To study the shape changes in precentral gyrus (PRG) and postcentral gyrus (PCG) during healthy brain development in childhood.
To study the shape changes in precentral gyrus (PRG) and postcentral gyrus (PCG) during healthy brain development in childhood.
Magnetic resonance (MR) images from 20 healthy children scanned twice at 6.3 ± 0.8 and 8.1 ± 0.9 years old were analyzed in this study. The analysis steps included: 1) The PRG and PCG were manually segmented from the MR images of each subject; 2) 3D mesh representing the PRG and PCG were built from the manually segmented images; 3) a series of shape description features were extracted and statistically analyzed by a permutation test method.
This study showed the following statistically significant findings (P < 0.05): left PCG and PRG are located more posteriorly and superiorly than their right compartments; the hemispheric asymmetry of the PRG in the inferior–superior direction decreased after 2 years; the superior part of the PCG and PRG shifted and rotated to the posterior side of the brain.
During normal pediatric brain development hemispheric asymmetry and shape of PCG and PRG are changed. These findings, together with previous reports in the literature, illustrate a region-specific brain structure maturation pattern in children and may be related to changing neurocognitive functions. J. Magn. Reson. Imaging 2011;33:62–70. © 2010 Wiley-Liss, Inc.
PEDIATRIC BRAIN MATURATION is a dynamic, structure-dependent, nonlinear biological process (1, 2), and study of the normal brain developmental process is essential to understand neurodevelopmental disorders. Pediatric brain developmental morphologic changes have been studied using global and regional analysis methods. Although global analysis methods provide valuable overall information on the development of pediatric brains, they may disguise the complexity of brain structure changes. Regional studies can investigate localized information of a region corresponding to specific functions. Brain developmental morphologic changes can be assessed by either cross-sectional or longitudinal study methods. In cross-sectional studies, magnetic resonance imaging (MRI) data from different subjects are compared with one another. Longitudinal studies investigate MR data from same subject cohort scanned twice or more with specific time gaps. Longitudinal methods are more sensitive than cross-sectional methods to individual brain developmental patterns due to their exclusion of the influence of large interindividual variations.
Global cross-sectional studies of volumes of the healthy maturation of neocortex gray matter, subcortical gray matter structures (eg, hippocampus and thalamus), and white matter are well documented in the neuroimaging literature (3–6). Many valuable volumetric findings about pediatric brain development have been obtained: total cerebral volume does not change significantly after 5 years of age (7); pediatric brain development is also gender- and hemisphere-dependent; for example, the putamen and globus pallidus are larger in males than in females even after adjusting for cerebral size, the volumes of the caudate and putamen decrease with age only in males (4), and the left putamen is significantly larger than the right (4). Giedd et al (8) performed a global longitudinal study and showed that gray matter volume of the frontal lobe reaches its maximum at about 11 to 12 years old, and then the volume starts to decrease.
Regional cross-sectional (9–11) and longitudinal (12–14) studies have been performed to investigate localized information about pediatric brain development. Cortical thickness (13) and gray matter density (12) studies showed an overall developmental or maturation sequence of the neocortex during childhood through early adulthood: the primary sensory and motor areas together with the frontal and occipital poles develop first, and then spread caudally over the parietal lobe and rostrally over the frontal lobe; the temporal lobe matures last. A voxel-based morphometry study (10) showed localized differences in dorsal frontal and parietal regions between childhood and adolescence. These studies (9–14) showed that brain development is dynamic and region-dependent and can be cubic, quadratic, or linear over time, depending on cerebral locations, all of which align well with the cortical types (13) and cognitive function development of the neocortex (12).
Hemispheric asymmetries have been well documented in some brain regions by postmortem or in vivo MR studies. For example, the postcentral sulcus on the left hemisphere is located more posteriorly than that on the right hemisphere (15), and the Sylvian fissure extends more posteriorly on the left than on the right (16). Binder et al (17) proposed that the left frontal lobe and peri-rolandic region shift toward the posterior brain side and the right posterior parietal lobe shifts toward the anterior side in humans, which explains the shape asymmetry of the posterior Sylvian fissure. However, little is known about the emergence of these gyral or sulcal hemispheric-asymmetries (11). Sowell et al (11) studied sulcal asymmetry differences between child and adult subjects and found very interesting and valuable results that showed the asymmetries in the perisylvian area were larger in adulthood than in childhood, which may show an environmental effect (eg, language) on brain development. However, a curve representation of sulci may lose spatial and regional change information, and little is known about whether this asymmetric developmental process is monotonous or not.
In this study we used a gyral region representation and a series of features to address the previously mentioned gyrus developmental issues in childhood; that is, to study location shift, hemispheric asymmetries, and shape and asymmetric developmental changes in the postcentral gyrus (PCG) and precentral gyrus (PRG) during childhood by a longitudinal method. The PCG and PRG are the primary motor and sensory areas that develop first (12, 13) and are well defined by surrounding sulci, resulting in a fairly easy and consistent segmentation, which is critical in a brain morphology developmental study.
Records of 20 healthy children (13 female, 7 male; 19 right-handed, 1 not right-handed) were arbitrarily retrieved from the pediatric brain database (18). Each record included 2 MR scans acquired at ages (mean ± SD) 6.3 ± 0.8 and 8.1 ± 0.9 years old. The use of these data for this retrospective study was approved by the Institutional Review Board at our institution. The 3D T1 MR data in the pediatric brain database were acquired primarily as 1-mm isotropic voxels using a spoiled gradient recalled echo with TR 22–25 (msec) and TE 10–11 (msec) by a 1.5 T GE or Siemens MR system (18). Demographic data on these subjects is summarized in Table 1. The MR scans formed two groups of image datasets designated group one and group two, which were used to study the age effect on PCG and PRG during childhood.
|ID||Scan 1||Scan 2||Gender||Handedness|
|6.272 ± 0.844||8.120 ± 0.859||13 F /7 M||19 R/1 NR|
All the retrieved T1-weighted MR head images were corrected for 3D intensity nonuniformity and registered to the standardized Talairach space using a 9-parameter linear transformation (18). Nonbrain tissues (eg, skull and fat) were removed from the head images (19).
The PCG and PRG are bounded by interhemispheric fissure medially, precentral sulcus anteriorly and rostrally, postcentral sulcus posteriorly and caudally, and lateral sulcus inferiorly. They are separated by central sulcus. An experienced rater manually segmented the PCG and PRG in the sagittal slice of the MR image data with reference to the 3D cortical surface mesh representing the gray matter and cerebrospinal fluid interface generated by FreeSurfer (20). When interruptions were encountered, we manually connected them in a shortest distance fashion.
We used the contour-based surface parameterization method (21) to parameterize the surface of a gyrus and generate its mesh representation. Given a stack of outline contours of a gyrus in the sagittal plane C1,…,Cm the 3D surface parameterization is to find the mapping function:
where u,v are integers and 0 ≤ u ≤ nu, 0 ≤ v ≤ nv, forming a rectangular grid. nu and nv are the numbers of points in the u and v directions, respectively. Thus, the 3D surface is represented as a set of position vectors in the standardized Talairach space. u and v are parametric coordinates of the surface . In fact, the manually segmented contour Ck in the image space itself is made up of a set of nk digitized points, where nk may vary for different contours. In the 3D surface parameterization process, we first generate nu + 1 points dividing the circumference of contour Ck equally by linear interpolation. In this step, we calculate pk0 as the point on the contour Ck nearest to the coordinate origin. After processing all the m contours C1,…,Cm, we obtain a stack of digitized contours parameterized only in the sagittal plane, that is, in the u direction. We then parameterize the surface using a similar interpolation process by equally dividing the length of curves made up of p1j,…,pmj, for j = 0,…,nu, which parameterized the surface in v direction.
After the 3D surface of a gyrus is parameterized into a regular rectangular grid of size nu×nv(nu = 200, nv = 500 in our experiment), surface points indexed by parameter coordinates (u,v) are analogous anatomic points on different subjects.
The most inferior and posterior points of the surface of a gyrus are denoted by PI and Pp, respectively, and the line passing through PI and parallel to the y′ axis of the Talairach space is defined as the y axis, whose direction is from the posterior to the anterior side of the brain. The interpolating point PO of the y axis and the plane passing through Pp, parallel to the x′z′ plane of the Talairach space is defined as the origin of the new coordinate system. Then, the z axis is defined as parallel to the z′ axis of the Talairach space, whose direction is from the inferior to the superior side of the brain. Finally, the x axis is perpendicular to both the y and z axes and directs from right to left. For a point P on the PCG or PRG contour, the coordinates yp and zp and the angle ap are used as shape descriptors (Fig. 1). Features yp and zp represent the shape shift changes in posterior–anterior and inferior–superior directions, respectively. Angle ap expresses the rotational shape change. For consistency of the reference points, we use the origin point PO of the first scan to compute shape features of the second scan of the same subject.
For a point P on the surface of a gyrus, we use the Talairach space coordinates y′ and z′ to describe its location in posterior–anterior and inferior–superior directions, respectively. A group of asymmetry descriptors are defined for a pair of analogous points PL and PR on the left and right hemispheres:
ALR(y′),ARL(y′),ALR(z′), and ARL(z′) describe the location differences of a pair of analogous points on the left- and right-hemispheric compartments of a gyrus in the posterior–anterior and inferior–superior directions. ALR(·) and ARL(·) are defined as left and right asymmetry effects, respectively. We did not consider asymmetry in the left–right direction because medial plane detection accuracy may affect the results.
For developmental changes in gyral shape and hemispheric asymmetry, a dependent t statistic for each feature was calculated at each triangle of the surface mesh representing the surface of a gyrus to generate a t statistic map by:
where is the average feature difference between the two groups of subjects, sD is the standard deviation of the differences, and n is the number of subjects. For hemispheric asymmetric features, in (3) is the average of ALR(y′),ARL(y′),ALR(z′), or ARL(z′). Each mesh triangle's features are computed by either its centroid's Talairach space coordinate (y′,z′) or coordinate (y,z) defined in the previous section.
When investigating developmental changes in shape or hemispheric asymmetry, the feature difference fD can be calculated either by features of group one (MR image dataset of the first scans) minus those of group two (MR image dataset of the second scans) or by group two features minus group one features, yielding a negative effect or positive effect, respectively.
A permutation test is a statistical significance test in which a reference distribution is obtained by computing all possible test statistic values by relabeling the observed data points (22). Permutation tests have been used extensively to determine whether a finding is statistically significant when multiple comparisons are considered (11, 14, 23). We used a cluster size analysis permutation test method (22) to evaluate group differences. Group memberships were rearranged 10,000 times to generate 10,000 t statistic maps. These t statistical maps were then thresholded using a threshold of probability 0.05 (two-tailed with 19 degrees of freedom). The maximum cluster size in each of these thresholded t statistic maps was recorded, and those together formed an accurate experimental distribution of the maximum cluster size denoted by h(bc). A p value was attached to each cluster in the actual thresholded t statistical map by:
where bc is the cluster size of cluster c.
The mesh generation, feature computation, and statistical analysis methods were implemented using C++ language. Experiments were performed on a Linux workstation with two Quad-Core AMD Opteron Processors (1.15 GHz) and 8GB memory. Pre- or postcentral gyri mesh generation and feature calculation takes about 2 minutes, while statistical analysis using permutation test spends about 19 minutes.
To evaluate the rater's manual segmentation reproducibility, the rater performed manual segmentation of two MR image datasets two times, a week apart. A similarity index (also known as a Dice coefficient) measure of reproducibility, which is defined as the overlapping size of two segmentations of an object divided by the average size of the two segmentations, was then computed between these repeat manual segmentations. The achieved similarity index value of 0.93 and greater by the rater demonstrated that the reproducibility of the rater was high enough to detect subtle brain developmental change patterns.
We performed hemispheric asymmetry analysis on the two groups of MR data acquired about 2 years apart. Due to the results of the two groups of MR data being similar, we only show the analysis results of the MR data acquired at the first timepoint in this article. Statistically significant areas of a right asymmetry effect in the PCG in the posterior–anterior direction were found, with a P value of 0.0015. Statistically significant areas of a left asymmetry effect in the inferior–superior direction were also found, with a P value of 0.0032. No statistically significant left asymmetry effects in the posterior–anterior direction or right asymmetry effects in the inferior–superior direction were found. The detected statistically significant areas were 2745.88 and 1506.02 mm2 (57.88% and 31.75% of the total surface area) for the posterior–anterior and inferior–superior directions, respectively. Similar hemispheric asymmetry findings of PRG were also observed, with P values of 0.0008 and 0.0007, significant areas of 2499.17 and 1485.63 mm2 (56.14% and 33.37% of the total surface area), respectively. The results are illustrated in Figs. 2 and 3, respectively. The right asymmetry effect is right hemispheric features minus corresponding left hemispheric ones and the left asymmetry effect is the opposite, so the results show that both the left PCG and PRG are located more posteriorly and superiorly than their right counterparts. In the posterior–anterior direction the hemispheric asymmetries (mean ± SD) were 4.09 ± 1.13 mm and 4.21 ± 0.75 mm for the PCG and PRG, respectively. In the inferior–superior direction, they were 2.74 ± 0.76 mm and 4.23 ± 0.95 mm for the PCG and PRG, respectively.
Statistically significant areas of a negative effect on left asymmetric changes in the PRG in the inferior–superior direction were found, with a P value of 0.0174. No statistically significant results were found for the PCG. The detected statistically significant area was 207.844 mm2 (4.66% of the total surface area). The result is illustrated in Fig. 4. For asymmetric developmental changes, a negative effect difference was computed by features of the first group minus those of the second group, so the result showed that the left asymmetry of the PRG in the inferior–superior direction decreased over 2 years in the range of 0.563 to 3.509 mm in the statistically significant area; that is, the left PRG was located less superiorly than the right compartment after 2 years.
Statistically significant areas of a negative effect on features yp and aP were found in both the PCG and PRG, with all P values less than 0.05. The detected significant areas are summarized in Table 2. The similarity index between the detected statistically significant areas of features yp and aP was 0.76 ± 0.04 (mean ± SD), which illustrates that the two detected significant parts were highly consistent. This may come from the fact that aP = tg−1(yp/zp). Thus, we illustrated only the detected statistically significant areas of feature yp for the PCG and PRG in Figs. 5 and 6, respectively.
|Gyri||P-values||Cluster size (mm2)||PTSA|
Because negative effect feature differences were computed by features of the first group minus those of the second group, the experimental results showed that differences in the significant areas' features yp and aP became smaller statistically over 2 years, in the range of 0.3 to 1.9 mm and 0.26 to 2.00 degrees for the PCG and 0.4 to 2.5 mm and 0.3 to 1.9 degrees for the PRG. Considering the definition of the shape descriptors (Fig. 1), the experimental results showed that the superior part of the PCG and PRG of the pediatric brain shifts and rotates toward the posterior side of the brain.
Finally, we illustrated a per-subject plot of differences in the detected significant areas of feature yp in the left PCG between group one and group two (Fig. 7) to show that our results were a real group trend, not caused by outlier measurements. In Fig. 7 the upper-left subimage in blue is the detected statistically significant area in the central sulcus side; for all the other subimages we display them in red, if the individual subject developmental changes of feature yp in the detected significant area are consistent with the group analysis result. Since all these sub-images are very similar, they show that the detected significant area is a real group trend, not a result of outlier measurements. We chose feature yp because the change in feature yp was smaller than the hemispheric asymmetry measures in the detected statistically significant areas, which would make it more likely to be affected by outlier measurements.
In this longitudinal study, we found that the left PCG and PRG are located more posteriorly and superiorly than their right compartments, the hemispheric asymmetry of the PRG in the inferior–superior direction decreased after 2 years, and the superior part of the PCG and PRG shifted and rotated to the posterior side of the brain.
The hemispheric asymmetry findings are consistent with previous work (15, 24, 25), which is also suggested in Binder et al's report (17) to explain the role of the planum temporale in language processing.
Sowell et al (11) found that the hemispheric sulcal asymmetry in the perisylvian region in adults was larger than those in children. However, little is known regarding whether this asymmetry continuously grows from childhood to adulthood or whether it decreases in some part of the brain development process. Our longitudinal study showed that the hemispheric asymmetry of the superior part of the PRG decreased in the pediatric subjects we studied in the age range of 6 to 8 years old. Although we do not know whether the asymmetry in this area will be larger in adults than in children, our results showed that the hemispheric asymmetry could decrease during brain development. This hemispheric asymmetry developmental finding is consistent with the findings in Shaw et al's study (13), which pointed out that the left PRG reaches its peak cortical thickness about 1 year later (10.5 years old) than the right (9.6 years old). As we already know that the left PRG is located more superiorly than the right, an earlier increase in the thickness of the right PRG would decrease the left-asymmetry effect in the inferior–superior direction. Regarding PCG, we did not detect any hemispheric asymmetric developmental changes, which also is in agreement with Shaw et al's findings that both the left PCG and right PCG follow a cubic developmental trajectory and reach their peak cortical thickness at almost the same time (8.4 and 8.5 years old for the right and left PCG, respectively) (13). Future studies on gender differences and using older (and younger) populations are worthwhile and necessary to address this issue further.
Pediatric brain developmental studies using cortical thickness (13) and gray matter density (12) showed that neocortex development could be cubic, quadratic, or linear over time, depending on locations. The pediatric brain developmental process is also temporally in accordance with the brain function maturation sequence. This complicated region-dependent dynamic gray matter developmental pattern may cause the developmental shape changes we detected here.
The detected shape changes were on the submillimeter scale, although the resolution of MR images was 1 × 1 × 1 mm. This was because the vertex's position in the mesh representation of the surface of a gyrus was represented in decimal numbers (not only integers) as a result of interpolation.
We used a permutation test to correct multicomparison issues in this study. This test is a nonparametric statistical analysis method providing flexible and intuitive methodology to study many kinds of statistical data, such as functional neuroimaging experimental data (22). Compared with the statistical parametric mapping approach (26) derived from random theory, the permutation test method requires fewer assumptions and can be used when the distribution of the statistic is unknown and the statistical parametric mapping approach would be inappropriate. More important is that using the permutation test method avoids the complicated correction issues of the statistical parametric mapping approach for data located on surfaces (23). Rettmann et al (27) considered multicomparisons for a group of symbolic features, by controlling the false discovery rate (FDR). Benjamini and Hochberg (28) pointed out that the FDR is always smaller than the family-wise error rate (FWER) that is controlled by the permutation test algorithms we used. This means that the FDR is implicitly controlled by controlling the FWER; however, controlling the FDR does not control the FWER in any way. In the brain structure mapping study literature (11, 14, 23), controlling the FWER by permutation test is preferred, so we also used a permutation test method.
Lenroot et al (29) found by volumetric study that brain developmental trajectories during childhood and adolescence differed between males and females; specifically, females usually achieve gray matter peak volume 1 to 2 years earlier than males. In our study, there were no statistically significant age differences between male and female subjects at either the first or second scans, with P values of 0.9 and 0.7, respectively. Considering the complicated, region-dependent, dynamic brain structure developmental process, a study of the gender dimorphism of shape and hemispheric asymmetry development would be interesting, and we plan to perform such a study in the near future using our current analysis method.
This study has two limitations: 1) the brain structures are manually segmented and 2) the data from 20 subjects were used due to the labor-intensive manual segmentation. Automatic gyrus segmentation and recognition methods are a must if hundreds of subjects are to be involved in this kind of study. Although Yang and Kruggel (30) developed an automatic brain sulci segmentation method, it is not straightforward to obtain a brain gyrus segmentation from the sulci segmentation for two reasons: 1) a sulcus may have one or several pieces (eg, the central sulcus may have one piece, two pieces, or even more (31), so manual interactions (cutting or linking) are required to segment the neighboring gyri; 2) the sulcal and gyral regions are overlapped in definition. Although there are other previous studies reporting automatic gyral segmentations (32), to the best of our knowledge no publicly accessible software tools are available. Considering the efforts of implementation of an algorithm into an actual computational tool and the potentially limited applicability, we still manually segmented the PRG and PCG in this preliminary study to evaluate the possible gyral shape changes during brain maturation in childhood. In the meantime, an ongoing project is to develop an automatic segmentation of gyri based cortical structures.
In conclusion, statistically significant age effects on hemispheric asymmetries and shapes of the PCG and PRG were found by analyzing head MR images. These findings, together with previous reports in the literature (8, 10, 12, 13), illustrated a region-specific brain structure maturation pattern in children and may be related to changing neurocognitive functions.