Relevant Methods for MRI/X Brain Image Registration
339
refer to both functional and structural aspects conjointly. In particular, MRI
is frequently coupled with CT, MRI atlas and PET. However, a registration
step is required in order to ensure effectively the complementarity of structural
and functional images. Research on registration process is driven either by the
type of attributes (geometric vs. iconic methods), the type of transformation
(rigid vs. non-rigid) or the involved images (monomodal vs. multimodal) (Fig. 1).
The principle of geometric registration methods consists in extracting geometric
primitives from the two images to be registered (e.g. points, curves, surfaces . . .),
whereas iconic registration methods operate directly on the intensities. Furthermore, the rigid registration methods aim to correct the geometric transformations, including translation, rotation, shear, and scaling, whereas the non-rigid
registration methods are carried out using localized stretch of the images. In this
type of transformation, all kinds of deformation fields can be used (e.g. splines,
B-spline, elastic model . . .) [1]. For the monomodal registration methods, the
two images are coming from the same modality (e.g. MRI scans, CT scans . . .),
whilst in the multimodal registration ones, the two images come from two different modalities (e.g. MRI and PET, MRI and CT . . .) [2]. Generally, one of the
key challenges in brain image registration is its veracity. This is because of the
limitations in the registration methods, which are dependent on the quality of
MRI/X parameters as well as the inaccuracy on the non-linear transformations.
In addition to the registration errors, several registrations methods suffer from
the extensive computational cost. To circumvent these limits, atlas-based registration coupled with standard softwares (such as SPM, ITK . . . ) are commonly
used. Indeed, various studies are using this framework in order to investigate
Parkinson disease [3], brain tumors using CT/MRI [4] or PET/MRI [5], and
Alzheimer disease [6]. Additionally, challenges can arise where mono- and multimodal registration is required sequentially. To this end, we evaluate a hybrid
method that may handle mono- and multi-modal registration according to the
same technique. In fact, this paper is dedicated to determine the best tool for
mono- and multi-modal registration for MRI/X brain images, such that X refers
in our case to PET, CT and MRI atlas. We compare three widely brain image
registration tools (SPM, ITK-Snap, 3D Slicer) against an accurate hybrid registration method from the state-of-the-art.
The rest of this paper is organized as follows. Section 2 shows the studied
registration methods. Then, we present the clinical datasets and the evaluation
protocol in Sect. 3. We detail experimental results in Sect. 4. Finally, a conclusion
with some directions for future work are discussed in Sect. 5.
2 Registration Methods
ITK-Snap. Insight Segmentation and Registration Toolkit (ITK-Snap) is a popular tool for segmenting and registering medical images such as MRI, PET and
CT [7]. It is an open source software widely used by clinicians and non-computer
researchers. ITK-Snap allows manual and automatic medical image registration.
This software groups several methods of registration based on the intensity. For
339
refer to both functional and structural aspects conjointly. In particular, MRI
is frequently coupled with CT, MRI atlas and PET. However, a registration
step is required in order to ensure effectively the complementarity of structural
and functional images. Research on registration process is driven either by the
type of attributes (geometric vs. iconic methods), the type of transformation
(rigid vs. non-rigid) or the involved images (monomodal vs. multimodal) (Fig. 1).
The principle of geometric registration methods consists in extracting geometric
primitives from the two images to be registered (e.g. points, curves, surfaces . . .),
whereas iconic registration methods operate directly on the intensities. Furthermore, the rigid registration methods aim to correct the geometric transformations, including translation, rotation, shear, and scaling, whereas the non-rigid
registration methods are carried out using localized stretch of the images. In this
type of transformation, all kinds of deformation fields can be used (e.g. splines,
B-spline, elastic model . . .) [1]. For the monomodal registration methods, the
two images are coming from the same modality (e.g. MRI scans, CT scans . . .),
whilst in the multimodal registration ones, the two images come from two different modalities (e.g. MRI and PET, MRI and CT . . .) [2]. Generally, one of the
key challenges in brain image registration is its veracity. This is because of the
limitations in the registration methods, which are dependent on the quality of
MRI/X parameters as well as the inaccuracy on the non-linear transformations.
In addition to the registration errors, several registrations methods suffer from
the extensive computational cost. To circumvent these limits, atlas-based registration coupled with standard softwares (such as SPM, ITK . . . ) are commonly
used. Indeed, various studies are using this framework in order to investigate
Parkinson disease [3], brain tumors using CT/MRI [4] or PET/MRI [5], and
Alzheimer disease [6]. Additionally, challenges can arise where mono- and multimodal registration is required sequentially. To this end, we evaluate a hybrid
method that may handle mono- and multi-modal registration according to the
same technique. In fact, this paper is dedicated to determine the best tool for
mono- and multi-modal registration for MRI/X brain images, such that X refers
in our case to PET, CT and MRI atlas. We compare three widely brain image
registration tools (SPM, ITK-Snap, 3D Slicer) against an accurate hybrid registration method from the state-of-the-art.
The rest of this paper is organized as follows. Section 2 shows the studied
registration methods. Then, we present the clinical datasets and the evaluation
protocol in Sect. 3. We detail experimental results in Sect. 4. Finally, a conclusion
with some directions for future work are discussed in Sect. 5.
2 Registration Methods
ITK-Snap. Insight Segmentation and Registration Toolkit (ITK-Snap) is a popular tool for segmenting and registering medical images such as MRI, PET and
CT [7]. It is an open source software widely used by clinicians and non-computer
researchers. ITK-Snap allows manual and automatic medical image registration.
This software groups several methods of registration based on the intensity. For