Quantitative comparison of algorithms for inter-subject registration of 3D volumetric brain MRI scans

doi:10.1016/j.jneumeth.2004.07.014

Journal of Neuroscience Methods

Volume 142, Issue 1, 15 March 2005, Pages 67-76

https://doi.org/10.1016/j.jneumeth.2004.07.014 Get rights and content

Abstract

The objective of inter-subject registration of three-dimensional volumetric brain scans is to reduce the anatomical variability between the images scanned from different individuals. This is a necessary step in many different applications such as voxelwise group analysis of imaging data obtained from different individuals. In this paper, the ability of three different image registration algorithms in reducing inter-subject anatomical variability is quantitatively compared using a set of common high-resolution volumetric magnetic resonance imaging scans from 17 subjects. The algorithms are from the automatic image registration (AIR; version 5), the statistical parametric mapping (SPM99), and the automatic registration toolbox (ART) packages. The latter includes the implementation of a non-linear image registration algorithm, details of which are presented in this paper. The accuracy of registration is quantified in terms of two independent measures: (1) post-registration spatial dispersion of sets of homologous landmarks manually identified on images before or after registration; and (2) voxelwise image standard deviation maps computed within the set of images registered by each algorithm. Both measures showed that the ART algorithm is clearly superior to both AIR and SPM99 in reducing inter-subject anatomical variability. The spatial dispersion measure was found to be more sensitive when the landmarks were placed after image registration. The standard deviation measure was found sensitive to intensity normalization or the method of image interpolation.

Introduction

An important methodological consideration for analysis of human brain imaging data is inter-subject registration or spatial normalization of images acquired from different individuals. The aim of inter-subject registration is to reduce the anatomical variability in three-dimensional (3D) volumetric brain scans obtained from different subjects. For example, inter-subject registration allows voxelwise group analysis of functional magnetic resonance imaging (fMRI) data (Svensen et al., 2002, Zeffiro et al., 1997), and studies of brain white matter using diffusion tensor imaging (Ardekani et al., 2003, Jones et al., 2002). Another class of applications of inter-subject registration can be categorized as automatic ‘image understanding’, where higher order information (tissue type, locations of anatomical landmarks, structural boundaries, specific sulci/gyri, etc.) that are known about a template image are obtained about a subject or test image after the subject image is registered to the template by non-linear spatial transformation or deformation (Collins et al., 1995, Marroquin et al., 2002, Webb et al., 1999). An additional important application is the quantification of small changes in volume observed in anatomical structures over time (Holden et al., 2002, Rey et al., 2002) that can be used in diagnosis and evaluation of disease progression and treatment.

Spatial registration is specified in terms of a 3D transformation or displacement field $w$ : ℜ³ → ℜ³ which is applied to the subject image, I_s(r), to obtain the spatially normalized or warped image I_w(r) = I_s(r + $w$ (r)). The main objective of most inter-subject image registration algorithms is to find a displacement field $w$ such that the warped image I_w(r) is as ‘similar’ as possible to a template image, I_t(r). For a given pair of subject and template images I_s and I_t, algorithms mainly differ in their approach to modeling and estimation of $w$ . In the SPM99 software package (Friston et al., 1995), the displacement field $w$ is modeled by a finite orthogonal series with trigonometric basis functions. The algorithm computes a displacement field by estimating the coefficients of the series using an iterative linearized least-squares approach. In the automated image registration (AIR) software package (Woods et al., 1998), the displacement field is modeled as a polynomial, and the algorithm computes the displacement field by estimating the polynomial coefficients using non-linear least-squares optimization. Other methods of inter-subject registration model the displacement field as the displacement field in an elastic object (Bajcsy and Kovačič, 1989) or a viscous fluid (Christensen et al., 1997, Christensen and Johnson, 2001) reacting to internal forces proportional to ‘image mismatch’ between template and subject volumes. These methods are governed by partial differential equations that model the physical phenomenon. In several other methods, the displacement fields are estimated as non-parametric vector fields subject to regularity constraints (Collins et al., 1995, Kjems et al., 1999, Kosugi et al., 1993).

Although many methods have been proposed for spatial normalization, there are few studies comparing the performance of various algorithms on a common set of real MRI data. At least two important questions can be raised: (1) how do the algorithms compare in terms of their ability to reduce anatomic differences between subjects? And (2) how does image registration accuracy impact the final analysis results (e.g., activation maps in fMRI)? The present paper is concerned with the first question. The latter question has been previously addressed primarily in the context of positron emission tomography (PET) functional imaging studies (Crivello et al., 2002, Kjems et al., 1999, Senda et al., 1998). Overall, little difference has been found between the functional activation maps obtained by processing PET activation data using different inter-subject registration methods. Crivello et al. (2002) attributed this finding to the limited spatial resolution of PET and the inherent functional variability across subjects. However, this conclusion may not be automatically extended to higher resolution fMRI studies or group analysis of diffusion tensor imaging data. Recently, Ardekani et al. (2004) studied the impact of inter-subject registration on group analysis of fMRI data. They showed that increased accuracy of inter-subject registration in removing anatomic variability between subjects results in significant increases in the sensitivity of activation detection and the reproducibility of activation maps. Thus, at least in fMRI studies, it is important for researchers to understand the relative accuracy of inter-subject registration tools available to them.

The first objective of the present paper was to present details of our implementation of an inter-subject registration algorithm included in our automatic registration toolbox (ART). This algorithm is a non-parametric method similar to those proposed by Collins et al. (1995), Kjems et al. (1999), and Kosugi et al. (1993) with some new components. The second objective was to quantitatively compare the performance of three different inter-subject registration programs: (1) SPM99; (2) AIR; and (3) ART. The third aim was to evaluate different independent criteria for assessment of registration accuracy. One measure is the post-registration spatial dispersion of sets of homologous landmarks located manually on the images. Another criterion compares the sample standard deviation (S.D.) of voxel intensity maps computed from registered image sets. The precise mathematical definitions of these measures are presented in the following section.

Section snippets

Implementation of ART inter-subject registration

Without loss of generality, we assume that the template image I_t(r) and the subject image I_s(r) are of the same matrix and voxel dimensions. If not, the subject image is resized and interpolated so that its voxel and matrix dimensions match those of the template image. The objective of ART is to find a displacement vector $w$ (r) = (u_x(r), u_y(r), u_z(r)) at each voxel r. To achieve this, each voxel is visited in a raster scan fashion. Let Ω_r be a neighborhood around and including voxel r. The

Results and discussion

The 16 MP-RAGE volumes were registered to the template volume using ART, AIR, and SPM99, resulting in a total of 48 spatially normalized volumes (16 per method). The 16 registered volumes obtained from applying each algorithm were averaged. Selected slices from the resulting three average volumes are shown in Fig. 1. Qualitatively, it can be clearly seen that the average registered volume corresponding to ART (first column) has higher resolution than those corresponding to AIR (second column)

Conclusions

Details of implementation of a non-parametric method (ART) for estimating displacement fields for inter-subject registration of high-resolution volumetric MRI images were presented. The ability of ART in reducing anatomical variability between subjects was compared to the registration accuracy of two other popular registration programs: AIR and SPM99. Homologous landmark sets manually identified on the images before registration were significantly less dispersed in ART after registration, as

Acknowledgement

This research was supported by Biomedical Engineering Research Grant RG-00-0350 from the Whitaker Foundation to BAA. MJH gratefully acknowledges the support of NIH R01 MH64783 and a NARSAD Young Investigator award.

References (28)

R. Bajcsy et al.
Multiresolution elastic matching
Comput. Vis. Graph. Im. Proc.
(1989)
I.D. Grachev et al.
A method for assessing the accuracy of intersubject registration of the human brain using anatomic landmarks
Neuroimage
(1999)
D.K. Jones et al.
Spatial normalization and averaging of diffusion tensor MRI data sets
Neuroimage
(2002)
D. Rey et al.
Automatic detection and segmentation of evolving processes in 3D medical images: application to multiple sclerosis
Med. Image Anal.
(2002)
M. Senda et al.
Influence of ANOVA design and anatomical standardization on statistical mapping for PET activation
Neuroimage
(1998)
M. Svensen et al.
ICA of fMRI group study data
Neuroimage
(2002)
J. Webb et al.
Automatic detection of hippocampal atrophy on magnetic resonance images
Magn. Reson. Imaging.
(1999)
Ardekani BA, Bachman AH, Strother SC, Fujibayashi Y, Yonekura Y. Impact of inter-subject image registration on group...
B.A. Ardekani et al.
A fully automatic multimodality image registration algorithm
J. Comput. Assist. Tomogr.
(1995)
B.A. Ardekani et al.
Automatic detection of the mid-sagittal plane in 3D brain images
IEEE Trans. Med. Imaging
(1997)

B.A. Ardekani et al.

MRI study of white matter diffusion anisotropy in schizophrenia

Neuroreport

(2003)

J. Ashburner et al.

Voxel based morphometry—the methods

Hum. Brain Mapp.

(2000)

F.L. Bookstein

“Voxel-based morphometry” should not be used with imperfectly registered images

NeuroImage

(2001)

G.E. Christensen et al.

Consistent image registration

IEEE Trans. Med. Imag.

(2001)

Cited by (206)

It is the locus coeruleus! Or… is it?: a proposition for analyses and reporting standards for structural and functional magnetic resonance imaging of the noradrenergic locus coeruleus
2023, Neurobiology of Aging
The noradrenergic locus coeruleus (LC) is one of the protein pathology epicenters in neurodegenerative diseases. In contrast to PET (positron emission tomography), MRI (magnetic resonance imaging) offers the spatial resolution necessary to investigate the 3–4 mm wide and 1.5 cm long LC. However, standard data postprocessing is often too spatially imprecise to allow investigating the structure and function of the LC at the group level. Our analysis pipeline uses a combination of existing toolboxes (SPM12, ANTs, FSL, FreeSurfer), and is tailored towards achieving suitable spatial precision in the brainstem area. Its effectiveness is demonstrated using 2 datasets comprising both younger and older adults. We also suggest quality assessment procedures which allow to quantify the spatial precision obtained. Spatial deviations below 2.5 mm in the LC area are achieved, which is superior to current standard approaches. Relevant for ageing and clinical researchers interested in brainstem imaging, we provide a tool for more reliable analyses of structural and functional LC imaging data which can be also adapted for investigating other nuclei of the brainstem.
Depression core network-based individualized targeting for transcranial magnetic stimulation
2023, Brain Stimulation
Transcranial magnetic stimulation (TMS) of the dorsolateral prefrontal cortex (DLPFC) is an established treatment for major depressive disorder (MDD). Recent attempts to improve TMS efficacy by individually targeting DLPFC subregions that are functionally connected to the subgenual anterior cingulate cortex (sgACC) appear promising. However, sgACC covers only a small subset of core MDD-related areas. Further, fMRI connectivity of sgACC is poorly repeatable within subjects.
Based on an fMRI database analysis, we first constructed a novel core network model (CNM), capturing voxelwise emotion regulation- and MDD-related DLPFC connectivity. Then, in a sample of 15 healthy subjects and 29 MDD patients, we assessed (i) within-subject repeatability of the DLPFC connectivity patterns computed from time segments of varying lengths of individual-level fMRI data and (ii) association of MDD severity with the individual DLPFC connectivity strengths. We extracted group-level connectivity strengths in CNM from individual DLPFC coordinates stimulated with neuronavigated TMS in a separate sample of 25 MDD patients. These connectivity strengths were then correlated with individual TMS efficacy.
Compared with sgACC connectivity, CNM increased intraindividual repeatability 5-fold. DLPFC connectivity strength from CNM was associated with MDD severity and TMS efficacy. While the locations of CNM-based individual TMS targets remained constant within individuals, they varied considerably between individuals.
CNM increased repeatability of functional targeting to a clinically feasible level. The observed association of MDD severity and TMS efficacy with DLPFC connectivity supports the validity of the CNM. The interindividual differences in target locations motivate future individualized clinical trials leveraging the CNM.
Impact of deformable registration methods for prediction of recurrence free survival response to neoadjuvant chemotherapy in breast cancer: Results from the ISPY 1/ACRIN 6657 trial
2022, Translational Oncology
Image registration plays a vital role in spatially aligning multiple MRI scans for better longitudinal assessment of tumor morphological features. The objective was to evaluate the effect of registration accuracy of six established deformable registration methods(ANTs, DRAMMS, ART, NiftyReg, SSD-FFD, and NMI-FFD) on the predictive value of extracted radiomic features when modeling recurrence-free-survival(RFS) for women after neoadjuvant chemotherapy(NAC) for locally advanced breast cancer.
130 women had DCE-MRI scans available from the first two visits in the ISPY1/ACRIN-6657 cohort. We calculated the transformation field from each of the different deformable registration methods, and used it to compute voxel-wise parametric-response-maps(PRM) for established four kinetic features.104-radiomic features were computed from each PRM map to characterize intra-tumor heterogeneity. We evaluated performance for RFS using Cox-regression, C-statistic, and Kaplan-Meier(KM) plots.
A baseline model(F1:Age, Race, and Hormone-receptor-status) had a 0.54 C-statistic, and model F2(baseline + functional-tumor-volume at early treatment visit(FTV₂)) had 0.63. The F2+ANTs had the highest C-statistic(0.72) with the smallest landmark differences(5.40±4.40mm) as compared to other models. The KM curve for model F2 gave p=0.004 for separation between women above and below the median hazard compared to the model F1(p=0.31). A models augmented with radiomic features, also achieved significant KM curve separation(p<0.001) except the F2+ART model.
Incorporating image registration in quantifying changes in tumor heterogeneity during NAC can improve prediction of RFS. Radiomic features of PRM maps derived from warping the DCE-MRI kinetic maps using ANTs registration method further improved the early prediction of RFS as compared to other methods.
A novel MRI-based quantitative water content atlas of the human brain
2022, NeuroImage
The measurement of quantitative, tissue-specific MR properties, e.g., water content, longitudinal relaxation time (T₁) and effective transverse relaxation time (T₂^⁎), using quantitative MRI at a clinical field strength (1.5 T to 3T) is a well-explored topic. However, none of the commonly used standard brain atlases, such as MNI or JHU, provide quantitative information. Within the framework of quantitative MRI of the brain, this work reports on the development of the first quantitative brain atlas for tissue water content at 3T. A methodology to create this quantitative atlas of in vivo brain water content based on healthy volunteers is presented, and preliminary, practical examples of its potential applications are also shown.
Established methods for the fast and reliable measurement of the absolute water content were used to achieve high precision and accuracy. Water content and T₂^⁎ were mapped based on two different methods: an intermediate-TR, two-point method and a long-TR, single-scan method. Twenty healthy subjects (age 25.3 ± 2.5 years) were examined with these quantitative imaging protocols. The images were normalised to MNI stereotactic coordinates, and water content atlases of healthy volunteers were created for each method and compared. Regions-of-interest were generated with the help of a standard MNI template, and water content values averaged across the ROIs were compared to water content values from the literature.
Finally, in order to demonstrate the strength of quantitative MRI, water content maps from patients with pathological changes in the brain due to stroke, tumour (glioblastoma) and multiple sclerosis were voxel-wise compared to the healthy brain.
The water content atlases were largely independent of the method used to acquire the individual water maps. Global grey matter and white matter water content values between the methods agreed with each other to within 0.5 %. The feasibility of detecting abnormal water content in the brains of patients based on comparison to a healthy brain water content atlas was demonstrated.
In summary, the first quantitative water content brain atlas in vivo has been developed, and a voxel-wise assessment of pathology-related changes in the brain water content has been performed. These results suggest that qMRI, in combination with a water content atlas, allows for a quantitative interpretation of changes due to disease and could be used for disease monitoring.
Measuring variability of local brain volume using improved volume preserved warping
2022, Computerized Medical Imaging and Graphics
Measuring local brain volume is clinically important in neuroimaging studies. Voxel preserved warping (VPW) and Jacobian determinant are effective methods for studying local brain volume changes and variations (LBVCV) across multiple brains. However, these LBVCV methods typically depend on the local deformation without using the global deformation, while both deformations are needed in co-registering the brains under examination so that the brains can be compared on a common and fair basis. However, instead of employing a uniformed strategy, different co-registration methods have developed their own unique strategy in performing global and local transformation of the co-registration of the brains, and how the global and local transformations may combine to achieve the final goal of co-registration is not their concern, as long as the final registration may accomplish the co-registering job satisfactorily. The aforementioned inconsistency thus makes the LBVCV measurement that relies on the registration methods for studying local brain volumes totally unstable and actually unreliable. To address the uncertainty in measuring local brain volume variability caused by the un-uniqueness of performing global and local deformations during co-registration, the present study proposes new VPW approaches (VPW_α and VPW_β), which no longer require the separation of the global and local transformation components but employ only the general deformation concatenating both components, as long as the general registration may achieve the task of co-registering brain images. The new VPW methods are validated in theory and in practice, using both simulated and real-world imaging data, respectively, based on two registration methods popularly in use by the neuroimaging research community, i.e., the Automatic Registration Toolbox (ART) and Symmetric Image Normalization Method (SyN) registration methods. Experiments using simulated data demonstrated that the proposed new VPW methods may reliably measure local brain volume changes and variability. In contrast, traditional methods typically may result in LBVCV maps containing significantly inconsistent even false findings. In the experiments using real neuroimaging datasets from a schizophrenia study, the results based on the proposed new VPW methods were highly consistent, no matter which registration method was employed. Otherwise, the LBVCV results based on traditional approaches would show significant difference, depending on the individual registration method that the analysis employed. LBVCV assessments based on traditional methods appear to be unreliable. The proposed new VPW methods for measuring local volume changes is independent of registration methods, and therefore can serve as alternative approaches for assessing LBVCV reliably.
MRI brain templates of the male Yucatan minipig
2021, NeuroImage
The pig is growing in popularity as an experimental animal because its gyrencephalic brain is similar to humans. Currently, however, there is a lack of appropriate brain templates to support functional and structural neuroimaging pipelines. The primary contribution of this work is an average volume from an iterative, non-linear registration of 70 five- to seven-month-old male Yucatan minipigs. In addition, several aspects of this study are unique, including the comparison of linear and non-linear template generation, the characterization of a large and homogeneous cohort, an analysis of effective resolution after averaging, and the evaluation of potential in-template bias as well as a comparison with a template from another minipig species using a “left-out” validation set. We found that within our highly homogeneous cohort, non-linear registration produced better templates, but only marginally so. Although our T1-weighted data were resolution limited, we preserved effective resolution across the multi-subject average, produced templates that have high gray-white matter contrast and demonstrate superior registration accuracy compared to an alternative minipig template.

View all citing articles on Scopus

View full text

Quantitative comparison of algorithms for inter-subject registration of 3D volumetric brain MRI scans

Abstract

Introduction

Section snippets

Implementation of ART inter-subject registration

Results and discussion

Conclusions

Acknowledgement

Comput. Vis. Graph. Im. Proc.

Neuroimage

Neuroimage

Med. Image Anal.

Neuroimage

Neuroimage

Magn. Reson. Imaging.

A fully automatic multimodality image registration algorithm

J. Comput. Assist. Tomogr.

Automatic detection of the mid-sagittal plane in 3D brain images

IEEE Trans. Med. Imaging

MRI study of white matter diffusion anisotropy in schizophrenia

Neuroreport

Voxel based morphometry—the methods

Hum. Brain Mapp.

“Voxel-based morphometry” should not be used with imperfectly registered images

NeuroImage

Consistent image registration

IEEE Trans. Med. Imag.