EMS

Expectation-Maximization Segmentation

Fully automated model-based segmentation of MR images of the brain

Developed by Koen Van Leemput at Medical Imaging Computing, Leuven, Belgium

Now press the button 'Check Reg' and select all three images. You should see something similar to the figure on the left. Try moving the cross-hairs by clicking on one of the views. If you do not see anything, make sure that you entered all the fields in the header file correctly.

Since the images are simulated, the three modalities are perfectly well co-registered. However, it is unlikely that this will be the case in real circumstances, where the patient might have moved in between the acquisition of the images. This section explains how to co-register the different modalities in such cases, and gives at the same time an understanding of how affine spatial transformations are handled in SPM.

Suppose that the PD- and T2-weighted images are acquired simultaneously but at a different time than the T1-weighted image. The spatial displacement between the PD/T2-pair and the T1-image, resulting from a different patient positioning at the two scanning sessions, can easily be simulated in SPM as follows. Click on the button 'Display' in the SPM menu window, and select the PD-weighted image. In the field 'pitch {rad}', fill in for instance '20/180*pi' and press ENTER. Subsequently, fill in '20' in the field 'right {mm}' and press again ENTER. This will cause the image to rotate 20 degrees and to shift 20 mm to the right. Press 'Reorient images...' and select both the T2- and the PD-weighted image. If you look at the files in the directory where the images are located, you will see that there now also exists a file 'T2.mat' and a file 'PD.mat'. These '.mat'-files contain a 4x4 affine transformation matrix that maps a specific voxel location (in voxel numbers) into a world location (in mm). Now press the button 'Check Reg' and select all three images. You should see something similar to the figure on the left: the PD- and T2-weighted images are now moved with respect to the T1-weighted image.

To co-register the images, press the 'Register' button in the EMS menu window (not in the SPM menu window!). Enter '1' for 'number of subjects', 'no' for 'Normalize to atlas?', select 'T2.img' or 'PD.img' for target image, 'T1.img' for object image, and nothing for "other" images. Since the images are all acquired from the same individual, we are looking for a rigid-body transformation, so select '6 params (Rigid Body)' for the number of affine parameters. This will start the coregistration of the T1-weighted with the T2- or PD-weighted image based on maximization of mutual information.

When the registration is finished, a file 'T1.mat' is created that contains the updated affine mapping from voxel location in the T1-weighted image into world location. Use the button 'Check Reg' to check that all images are again in spatial correspondence with each other.

Finally, press the 'Segment' button in the EMS menu window (not in the SPM menu window!), select '1' for 'number of subjects', all three images (order is not important), '4' for the order of the bias field polynomial model, '3D' for the type of polynomial, and 'no' for 'Use Markov random field?'. Now our multi-spectral data set is automatically segmented; the results are written in files with suffix '_seg' and '_bias' for the tissue class probability images and the estimated bias fields, respectively. Using the 'Check Reg' option of SPM, and selecting the first three class probability maps and the three bias field images will result in the figure shown on the left. You can also calculate the bias corrected images by pressing 'Correct' in the EMS menu window and selecting the original image files. The bias corrected images are written to file with suffix '_corr'.