: uvw is a matrix consisted of indices in the original image space.


In order to use sliceVolume() to do the interpolation, we need to separate uvw in three matrices.

  u =  ( reshape(uvw(1,:), [Nr, Nc]) ) + 1;
  v =  ( reshape(uvw(2,:), [Nr, Nc]) ) + 1;
  w =  ( reshape(uvw(3,:), [Nr, Nc]) ) + 1;
  
  * Note: the indices in original image volume are defined to start from 1. After performing the inverse mapping, the indices need to be added by 1.


As we described before, the raw background value of PET can be calculated as follows:

  s = get(ivd, 'VolumeFrameBufferScaleFactor');
  o = get(ivd, 'VolumeFrameBufferRescaleIntercept');
  
  rawBackgroundPixelValue = -o/s;


Finally, we can interpolate the new slice using sliceVolume().

  slice = sliceVolume(ivd, v , u , w, rawBackgroundPixelValue, 'linear');


Show the new slice (the new image can be seen the first image in the figure).

  figure, imagesc(slice), axis image, colormap(hot) % show org image



Zooming

We're going to use the approach II to zoom-in the slice in a factor of three.


Set the zoom factor.

  zoomFactor = 3;  % set zoom factor


Calculate the new pixel size.

  pixelSpacing = pixelSpacingOrg / zoomFactor;  % adjust pixel size


After zooming you may need to translate the image center, or you may see nothing.

  originalFOV = [Nc; Nr] .* pixelSpacingOrg;  % calculate the original FOV (mm)
  zoomShift = originalFOV / 2 * (zoomFactor - 1) / zoomFactor ;
  planePos = planePosOrg + [zoomShift ; 0] ;   


Generate new slice using coordinateGen() and sliceVolume().

  [u, v, w] = coordinateGen(ivd, Nc, Nr, pixelSpacing, planePos, orientationInput);
  
  slice = sliceVolume(ivd, v , u , w, rawBackgroundPixelValue,'linear');


Show the new slice.

  figure, imagesc(slice), axis image, colormap(hot) % show org +zoom


The new slice is the second image in the figure. It required fewer lines when coding using the approach II.



Translation

Let's try to do the translation using the approach II.


First of all, set the translation matrix.

  transMat = [-35.0; -5; 0];  % the translation in x, y, and z directions


Apply the translation matrix to obtain new plane position.

  planePosTrans = planePos + transMat; 


Generate new slice.

  [u, v, w] = coordinateGen(ivd, Nc, Nr, pixelSpacing, planePosTrans, orientationInput);
  
  slice = sliceVolume(ivd, v , u , w, rawBackgroundPixelValue,'linear');


Show the new slice.

  figure, imagesc(slice), axis image, colormap(hot) % show org +zoom+translation


The third picture in the figure is the result of the new slice.



Rotation

In this example, we're going to rotate the original image (counter-clockwise).


Set the angle we want to rotate.

  rotAngle = 25/180*pi;   % rotate 25 degree


Firstly, we create the rotation matrix. You can find more detail about the rotation matrix in [2]. Let's try to rotate the image counter-clockwisely about z-axis.

  rotMat= [ cos(rotAngle), -sin(rotAngle),       0 ;
                  sin(rotAngle),  cos(rotAngle),       0 ;
                                    0,                     0,       1 ];


By combining the rotation matrix and the original orientation marix, we will get a new orientation matrix with rotation.

  orientationInputRot = rotMat * orientationInput;


After rotation, we adjust the center of the rotated image.

  newFOV = [Nc; Nr] .* pixelSpacing;
  transMat2 = orientationInputRot([1, 2], [1,2]) * ( newFOV / 2 ) - newFOV / 2;  % calculate the translation of the original center after rotation
  planePosTrans2 = planePosTrans - [transMat2 ; 0];  % backward translating the rotated image center


Generate new slice.

  [u, v, w] = coordinateGen(ivd, Nc, Nr, pixelSpacing, planePosTrans2, orientationInputRot);
  slice = sliceVolume(ivd, v , u , w, rawBackgroundPixelValue,'linear');


Show the new slice (the new slice can be seen in the fourth image in the figure).

  figure, imagesc(slice), axis image, colormap(hot) % show org+zoom+translate+rotate




The results of the slicing images:

Fig eg3Dslicing.png