Page 55 - Fister jr., Iztok, and Andrej Brodnik (eds.). StuCoSReC. Proceedings of the 2017 4th Student Computer Science Research Conference. Koper: University of Primorska Press, 2017
P. 55
demonstrate it on BrainWeb images.
We started by creating a new project on the top of the reg-
istration environment. Without any other interference we
initialized the graphical user interface wich is shown in fig-
ure 4.
Figure 2: Image histogram as an example of the first
set of tools.
Figure 3: Absolute difference as an example of the Figure 4: Graphical user interface.
second set of tools.
After that we loaded the reference and moving image into
The graphical user interface is build with Matlab GUI [6] the registration data structure through the user interface.
toolbox and can be easily initialized within Matlab workspace. For the reference image we used a MRI image of brain with
181 × 217 × 181 voxels, 1mm slice thickness and with 3% of
2.3 Registration Toolbox noise from the BrainWeb database. To create the moving
image we applied a rigid rotation of 10 degrees over the ref-
Image registration procedures typically consists of three com- erence image. The user interface displayed images as show
ponents. The first component is the geometric transforma- in figure 1. Without any additional coding we could slide
tion which describes the movement, rotation, translation through all slices of the reference and moving image. In ad-
and/or deformation of objects in the image. The second dition to that we analyzed the relation between the two im-
component is the similarity metrix, which is used to esti- ages by displaying the joint intensity distribution histogram
mate degree of image alignement. The third component is or the absolute difference as shown in figure 3.
the optimization process. The goal of it, is to find the opti-
mal geometric transformation which maximizes or minimizes At this point we only needed to focus on the core imple-
the similarity of two images. mentation of the rigid registration procedure. The step of
creating the necessary environment was done in a very short
Implementation of these components may require a lot of time. To implement the rigid registration procedure we cre-
effort if not supported by lower level processing functions. ated three components. The first component was designed to
We have collected a large set of such lower level processing controll the overall registration workflow. The second com-
functions that were implemented in past research activities. ponent was designed to compute the similarity between the
This functions are grouped into a toolbox that can be used reference and the moving image. This component was im-
through the graphical user interface or directly in the Matlab plemented by using low level processing functions from the
workspace. This toolbox will be constantly updated accord- integrated toolbox. The third component was the optimiza-
ing to the needs of future image registration procedures. tion process. It was implemented with the free/open-source
NLopt [5] library which provides implementation of several
3. RESULTS optimization methods.
We have tested the image registration environment by using After we implemented the rigid registration procedure on the
it for the development of a rigid registration procedure and top of the registration data structure we started testing the
implementation. In the graphical user interface we selected
the implemented rigid registration method as shown in figure
5. We executed the procedure many times until we solved
all issues and got the preferred results as shown in figure 6.
4. DISCUSSION AND CONCLUSION
From the described results we can conclude that the reg-
istration library has satisfied all the initial requirements.
The predefined data structure helped to implement more
structured registration methods and to easily display regis-
StuCoSReC Proceedings of the 2017 4th Student Computer Science Research Conference 55
Ljubljana, Slovenia, 11 October
We started by creating a new project on the top of the reg-
istration environment. Without any other interference we
initialized the graphical user interface wich is shown in fig-
ure 4.
Figure 2: Image histogram as an example of the first
set of tools.
Figure 3: Absolute difference as an example of the Figure 4: Graphical user interface.
second set of tools.
After that we loaded the reference and moving image into
The graphical user interface is build with Matlab GUI [6] the registration data structure through the user interface.
toolbox and can be easily initialized within Matlab workspace. For the reference image we used a MRI image of brain with
181 × 217 × 181 voxels, 1mm slice thickness and with 3% of
2.3 Registration Toolbox noise from the BrainWeb database. To create the moving
image we applied a rigid rotation of 10 degrees over the ref-
Image registration procedures typically consists of three com- erence image. The user interface displayed images as show
ponents. The first component is the geometric transforma- in figure 1. Without any additional coding we could slide
tion which describes the movement, rotation, translation through all slices of the reference and moving image. In ad-
and/or deformation of objects in the image. The second dition to that we analyzed the relation between the two im-
component is the similarity metrix, which is used to esti- ages by displaying the joint intensity distribution histogram
mate degree of image alignement. The third component is or the absolute difference as shown in figure 3.
the optimization process. The goal of it, is to find the opti-
mal geometric transformation which maximizes or minimizes At this point we only needed to focus on the core imple-
the similarity of two images. mentation of the rigid registration procedure. The step of
creating the necessary environment was done in a very short
Implementation of these components may require a lot of time. To implement the rigid registration procedure we cre-
effort if not supported by lower level processing functions. ated three components. The first component was designed to
We have collected a large set of such lower level processing controll the overall registration workflow. The second com-
functions that were implemented in past research activities. ponent was designed to compute the similarity between the
This functions are grouped into a toolbox that can be used reference and the moving image. This component was im-
through the graphical user interface or directly in the Matlab plemented by using low level processing functions from the
workspace. This toolbox will be constantly updated accord- integrated toolbox. The third component was the optimiza-
ing to the needs of future image registration procedures. tion process. It was implemented with the free/open-source
NLopt [5] library which provides implementation of several
3. RESULTS optimization methods.
We have tested the image registration environment by using After we implemented the rigid registration procedure on the
it for the development of a rigid registration procedure and top of the registration data structure we started testing the
implementation. In the graphical user interface we selected
the implemented rigid registration method as shown in figure
5. We executed the procedure many times until we solved
all issues and got the preferred results as shown in figure 6.
4. DISCUSSION AND CONCLUSION
From the described results we can conclude that the reg-
istration library has satisfied all the initial requirements.
The predefined data structure helped to implement more
structured registration methods and to easily display regis-
StuCoSReC Proceedings of the 2017 4th Student Computer Science Research Conference 55
Ljubljana, Slovenia, 11 October
