Biomedical Engineering Reference
In-Depth Information
8.3.3.4
Experiments on a Dataset of 18 Subjects
In order to validate the registration method on a larger database, we acquire MRI-
T1 volumetric data of 18 patients. One subject is chosen as the reference subject.
We then perform the registration between the reference volume (source) and
each of the other subjects (target) using
always
the same set of parameters for
the algorithm. Finally, we get 17 reconstructed volumes that can be compared
to the reference volume. We average all the reconstructed volume in order to
have a global overview of the quality of the method.
Average Deformed Volume.
Figures 8.18 and 8.19 present the averaging
between 17 patients after a global affine registration (top), after a quadratic
multigrid registration, i.e., the method without robust estimators (middle), and
the average volume after a robust multigrid registration (bottom). After global
affine registration and averaging, we notice that the internal anatomical struc-
tures are blurred, because the registration is not precise enough. However, after
a robust multigrid registration, we may distinguish precisely the contours of
anatomical structures, such as ventricles, deep nuclei, white matter tracks, and
even cortical regions (sylvian fissure and parietal region for instance).
The comparison between the quadratic registration and the robust registra-
tion shows the benefit of robust functions, because cortical regions are better
registered. The MSE between the reference volume and the averaged volume is
892 for quadratic registration, and drops to 584 for robust registration. We must
note that, considering two subjects, the MSE is not a good absolute measure of
the quality of the registration because of the acquisition (a simple translation
between the two histogram can lead to large MSE). However, the MSE is a good
relative measure to compare two registration processes over a large database.
These experiments clearly show the significant impact of robust estimators.
All the more, it validates the assumption that it is necessary to let discontinuities
appear in the deformation field to register brains correctly. These experiments
also demonstrate the robustness of the method (robustness with respect to the
acquisitions and also with respect to the algorithm parameters) over a realistic
database of subjects.
Overlapping of Brain Tissues.
The evaluation must not be based only on
a measure that is more or less related to the image similarity. Therefore, as in
