In the behavioral tasks, digit and letter memory spans were found significantly higher in the AMC group when compared with the controls (Fig. (Fig.2a)2a) and significantly correlated across all subjects (Fig. (Fig.2b).2b). Using TBSS method, we demonstrate for the first time the changes in brain WM integrity induced by long‐term training of AMC in Chinese abacus children. Increased WM FA in the abacus group was observed in the corpus callosum, the left occipitotemporal junction and the right premotor projection (see Fig. Fig.4),4), while no regions with decreased FA in the abacus group were detected. Furthermore, lower radial diffusivity was found in the abacus group (see Fig. Fig.5),5), but no significant difference was found in axial diffusivity between groups. Besides, working memory spans were found associated with brain white matter integrity (Figs. (Figs.33 and and66).
Whole‐Brain Average Skeleton FA
FA is thought to be related to WM integrity, and the decline in FA is often viewed as an index of decreasing WM health. As shown in Figure Figure3,3, the whole‐brain average skeleton FA is significantly higher in the abacus group than in the controls. In this study, the participants were about 10‐years old and this age bracket is crucial for brain development. During the training period of over 3 years, the children with AMC training gradually develop to adapt to the operation with either a true or an imagined abacus. The training‐induced gradual brain structural modulation could eventually accumulate to make significant differences in WM integrity between abacus children and their control peers. As the superior digit working memory in the AMC group likely results from the AMC training, the positive correlation between digit working memory and global brain white matter integrity could be an evidence of training‐induced brain modulation.
Voxelwise Statistics of FA, Axial, and Radial Diffusivity
Voxelwise statistics revealed that FA values increased in the abacus group in left occipitotemporal junction, right premotor projection and corpus callosum (see Fig. Fig.4),4), and radial diffusivity decreased in internal capsule, posterior thematic radiation, right occipitotemporal junction and the regions with increased FA mentioned earlier (see Fig. Fig.5).5). Axial diffusivity, however, showed no difference between groups. That is, the increase of FA in the abacus group was mainly driven by the decrease of radial diffusivity. As radial diffusivity was thought to reflect the myelination degree while radial diffusivity axon maturation [Song et al., 2002, 2003], it makes sense that AMC training likely results in the significant acceleration of myelination process but exerts little influence on axonal maturation.
The regional differences in DIT measurements between the AMC group and the controls are discussed below.
The occipitotemporal junction contains fibers from many tracts, including inferior longitudinal fasciculus (ILF), inferior fronto‐occipital fasciculus (IFOF), and optic radiation (OR). As a major associative connection between occipital and temporal lobes, ILF is suggested to mediate fast transfer of visual signal to anterior temporal regions [Catani et al., 2003], as well as to be related to visual recent memory [Tusa and Ungerleider, 1985]. IFOF, running from the frontal lobe to the occipital and temporal lobes ipsilaterally, is suggested to be crucial for language semantic processing [Mandonnet et al., 2007]. As known, OR is the pathway carrying information from the thalamic lateral geniculate nucleus to the visual cortex. Functions of these fiber tracts mentioned earlier imply that the occipitotemporal junction is likely to play an important role in integrating visual and semantic information. That is to say, the acceleration of the myelination in this region may reflect the integration of abacus, lexical and phonetic representations of numbers in the AMC tasks.
Right premotor projection
As we know, abacus players operate abacus with both hands simultaneously in the initial stage of the AMC training. Bimanualness in operation with abacus could be one of the factors that affect myelination of the premotor projection. Besides, premotor cortex is important in dynamic visuospatial imagery [Lamm et al., 2001] and plays an important role in the translation from a visual or auditory cue to an associated movement [Passingham, 1989]. Furthermore, premotor cortex was reported to be critical in updating verbal information and spatial information [Tanaka et al., 2005]. Thereby, the enhancement of WM integrity in the right premotor projection could mainly result from two factors: (1) the frequent use of left fingers for abacus operation and (2) the greater involvement of neuro‐resources for spatial information processing in the right hemisphere in the AMC tasks.
FA enhancement in the abacus group was observed in splenium and anterior midbody at the midsagittal corpus callosum (Fig. (Fig.4c).4c). With diffusion tensor imaging technique, Hofer and Frahm [ 2006] demonstrated that fibers in splenium project to parietal, occipital and temporal regions while fibers in anterior midbody project to premotor and supplementary motor areas. Additionally, a neuroimaging research [Formisano et al., 2002] indicated that there is functional segmentation between left and right parietal cortices. They suggested that the left parietal cortex was likely involved in the generation of spatial mental images while the right parietal cortex engaged in the manipulation and comparison of such images. Bilateral premotor areas and posterior parietal regions were found engaged in the AMC tasks [Chen et al., 2006; Hanakawa et al., 2003]. According to the functional segmentation viewpoint, imagined abacus may be generated in the left hemisphere and manipulated in the right half. Long‐term AMC training, relying on visuospatial processing, is likely to enhance the functional interaction between the left and right hemispheres, hence enhanced fiber integrity in the corpus callosum.
Areas observed with radial diffusivity difference
As shown in Figure Figure4,4, corpus callosum, right premotor projection and left occipitotemporal junction demonstrated increased FA and decreased radial diffusivity in AMC group. Moreover, decreased radial diffusivity was also observed in the right hemisphere regions including posterior limb internal capsule, posterior thalamic radiation, corona radiata, superior longitudinal fasciculus, and inferior longitudinal fasciculus (see Fig. Fig.55).
Posterior thalamic radiation and occipitotemporal junction are thought to be components of visual spatial attention network [Tuch et al., 2005]. And, temporal part of superior longitudinal fasciculus and inferior longitudinal fasciculus are suggested to mediate fast transfer of visual signal to anterior temporal regions [Catani et al., 2003] and to play an important role in visual recent memory [Tusa and Ungerleider, 1985]. The employment of visuospatial strategy in AMC is likely to contribute to the acceleration of myelination in these visuospatial related regions mentioned earlier. The internal capsule is a major route connecting cerebral cortex to brainstem and spinal cord while corona radiata, connecting with the internal capsule, contains both descending and ascending axons that carry nearly all of the neural traffic from and to the cerebral cortex. The frequent use of left fingers in AMC might contribute much to the enhanced myelination in these two fiber tracts.
In short, more regions in right than left hemisphere were found with decreased radial diffusivity in the AMC group (see Fig. Fig.5).5). This finding supports the hypothesis that spatial information is processed predominately in right hemisphere [Formisano et al., 2002; Kohler et al., 1998].
A further question remains that why radial diffusivity seems more sensitive than FA to the changes of neurostructures in our study. We examine this question from a mathematic view. The first, second, and third eigenvalues (λ1, λ2, and λ3) estimated from DTI data are related to the major, intermediate, and minor axes of a diffusion ellipsoid. For isotropic diffusion (λ1 = λ2 = λ3), FA is zero, and in the case where there is a strongly preferred direction of diffusion [λ1 (λ2 + λ3)/2], FA approaches to one. Without any difference in axial diffusivity (λ1) between groups, as found in our study, FA values depend heavily on radial diffusivity [(λ2 + λ3)/2]. In such circumstance, radial diffusivity is more sensitive than FA to the changes of neuro‐structures, for the variances in axial diffusivity weaken the sensitivity of FA.
Relations Between Behavioral Performances and DTI Measurements
Working memory refers to the mechanism or system underlying the maintenance and processing of task‐relevant information during the performance of a cognitive task [Baddeley et al., 1974; Daneman and Carpenter, 1980]. Using an event‐related functional MRI study, Tanaka et al. [ 2002] found that superior short‐term memory for digit was associated with greater activity in cortical areas related to visuospatial working memory. Additionally, previous behavioral studies indicated that abacus training is helpful to improve short‐memory capacity in children [Bhaskaran et al., 2006; Lee et al., 2007]. In this study, digit working memory spans were found larger in the AMC group (see Fig. Fig.2),2), which confirms the effects of AMC training. Besides, digit working memory spans were positively correlated with the global average of skeleton FA (Fig. (Fig.3b).3b). Further analysis showed that, within regions showing different fiber integrity between the two groups, digit working memory spans were positively correlated with the regional average FA, negatively correlated with the regional average radial diffusivity and not correlated with regional average axial diffusivity (see Fig. Fig.6).6). These results indicate that improvement in working memory capacity by AMC training could be related with enhancement in white matter tracts integrity.
Intelligence, Training, Brain Developing, and Diseases
In the intelligence tests with WISC‐RC, the six verbal tests include information, digit span, vocabulary, arithmetic, comprehension and similarities, whereas the six performance tests include picture completion, picture arrangement, block design, object assembly, coding and mazes. In our study, scale scores were found significantly higher in the AMC group than in the control group in such three subtests as digit span, information, and coding, which confirms the influence of AMC training on digit memory [Bhaskaran et al., 2006; Tanaka et al., 2002] and further suggests that AMC training improves the ability in digit‐symbol substitution (as indicated in coding subtest). Interestingly, digit memory spans, higher in the AMC group than in the controls, were found positively correlated with WM integrity. Results from other studies also showed positive correlations between FA and behavioral measures such as reading ability [Beaulieu et al., 2005; Klingberg et al., 2000; Niogi and McCandliss, 2005], performance in a speeded lexical decision task [Gold et al., 2007], and musical sensorimotor practice [Bengtsson et al., 2005]. The positive correlation between digit memory spans and individual FA values in current study provides new evidence for the argument that higher FA might be behaviorally beneficial.
From a view of neurodevelopment, brain maturation is a complex process that continues well beyond infancy [Gao et al., 2009], and adolescence is thought to be a key period of brain rewiring. During the process of brain maturation, brain functions and underlying neurostructures could be modulated by training. Properties and intensity of the training are important in domain‐specific brain modulations, while the onset of the training is also a crucial factor. It was reported that, compared with the controls, enlarged anterior callosa [Schlaug et al., 1995] and increased mean diffusivity in corticospinal tracts [Imfeld et al., 2009] were only found in the early onset subgroup of musicians which had started musical training before the age of 7. Participants in this study all started the AMC training before the age of 7, which may partly account for the enhancement of white matter integrity.
Besides, in normal populations, FA values development followed an inverted U‐shaped curve, while the radial diffusivities followed a U‐shaped curve [Hasan et al., 2009]. Our results indicate that variance of whiter mater health indices (such as FA and radial diffusivity) might be experience‐related. Especially, long‐time intensive training may modulate the development of brain WM, but the underlying mechanism remains unknown. Increased and decreased FA resulting from training could be found in separate brain regions [Schmithorst and Wilke, 2002]. Interestingly, both increased FA [Bengtsson et al., 2005] and decreased FA [Imfeld et al., 2009; Schmithorst and Wilke, 2002] in the internal capsule have been reported to result from music training. In this study, we detected only increased FA regions in the AMC group. We speculate that the fast development‐induced increase in FA may suppress the training‐induced decreasing trends in FA.
Contrast to training‐induced FA increase, reduced FA was reported in a broad spectrum of diseases [Fling et al., 2008; Keller et al., 2003; Nave et al., 2008; Price et al., 2007; Sowell et al., 2008]. The corpus callosum receives a great many interests, for FA reduction in corpus callosum is related to many diseases such as schizophrenia [Keller et al., 2003], Multiple Sclerosis [Fling et al., 2008], fetal alcohol spectrum disorders [Sowell et al., 2008], spinocerebellar ataxia type 1 and 2 [Nave et al., 2008], first‐episode psychosis [Price et al., 2007], and so on. Despite the relations between training‐induced plastic modulations and DTI measurements still remain poorly understood and a matter of controversy [Alexander et al., 2007; Ashtari et al., 2007; Beaulieu, 2002], whether the training‐induced enhancement of WM integrity can reduce the risk of demyelination diseases as mentioned earlier is an interesting question.
In WISC‐RC intelligence tests, scaled scores of the digit span and the coding subtests were significantly higher in the AMC group, we speculated that the AMC training could improve children's ability in number‐related operations, including digit‐symbol mapping (as indicated in the coding subtest). Furthermore, in forward digit/letter memory span tests, the positive correlation between digit memory spans and letter memory spans implies that the AMC training not only improves the working memory capacity for digits but also for letters. The strength of these arguments, however, was weakened by the absence of pretests administered before the AMC training in this study. As for brain white matter integrity, the variation in FA relates to multiple factors including age, gender, handedness, and cognitive abilities [Schmithorst et al., 2008