1 Eur J Nucl Med Mol Imaging (2009) 36: DOI /s ORIGINAL ARTICLE Effects of long-term practice and task complexity on brain activities when performing abacus-based mental calculations: a PET study Tung-Hsin Wu & Chia-Lin Chen & Yung-Hui Huang & Ren-Shyan Liu & Jen-Chuen Hsieh & Jason J. S. Lee Received: 7 December 2007 / Accepted: 23 August 2008 / Published online: 5 November 2008 # Springer-Verlag 2008 Abstract Purpose The aim of this study was to examine the neural bases for the exceptional mental calculation ability possessed by Chinese abacus experts through PET imaging. Methods We compared the different regional cerebral blood flow (rcbf) patterns using 15 O-water PET in 10 abacus T.-H. Wu : C.-L. Chen Department of Medical Imaging and Radiological Sciences, Chung Shan Medical University, Taichung, Taiwan Y.-H. Huang Department of Medical Imaging and Radiological Sciences, I-Shou University, Kaohsiung County, Taiwan R.-S. Liu Department of Nuclear Medicine, Faculty of Medicine, National Yang-Ming University, Taipei, Taiwan R.-S. Liu Department of Nuclear Medicine, Taipei Veterans General Hospital, Taipei, Taiwan J.-C. Hsieh Brain Research Center and Institute of Brain Science, National Yang-Ming University, Taipei, Taiwan J.-C. Hsieh Department of Medical Research and Education, Taipei Veterans General Hospital, Taipei, Taiwan T.-H. Wu : J. J. S. Lee (*) Department of Biomedical Imaging and Radiological Sciences, National Yang-Ming University, 155 Li-Nong St., Sec. 2, Taipei, Taiwan112 experts and 12 non-experts while they were performing each of the following three tasks: covert reading, simple addition, and complex contiguous addition. All data collected were analyzed using SPM2 and MNI templates. Results For non-experts during the tasks of simple addition, the observed activation of brain regions were associated with coordination of language (inferior frontal network) and visuospatial processing (left parietal/frontal network). Similar activation patterns but with a larger visuospatial processing involvement were observed during complex contiguous addition tasks, suggesting the recruitment of more visuospatial memory for solving the complex problems. For abacus experts, however, the brain activation patterns showed slight differences when they were performing simple and complex addition tasks, both of which involve visuospatial processing (bilateral parietal/ frontal network). These findings supported the notion that the experts were completing all the calculation process on a virtual mental abacus and relying on this same computational strategy in both simple and complex tasks, which required almost no increasing brain workload for solving the latter. Conclusion In conclusion, after intensive training and practice, the neural pathways in an abacus expert have been connected more effectively for performing the number encoding and retrieval that are required in abacus tasks, resulting in exceptional mental computational ability. Keywords Mental calculation. Abacus. Brain activities. PET Introduction It is well known that practice and experience can result in substantial changes in the organization of the adult cerebral
2 Eur J Nucl Med Mol Imaging (2009) 36: cortex at multiple levels (from the molecular or synaptic level, to cortical maps and large-scale neural networks) of the central nervous system [1 3]. Modern neuroimaging methods such as positron emission tomography (PET) and functional magnetic resonance imaging (fmri) are excellent tools to study these changes, enabling examination of how practice in certain task affects the brain. Performing fast and accurate mental calculations which requires significant practice is a good example of high-level cognitive skill that involves the coordination of various basic and complex cognitive processes [4 6]. In our present study, we used PET imaging to investigate the neurophysiological mechanisms underlying superior mental computational ability of well-trained abacus experts. From the cognitive point of view, the mental solution of arithmetic problems requires integration of multiple cognitive functions, including recognizing and manipulating numbers in a working memory, temporarily storing and retrieving the intermediate results, and meanwhile applying basic arithmetical rules for sequential control of various steps. For solving more complex problems efficiently, one requires not only various working memories, but also dedicated resolution algorithms and long-term practice, and all of which are critical factors influencing calculation performance. This could suggest that brain activity during the performance of mental calculations will also depend on the complexity of arithmetic problems, the resolution algorithms applied, and the specific test groups (trained vs. not trained) recruited. Abacus-based mental calculation is a unique traditional Chinese cultural practice and in a broad sense is a special method for accomplishing mental calculations. A well-trained abacus expert can perform mental calculation, even very complex ones, exceptionally fast and with high accuracy. For example, abacus experts can quickly retrieve the answer to a problem with contiguous two-digit additions (nine problems), which is normally considered difficult for non-experts, with response latencies of 2 to 3 seconds and only require about 8 to 12 seconds to correctly answer more complex four-digit contiguous additions. According to previous studies of the processes in mental abacus calculations , the experts were able to mentally implement the abacus operations depicted in Fig. 1 by interiorizing the actual manipulations of the abacus beads through a particular algorithm on a virtual abacus after long practice. Because all the sequential arithmetic steps and intermediate results could be directly encoded and retrieved through this virtual abacus which circumvents the limited capacity and slowness of existing mental arithmetic strategies, abacus learners could circumvent the limited capacity and slowness of the existing mental arithmetic strategies to also achieve fast and accurate calculations. Similar computing mechanisms were observed in the non-experts on contiguous addition problems, which need (A) (B) upper deck lower deck 27+16=? 2 7 thousands hundreds tens ones 4 3 step 1. step 2. step 3. Fig. 1 Abacus introduction and operation. a The abacus is typically constructed of various parts. Each bead in the upper deck has a value of 5; each bead in the lower deck has a value of 1. Beads are considered counted, when moved towards the horizontal beam that separates the two decks. Once a specific column of beads is defined as the ones (right-most column), then the next adjacent column to the left is the tens, the next adjacent column the hundreds, and so on. b For example, when performing the addition 27+16, the numeric representation of the number 27 is first placed. The simple addition +10 is applied by moving one bead to the left column of the lower deck. Then 6 (10 5+1) is added by adding both one bead from the lower deck on the row directly to the left (+10) and one bead from the lower deck (+1) and removing one bead from the upper deck ( 5), which completes the operation to retrieve simple arithmetic and recollect intermediate results directly from short-term memory. However, contiguous two-digit number operations appeared to be more difficult, as the non-experts suffered from heavy workload while performing complex computations requiring recollection of intermediate results quickly in their working memory. Each intermediate result in computations could be divided into three steps: (1) perform second-digit addition, (2) perform first-digit addition, and (3) retrieve the first step result from the working memory and continue to perform the addition to the second step result. This type of number processing for complex calculations in non-experts was relatively complicated and time-consuming. Several functional neuroimaging studies have been carried out to explore neural correlates of mental calculation [4 6, 8 13]. Current cognitive models, supported by some, postulate at least two representational formats for number: language-based and language-independent representation. The former is used to store tables of exact arithmetic knowledge, and the latter is for quantity manipulation and approximation relying on visuospatial
3 438 Eur J Nucl Med Mol Imaging (2009) 36: networks . Dehaene and Cohen further proposed a triplecode model of number processing that suggests that numerical information can be manipulated mentally in three formats: an analogical representation of quantities, a verbal format, and a visual Arabic representation . Specifically, the model also postulates the coordination of verbal processing and visuospatial strategies while performing multidigit operations in which rote knowledge and mental visualization of arithmetic procedure are required. However, little is known about the cognitive mechanisms involved in abacus-based mental calculation. To date, only a few research groups have investigated the underlying neural correlates of abacus-based mental calculations [14 16]. Tanaka et al.  and Hanakawa et al.  separately indicated that digital working memory and mental calculation in abacus experts are associated with enhanced involvement of neural resources in the frontalparietal circuit for visuospatial information processing. However, Chen et al.  pointed out underlying dissociated neural correlates (the frontotemporal and frontoparietal circuits) in child abacus experts. They all attempted to explore neural correlates and examine the effect of number size on mental operations, but their results were somehow inconsistent, including patterns of increasing and functional reorganization of regional activations. In this study, we measured the relative regional cerebral blood flow (rcbf) with 15 O-water PET in abacus experts and non-experts to investigate the neural processing bases of the cognitive ability. By comparing brain activation between the two groups during the processing of computational problems of different complexity, our study focused on the following unanswered questions: (1) what are the brain activation patterns in the abacus experts during calculation, (2) what are the neural processes shared with the non-experts, (3) what could be the cortical representation related to problem complexity, and (4) what activation areas in the abacus experts are unique and correlated with their exceptional computational ability. Materials and methods Subjects: abacus experts and non-experts The non-expert group comprised 12 right-handed subjects (6 male and 6 female; average age 25.8 years). All the subjects had received more than 14 years of formal education and had an average score for computational ability as tested by The College Entrance Exam in Mathematics. The expert group comprised ten right-handed abacus experts (five male and five female; average age 24.9 years) recommended by the Abacus Calculation Promoting Association. All experts were certified with at least fifth-level ability in both mental calculation and actual abacus operations. To be qualified as an expert (level 4 masters and above), one should be able to handle 5 10 digits per second and make few errors in mixed addition and subtraction problems. Most of the abacus experts began to develop their calculation ability at about the age of 10 years and continuously practised for 1 to 2 hours per day to maintain their exceptional ability in mental calculation. The initial test results for the two groups, with different types of calculation problems are given in Table 1. Both groups were situated free from anxiety and on the basis of their performance. We also contrasted problem-solving requirements with the response time and the correct rate to equalize problem complexity in the two groups. Answering nine contiguous addition problems with single-digit and two-digits, the experts performed nearly perfectly. In the four-digit calculations, the correct rate was down to 70%, and the response time was increased about fourfold compared to the single- and two-digit calculations. For the non-experts, similar results were observed for single- and two-digit calculations. The four-digit calculations were considered too difficult for the non-experts. Therefore, the simple calculations for the experts and non-experts were set to the two-digit and single-digit Table 1 Examples of mental calculations of different complexity performed by abacus experts and non-experts Type of problem Example Number of calculations Response latency (s) Correct answer (%) Non-experts Experts Non-experts Experts Single-digit addition a =? 5 48±11 10±3 94± Two-digit addition a =? 5 207±47 12±2 73± Four-digit addition a =? 5 47±10 78±18 Multiplication =? 5 21±3 88±13 Division =? 5 35±13 68±11 a Nine contiguous additions were performed.
4 Eur J Nucl Med Mol Imaging (2009) 36: addition calculations, and the complex calculations were set to the four-digit and two-digit problems, respectively. Experimental tasks Covert reading condition To reveal the neural network specifically for the procedure of mental calculation, a covert reading paradigm was designed for subsequent statistical comparison. In the covert reading condition, the experimental arrangement was similar to the calculation condition in which ten stimulation slices including nine slices of Arabic digits for continuous additions and the last one for the report of their answers were presented to subjects, the difference being that the subjects were asked solely to read the number covertly during the first nine slice presentations without performing the calculation. Simple calculation condition All the non-experts were presented with nine contiguous one-digit additions, and asked to report the answers aloud. To minimize error rates in the non-experts, each addition was limited to a sum of less than 50. Similar arrangements were designed for the experts but with two-digit additions, and each addition was limited to a sum of 500. Complex calculation condition All the non-experts were presented with nine contiguous two-digit additions and asked to report the answers aloud. For the non-experts each addition was limited to a sum of less than 500. Four-digit additions problems were prepared for the experts with no limit to the sums. PET image acquisition Nine sequential rcbf PET measurements with 15 O-labelled water were obtained during the three conditions repeated three times in random order. Each rcbf measurement was acquired on a Scanditronix PC WB (Uppsala, Sweden) PET scanner using the interleaved scan technique comprising thirty contiguous brain images with a voxel size of mm. For each study, one set of static PET data were acquired in 2-D mode based on the following protocol: at time 0, PET data acquisition started and the subject presented one by one with a set of eight calculation tasks (30 s for each calculation for a total of 240 s); at 30 s (immediately after conclusion of the first calculation), intravenous bolus injection of 30 mci 15 O-labelled water; data acquisition stopped after conclusion of all eight tasks (i.e., 210 s data acquisition after tracer injection). The image was then reconstructed with correction for head attenuation using a measured transmission scan. During the experimental tasks, the Arabic number stimuli in black on a light-grey background were projected onto a 60-inch screen hanging from the ceiling such that the subjects could see the entire screen. Each task consisted of eight problems, and ten stimulation slices were presented at fixed time intervals of 3 s in a problem. A digital recorder was prepared to record the verbal answers from the subjects and was used for checking the accuracy of their answers. To effectively monitor the correct rate in the non-expert group, the problems were prolonged to 50 s for the complex calculation condition, and only six problems were given for an rcbf measurement. Image data analysis After PET acquisition, all imaging data were realigned with automated image realignment (AIR) software  using rigid body transformations (six degrees of freedom). The original images were then transformed onto a standardized stereotactic Talairach space using the SPM2 (Wellcome Department of Imaging Neuroscience, London, UK) and the MNI (Montreal Neurological Institute) templates, sampled at a voxel size of mm. The transformed images were further smoothed using a threedimensional 12-mm full-width at half-maximum (FWHM) Gaussian filter to compensate for intersubject differences and to suppress high-frequency noise in the images. Global differences in CBF within and between the subjects were removed by scaling. Statistical comparisons across conditions were made using a t test. Statistical parametric maps corresponding to comparisons between the simple and complex computational tasks were generated with SPM2 to identify voxels that showed a statistically significant change, implemented in Matlab (Mathworks, Sherborne, MA). Three main comparisons were carried out using cognitive subtraction analysis. Increases in brain activity were evaluated (1) during simple calculation minus covert reading, (2) during complex calculation minus covert reading, and (3) during complex calculation minus simple calculation in both the experts and the non-experts. For each comparison, the voxel amplitude t map was transformed into a Z volume with a threshold Z 0 =3.09, which corresponds to p (uncorrected for multiple comparisons). For the complex calculation versus simple calculation comparison, the simple calculation versus covert reading comparisons were masked (mask set at p<0.05) in order to avoid artificial differences due to decreases in simple calculation conditions. To explore the task-related interest in each group, within-group analysis was performed in terms of the conjoint effects across the subjects. The statistical threshold was set at p<0.001 (uncorrected for multiple comparisons). In addition, conjunction analyses using orthogonalized contrasts 
5 440 Eur J Nucl Med Mol Imaging (2009) 36: were performed to uncover the activated areas common to abacus experts and non-experts while performing complex calculation tasks. To further identify activities specific to the two groups during calculations, betweengroup analysis was performed using a two-sample Student s t test. We focused only on differences in brain activation during complex calculation versus covert reading. Analysis was performed using a threshold at p<0.001 (uncorrected for multiple comparisons). The statistical images were colour-coded and then overlaid onto the MNI MR template to produce a combined PET/ MR image suitable for visualization. Results a simple calculation vs. covert reading b complex calculation vs. covert reading Arithmetic skill Correct rate and response latencies for each subject group (non-expert and expert) in the initial test study (Table 1) were used to decide the experimental set-up for the simple and complex calculation paradigm. During experiment tasks, correct rates in solving the simple and complex calculations were 94% and 73% for the non-experts and 100% and 78% for the experts, respectively. Simple calculations were solved by both groups with very satisfactory accuracies (94% and 100%). However, computational performance was worse for complex calculations in both groups, confirming a certain degree of difficulty (73% and 78%) in the tasks, which allowed the neural correlates under states of higher loading to be investigated. During actual scanning, correct rates in solving simple and complex calculations were 98% and 67% for the nonexperts and 99% and 83% for the experts, respectively. Performances during simple calculations by both groups were similar to those of the initial test studies which both revealed very satisfactory accuracies (98% and 99%). However, the differences became more significant for complex computations compared to those for simple computations, confirming the existence of different arithmetical skills between the two groups (67% and 83%). PET results c complex calculation vs. simple calculation Fig. 2 PET activation patterns in non-experts during (a) simple calculation vs. covert reading task, (b) complex calculation vs. covert reading task, and (c) complex calculation task vs. simple calculation task are shown overlaid on a surface-rendered standard anatomical space, and representative coronal slices are chosen for display (p< 0.001, uncorrected for multiple comparisons) The brain activation patterns of the non-experts are shown in Fig. 2 and the activated areas are shown in Table 2. Simple and complex tasks induced almost the same activation areas with left-hemisphere predominance. Activities were found in the bilateral superior parietal lobule (SPL, BA 7), left precentral gyrus (premotor cortex) and right superior medial frontal gyrus (supplementary motor area, pre-sma) including the anterior cingulated cortex (ACC) for both simple calculation and complex calculation tasks. Moreover, significant activation was also found in the left inferior frontal gyrus (BA 9, 45, 46) for complex calculation tasks, and the region in the inferior parietal lobule (IPL, BA 40) including the intraparietal sulcus extending anteriorly to the postcentral gyrus and laterally into the angular gyrus (AG). The brain activation patterns of the abacus experts are shown in Fig. 3 and the areas are shown in Table 3. The simple calculation tasks induced activation in the in the left precentral gyrus and the right precentral gyrus (premotor cortex). Activation in the bilateral precuneus and the left IPL (BA 40) was also found. In addition, there was activation in the right cingulate gyrus (BA 31). The complex calculation tasks induced more focused and symmetric activation patterns in the precentral gyrus (premotor cortex), and SPL (BA 7). Activation extending from the left precuneus (BA 7) through the SPL laterally to the IPL (BA 40) was also observed. No activation was observed in classical language areas of the left hemisphere. Finally, activation was observed in the right medulla and cerebellum. During complex calculation versus simple calculation tasks, the non-experts and experts (Figs. 2c and 3c) showed activation areas primarily in the medial frontal gyrus and SPL (including in the vicinity of the precuneus). However, the ACC and pre-sma showed increasing activation in the non-experts.
6 Eur J Nucl Med Mol Imaging (2009) 36: Table 2 Foci of significant rcbf increases in non-experts between different tasks Task Region Coordinates (mm) Z score x y z Simple calculation versus covert reading Frontal lobe Right pre-sma (BA 6) Left precentral gyrus (BA 6) Right precentral gyrus (BA 6) Parietal lobe Left postcentral gyrus (BA 4) Left inferior parietal lobe (BA 40) Temporal Left fusiform gyrus (BA 37) cortex Right fusiform gyrus (BA 37) Complex calculation versus covert reading Frontal lobe Right anterior cingulated cortex (BA 24/ ) Right superior medial frontal gyrus (BA 6) Left precentral gyrus (BA 6) Left middle frontal gyrus (BA 45/46) Left inferior frontal gyrus (BA 6/9) Parietal lobe Left superior parietal lobe (BA 7) Right superior parietal lobe (BA 7) Left inferior parietal lobe (BA 40) Right inferior parietal lobe (BA 40) Complex calculation versus simple calculation Frontal lobe Right anterior cingulated cortex (BA 32) Left pre-sma (BA 6) Left precentral gyrus (BA 6) Parietal lobe Right superior parietal lobe (BA 7) Left inferior parietal lobe (BA 40) Note: Within these regions, the anatomical localization of the extrema was based on MNI template anatomical analysis. The Z map was obtained using a threshold Z 0 =3.09 (BA Brodmann area; p<0.001, uncorrected for multiple comparisons). Brain activities specific to the non-expert and expert groups, and areas common to the two groups during calculation are depicted in Fig. 4 and details are shown in Table 4. The non-expert group showed hyperactivity in the superior frontal and left frontal areas (Fig. 4a). It has been suggested that these areas are associated with executive functions such as the ability to plan, initiate, coordinate a sequence of processes and place them in appropriate orders . On the other hand, activity in the left IPL (BA 40) was greater in the experts than the non-experts (Fig. 4b). This area is considered to be involved in the storage of immediate results on the virtual abacus image. The Fig. 4c also shows that there were similarities in the activation areas involved in visuospatial processing between the two groups. Discussion In the non-expert group (Fig. 2), activation areas in the prefrontal, premotor and parietal cortices, and the ACC with a tendency toward left lateralization, which have been often indicated as neural substrates constituting verbal and visuospatial strategy, were observed in both simple and complex tasks. Overall, our results are generally consistent with those of previous work [4, 6, 8, 9, 12, 19, 20] in which these activation regions reveal coordination of verbal processing for retrieving rote arithmetic facts, and back-up visuospatial strategy for mental visualization of the arithmetic procedure when rote knowledge is not available . We also observed that the calculation procedure adopted by the non-experts facing difficult tasks was mediated by coordination of several neural networks with different weights. Thus, comparing the results with complex calculation conditions, the non-experts tended to shift to a visuospatial strategy (left precentral gyrus and right SPL) to attempt to solve increasingly complicated problems. We therefore postulate that there is greater involvement of a verbal strategy for simple calculations and of a visuospatial strategy for complex problems. Moreover, involvement of the ACC and superior frontal gyrus may imply that neural networks are not effectively connected resulting in the need for more effort to coordinate a sequence of processes and place them in an appropriate order for higher weighted computation loadings. While examining the patterns of cerebral activation in the experts (Fig. 3), we observed comparable results in nonexperts and experts especially in the left hemispheric parietal areas, showing that both groups used visuospatial processes during mental calculation, whereas the non-
7 442 Eur J Nucl Med Mol Imaging (2009) 36: a simple calculation vs. covert reading b complex calculation vs. covert reading c complex calculation vs. simple calculation Fig. 3 PET activation patterns in abacus experts during (a) simple calculation vs. covert reading task, (b) complex calculation vs. covert reading task, and(c) complex calculation task vs. simple calculation task are shown overlaid on a surface-rendered standard anatomical space, and representative coronal slices are chosen for display (p<0.001, uncorrected for multiple comparisons) experts relied more on verbal strategies. Our results also showed that brain activation patterns in the experts during mental calculation were more focally and symmetrically distributed than those in the non-experts, in which the activities were more extensive. These dissimilar brain activation patterns during mental calculation probably reflect different cognitive strategies adopted by the two groups, accounting for discrepancies in their computation ability. Looking further into cortical activation in the abacus experts, their calculation networks were much more symmetrically distributed in the precentral gyrus at the levels of middle frontal gyrus and parietal lobe. Participation of the right hemisphere indicates that visuospatial information processing is involved in mental calculation , because the posterior superior parietal cortex is responsible for generation of mental numerical images (precuneus) . Also these visual images can be transferred to the right parietal cortex (precuneus) for further manipulation and comparison . These results possibly suggest that the abacus experts had developed a spatial representation of numbers as bead positions on a virtual abacus in their mind (left parietal cortex) and performed all computation steps through rule-based visuomotor processing of the abacus beads (right precentral gyrus and right parietal cortex) [24, 25]. In conclusion, mental calculation by the abacus experts was likely associated with enhanced involvement of neural resources for visuospatial information processing in a two- Table 3 Foci of significant rcbf increases in abacus experts between different tasks Task Region Coordinates (mm) Z score x y z Simple calculation versus covert reading Frontal lobe Left precentral gyrus (BA 6) Right precentral gyrus (BA 6) Parietal lobe Left precuneus (BA 7) Right precuneus (BA 7) Left inferior parietal lobe (BA 40) Right cingulate gyrus (BA 31) Complex calculation versus covert reading Frontal lobe Left precentral gyrus (BA 6) Right precentral gyrus (BA 6) Parietal lobe Left superior parietal lobe (BA 7) Right superior parietal lobe (BA 7) Left precuneus Left inferior parietal lobe (BA 40) Temporal cortex Left superior temporal gyrus (BA 22) Cerebellum Right cerebellum (anterior part) Brainstem Right medulla Complex calculation versus simple calculation Frontal lobe Left precentral gyrus (BA 6) Parietal lobe Left superior parietal lobe (BA 7) Right superior parietal lobe (BA 7) Note: Within these regions, the anatomical localization of the extrema was based on MNI template anatomical analysis. The Z map was obtained using a threshold Z 0 =3.09 (BA Brodmann area; p<0.001, uncorrected for multiple comparisons).
8 Eur J Nucl Med Mol Imaging (2009) 36: a Non-expert > Expert the same activation clusters in the visuospatial network, suggesting that both simple and complex tasks are relying on the same computational strategy. Specifically, additional activation was found in the right precentral gyrus, and this may be the reason why the abacus experts found simple calculation tasks too easy needing few actual calculation operations. These areas similarly constitute a circuit involved in visuospatial-dependent encoding and retrieval of the imaginary abacus. However, more complex operb Expert > Non-expert c Common area shared by Non-expert and Expert Fig. 4 Between-group comparison. a, b PET activation difference map between the two subject groups during the complex calculation task relative to the simple calculation task: a activation areas that the expert group recruited more than the non-expert group; b activation areas that the non-expert group recruited more than the expert group. c Overlap of areas in the two subject groups exhibiting significant activity during the complex calculation task relative to the covert reading task overlaid on a surface-rendered standard anatomical image (p<0.0001, uncorrected) dimensional space. This result shows that the abacus experts mainly utilized visuospatial strategy for computation through extensive abacus training, and this strategy increased their computational ability and accuracy because it may be more efficient to mentally manipulate large numbers using a spatial representation than a sequentially organized phonological representation. Comparing the differences between complex and simple calculations in the expert group (Table 3), one can observe Table 4 Between-group comparisons during complex calculation versus covert reading tasks Comparison Region Coordinates (mm) Z score x y z Non-expert > expert Frontal lobe Left pre-sma (BA 6) Left precentral gyrus (BA 6/9) Left precentral gyrus (BA 6) Parietal lobe Left inferior parietal lobe (BA 40) Left postcentral gyrus (BA 5) Temporal lobe Right middle temporal gyrus (BA 19/39) Expert > non-expert Parietal lobe Left inferior parietal lobe (BA 40) Areas common to non-experts and experts Frontal lobe Left precentral gyrus (BA 6) Parietal lobe Left superior parietal lobe (BA 7) Right superior parietal lobe (BA 7) Left inferior parietal lobe (BA 40) Note: Within these regions, the anatomical localization of the extrema was based on MNI template anatomical analysis. The Z map was obtained using a threshold Z 0 =3.09 (BA Brodmann area; p<0.0001, uncorrected)
9 444 Eur J Nucl Med Mol Imaging (2009) 36: ations must be recapitulated by visualizing a series of movements or visuospatial relationships, and to some degree this is an action-oriented object representation  of the imaginary abacus, which may explain the activation of the right precentral gyrus. In addition, previous studies [24, 25] have shown an increased right precentral activation known to play a role in attention, when subjects use a visual strategy during calculation. This finding suggests heavier computational loading would be needed to sustain selective attention to mental numerical images generated in the left medial parietal cortex (precuneus). In general, abacus experts can perform complex computations mentally with exceptional speed and accuracy since they only need to read answers from bead positions on a virtual abacus, just like using a calculator. However, complex problems involve relatively long numerical strings, and such long strings of digits must be memorized for further calculation, and hence temporary storage of results during mental calculation via a virtual abacus is crucial (left IPL, BA 40; Fig. 4b). The ability to record digits from the string shown on the virtual abacus seems to be a critical step in determining how many digits the experts can manipulate and hence the complexity of the problems they can accurately solve. Comparison of brain activation between the two groups showed greater involvement of the superior frontal and left frontal areas in the non-expert group (Fig. 4a). Involvement of these areas is related to the global workspace executive function, suggesting that these areas may play an important role in launching fairly time-consuming sequentially organized processes, including coordination of verbal processing and back-up visuospatial strategies. These activations during complex calculation tasks in the non-experts may be interpreted as deactivations in the abacus experts. Such pruning of attention and control ( scaffolding ) areas may be a general result of increasing familiarity with the tasks, and the final cleaner functional map of the processes, implying that exceptional speed and high accuracy in computation was due to increased activation of task-related neural systems and decreased activation of those implicated in attention, retrieval and monitoring. We have observed that intensive practice of cognitive tasks increases the efficiency of the distributed task network, and that this increased efficiency is observed as changes in size within separate areas as these areas become more efficient at performing their particular function. Conclusion The present results confirm the notion that through extensive training, abacus experts build up an effective computational pathway to circumvent an original relatively slow computational strategy enabling a further reduction in computation times. Coordination of a visuospatial network (bilateral frontal parietal network) facilitates number representation and operation by reassigning numbers onto a virtual abacus, which allows the whole calculation process to be retrieved and visualized directly from positions of the beads in this virtual abacus, not surprisingly shortening computation times significantly. Acknowledgements This study was financially supported from Ministry of Education (Promote Academic to Excellency Project, 89-B- FA22-1-4) and National Science Council (NSC B MY3; B MY2) of Taiwan. The project was conducted in the campus of NYMU/TVGH. References 1. Karni A, Meyer G, Rey-Hipolito C, Jezzard P, Adams MM, Turner R, et al. The acquisition of skilled motor performance: fast and slow experience-driven changes in primary motor cortex. Proc Natl Acad Sci USA 1998;95: Kolb B, Whishaw IQ. Brain plasticity and behavior. Annu Rev Psychol 1998;49: Buonomano DV, Merzenich MM. Cortical plasticity: from synapses to maps. Annu Rev Neurosci 1998;21: Dehaene S, Spelke E, Pinel P, Stanescu R, Tsivkin S. Sources of mathematical thinking: behavioral and brain-imaging evidence. Science 1999;284: PesentiM,ZagoL,CrivelloF,MelletE,SamsonD,DurouxB, et al. Mental calculation in a prodigy is sustained by right prefrontal and medial temporal area. Nat Neurosci 2001;4: Zago L, Pesenti M, Mellet E, Crivello F, Mazoyer B, Tzourio- Mazoyer N. Neural correlates of simple and complex mental calculation. Neuroimage 2001;13: Hatano G. Becoming an expert in mental abacus operation: a case of routine expertise. Adv Jpn Cogn Sci 1983;1: Burbaud P, Camus O, Guehl D, Bioulac B, Caille JM, Allard M. A functional magnetic resonance imaging study of mental subtraction in human subjects. Neurosci Lett 1999;273: Burbaud P, Camus O, Guehl D, Bioulac B, Caille JM, Allard M. Influence of cognitive strategies on the pattern of cortical activation during mental subtraction. A functional imaging study in human subjects. Neurosci Lett 2000;287: Dehaene S, Cohen L. Towards an anatomical and functional model of number processing. Math Cogn 1995;1: Dehaene S, Piazza M, Pinel P, Cohen L. Three parietal circuits for number processing. Cogn Neuropsychol 2003;20: Kawashima R, Taira M, Okita K, Inoue K, Tajima N, Yoshida H, et al. A functional MRI study of simple arithmetic a comparison between children and adults. Brain Res Cogn Brain Res 2004;18: Rickard TC, Romero SG, Basso G, Wharton C, Flitman S, Grafman J. The calculating brain: an fmri study. Neuropsychologia 2000;38: Hanakawa T, Honda M, Okada T, Yonekura Y, Fukuyama H, Shibasaki H. Neural correlates underlying mental calculation in abacus experts: a functional magnetic resonance imaging study. Neuroimage 2003;19:
10 Eur J Nucl Med Mol Imaging (2009) 36: Tanaka S, Michimata C, Kaminaga T, Honda M, Sadato N. Superior digit memory of abacus experts: an event-related functional MRI study. Neuroreport 2002;13: Chen F, Hu Z, Zhao X, Wang R, Yang Z, Wang X, et al. Neural correlates of serial abacus mental calculation in children: a functional MRI study. Neurosci Lett 2006;403: Woods RP, Grafton ST, Holmes CJ, Cherry SR, Maziotta JC. Automated image registration: I. General methods and intrasubject validation. J Comput Assist Tomogr 1997;22: Price CJ, Friston KJ. Cognitive conjunctions: a new approach to brain activation experiments. Neuroimage 1997;5: Dehaene S, Tzourio N, Frak V, Raynaud L, Cohen L, Mehler J, et al. Cerebral activations during number multiplication and comparison: a PET study. Neuropsychologia 1996;34: Paulesu E, Frith CD, Frackowiak RS. The neural correlates of the verbal component of working memory. Nature 1993;362: Molko N, Cachia A, Rivière D, Mangin JF, Bruandet M, Le Bihan D, et al. Functional and structural alterations of the intraparietal sulcus in a developmental dyscalculia of genetic origin. Neuron 2003;40: Mellet E, Tzourio N, Crivello F, Joliot M, Denis M, Mazoyer B. Functional anatomy of spatial imagery generated from verbal instructions. J Neurosci 1996;16: Formisano E, Linden DE, Di Salle F, Trojano L, Esposito F, Sack AT, et al. Tracking the mind s image in the brain I: timeresolved fmri during visuospatial mental imagery. Neuron 2002;35: Wise SP, Boussaoud D, Johnson PB, Caminiti R. Premotor and parietal cortex: corticocortical connectivity and combinatorial computations. Annu Rev Neurosci 1997;20: Courtney SM, Petit L, Maisong JM, Ungerleider LG, Haxby JV. An area specialized for spatial working memory in human frontal cortex. Science 1998;279: Jancke L, Loose R, Lutz K, Specht K, Shah NJ. Cortical activations during paced finger-tapping applying visual and auditory pacing stimuli. Brain Res Cogn Brain Res 2000;10:51 66.