In accordance with a previous study [51], we defined gaming group players based on skill and time commitment. For gaming skill criteria, we referenced rankings published in November 2023 [52]. Inclusion criteria for the FPS group were (1) ≥5 years of experience in PC-based FPS, ≥5 hours per week of gameplay in the last 6 months, and ranked within the top 15% of all players; and (2) no MOBA experience or no MOBA gameplay within the last 2 years. Inclusion criteria for the MOBA group were (1) ≥5 years of experience in PC-based MOBA, ≥5 hours per week of gameplay in the last 6 months, and ranked within the top 15% of all players, and (2) no FPS experience or no FPS gameplay within the last 2 years. Inclusion criteria for the control group were no MOBA or FPS experience, no gameplay within the last 2 years, and no gaming experience in the last 6 months.

Exclusion criteria included visual or auditory impairments, metallic implants or electronic devices in the head, neurological or psychological disorders, alcohol or drug abuse, smoking, and claustrophobia.

Consistent with previous research [53], it was challenging to recruit a sufficient number of high-level female players; therefore, all participants in this study were men. All participants were right-handed undergraduates or postgraduates.

Participants were recruited through online advertisements posted at Shanghai University of Sport. As part of the initial screening process, potential candidates completed a comprehensive questionnaire designed to verify the predefined inclusion and exclusion criteria. This questionnaire also gathered genre-specific information from the MOBA and FPS groups, including their competitive rankings and average weekly gaming hours over the previous 6 months. The next step involved collecting high-resolution brain structural images on participants who met the experimental criteria.

Sample size determination was informed by prior cross-sectional neuroimaging studies investigating video game genres [54,55]. To ensure adequate statistical power and robust effect sizes, we established a minimum target of 35 participants per group. A total of 116 participants were successfully recruited, including 39 (33.6%) FPS players, 40 (34.5%) MOBA players, and 37 (31.9%) healthy controls.

The primary measures in this study were individualized structural covariance edges (SCEs) constructed based on CT across 68 brain regions. Potential covariates, including age and years of education, were prespecified for inclusion in the statistical models if significant between-group differences were observed during demographic comparisons. Additionally, weekly gaming hours were collected as a behavioral measure for the predictive analysis.

Structural images were acquired using a 3-T Siemens Prisma MRI scanner with a 64-channel phased-array head coil and using a magnetization-prepared rapid gradient echo sequence (repetition time=2300 ms, echo time=2.98 ms, inversion time=450 ms, field of view=256 mm, flip angle=9°, voxel size=1×1×1 mm³, and 176 contiguous slices). The duration of the structural acquisition scan was 5 minutes.

High-resolution brain structural images were converted from DICOM to NIFTI format using MRIconvert (version 2.1.0; Lewis Center for Neuroimaging), followed by a quality check of each participant’s anatomical images to confirm the absence of significant artifacts. Anatomical images for each participant were realigned to the anterior commissure–posterior commissure plane using the Statistical Parametric Mapping (SPM12) software package [56] within the MATLAB 2020b environment [57], ensuring a consistent origin at the anterior commissure [58]. Subsequently, the Computational Anatomy Toolbox (CAT12.8.1 [59]) based on SPM12 was used to preprocess the MRI anatomical images of the 3 groups and to compute the CT. The specific steps included (1) standardization was conducted using Diffeomorphic Anatomical Registration using Exponentiated Lie algebra [60], followed by segmentation of the images into GM, WM, and cerebrospinal fluid; (2) calculation of CT: CT was calculated as the distance between the inner and outer surfaces of the GM [61]; (3) images were smoothed with a 15 mm FWHM Gaussian kernel; and (4) each hemisphere was segmented into 34 cortical regions using the Desikan-Killiany atlas [62], and the CT for all 68 brain regions across participants was extracted to construct the IDSCN.

The mean CT of all participants was within 3 SDs of the group mean. All participants met the image quality criterion of grade C or above, and therefore, no data were excluded from further analysis.

The construction process of the IDSCN was as follows [39]:

The edges in the IDSCN represent how the inclusion of gamer k altered the covariance of pairs of brain regions in CT compared to the reference group.

We used the Gretna toolbox [63] to conduct a one-way ANOVA based on the z scores of 4624 edges across the 3 groups. Multiple comparisons were controlled by applying a false discovery rate correction using the mafdr function, which implements the Benjamini-Hochberg procedure, at a threshold of P<.001 [64]. Significant edge differences for all participants were extracted, and post hoc comparisons were performed using Bonferroni correction in IBM SPSS Statistics (version 26.0; P<.05/3=.02) [65]. The results were visualized using BrainNet Viewer [66]. To further enhance the robustness of the results, we conducted an additional nonparametric permutation test (10,000 permutations) using network-based statistics (NBS) on the 4624 edges among the 3 groups [67], with the α level set at .05 and the F threshold corresponding to the minimum significant F value derived from the parametric tests. The demographic information was tested for normality using the Shapiro-Wilk test in SPSS (version 26.0). Age and years of education were analyzed using the Kruskal-Wallis H test, while weekly gaming hours were assessed using the Mann-Whitney U test for independent samples. The distribution of gaming levels between the FPS group and the MOBA group was analyzed using the 2-sample Wilcoxon rank-sum test.

The analysis predicting weekly gaming hours based on structural covariance matrices was conducted in accordance with the official NBS-Predict guidelines [50]. The significance threshold for the suprathreshold edge selection algorithm was set at P<.01, with 10,000 permutations used for statistical testing in correlation mode. Input features were scaled using MinMaxScaler, hyperparameter optimization was conducted via a grid search algorithm, and the number of optimization steps was set to 5.

In machine learning, SVM represents a class of supervised learning models and serves as a classic binary classification tool [68]. The construction of the SVM is consistent with a previous study [51]. We used SVM to assess the stability of connectivity differences between groups. The SVM analysis was conducted using the fitsvm function in MATLAB 2023a [57]. All differential edges in the FPS and MOBA groups were considered as feature values, with mutual information used as the feature selection algorithm [69], ranking feature values based on their importance. Leave-one-out cross-validation is a special case of k-fold cross-validation, where the number of folds equals the number of participants. It is particularly suitable for use when the dataset is small, and there is no need to randomly split the dataset into training and testing sets [70]. Therefore, we used a linear kernel SVM classifier and used the leave-one-out cross-validation strategy to evaluate its performance. The evaluation metrics included area under the curve (AUC), accuracy, sensitivity, and specificity [71]. Furthermore, SVM analysis was conducted on all differential edges between the FPS and the control groups, as well as between the MOBA and control groups. The visualization of the SVM results was performed using GraphPad Prism 9.5 [72].

This study involving human participants was reviewed and approved by the Ethics Committee of Shanghai University of Sport (approval 102772024RT028). All procedures were conducted in accordance with the Declaration of Helsinki, as well as relevant local legislation and institutional requirements. Written informed consent was obtained from all participants prior to their involvement, explicitly including provisions that allow for the secondary analysis of the collected research data without the requirement for further consent. Participants were informed that their participation was voluntary and their right to withdraw from the study at any time without penalty or the need to provide a reason. To ensure privacy and confidentiality, all data were anonymized. Furthermore, we confirm that no individual participant can be identified in any figures, tables, or supplementary materials presented in this manuscript. Upon completion of the experimental procedures, each participant received compensation of Chinese Yuan 35 (US $5) for their time and participation.

This study included 116 male participants, divided into 3 groups, aged 18 to 26 (mean 21.2, SD 1.9) years. The demographic and game information of the 3 groups is presented in Table 1. In the FPS group, 90% (35/39) of the players played Counter-Strike 2, and 26% (10/39) played Apex Legends, with 15% (6/39) playing both games. The MOBA group included 98% (39/40) League of Legends players and 2% (1/40) Dota 2 players. No significant differences were found in age or years of education among the study groups. We also observed no significant differences in the weekly gaming hours or rank distribution between MOBA and FPS groups.

The results of the one-way ANOVA for SCEs among the 3 groups are presented in Figure 1A and Table 2. A total of 30 SCEs were observed across the whole brain at the significant threshold of P<.001 (false discovery rate corrected). A consistent result was also obtained using an NBS analysis with 10,000 permutations (Multimedia Appendix 1).

Post hoc analysis indicated that compared to the MOBA and control groups, the FPS group displayed 2 dominant networks. These consisted of a temporo-fronto-parietal network anchored in auditory regions, specifically the superior and transverse temporal gyri, and a visuo-sensorimotor network involving the inferior temporal gyrus and postcentral gyrus (Figure 1B and C and Table 3).

Additionally, both gaming groups exhibited significantly enhanced connectivity within visual-attentional networks relative to the control group (Figures 1C and 1D and Table 3).

All the significant SCEs mentioned earlier were treated as feature values using the MI algorithm for importance ranking, with lower values for greater significance of the feature in distinguishing between each pair of groups. Detailed results are presented in Multimedia Appendix 2.

The weekly gaming time of the FPS group was predicted using the NBS-Predict based on individual structural covariance matrices. The predicted weekly gaming time showed a significant correlation with the actual values, with a Pearson correlation coefficient (r) of 0.34 (95% CI 0.26‐0.42). Within the linear regression model, we identified 100 positive edges and 7 negative edges (connecting 34 brain regions) with weight values >0.8 (Table 4 and Figure 2), where higher edge weights indicate greater contribution to the model. The positive edges constituted a connected network centered on the auditory-related transverse temporal gyrus and extending into frontal, parietal, and occipital lobes, involving regions such as the inferior temporal gyrus, superior temporal gyrus, pars orbitalis, rostral middle frontal gyrus, superior parietal gyrus, inferior parietal gyrus, and postcentral gyrus (Figure 2A). The negative edges primarily reflected connections between the entorhinal cortex and several frontal regions (eg, rostral middle frontal, caudal middle frontal, and frontal pole), the pericalcarine cortex, and the superior temporal cortex (Figure 2B). Furthermore, this pattern of positive edges showed substantial overlap with the differential network of the FPS group relative to the MOBA group (Figure 1B), with overlapping connections illustrated in Figure 2C. Additionally, a validation analysis was performed by increasing the weight threshold to 0.9, which revealed that the identified significant connections remained highly consistent with those obtained under the 0.8 threshold (Table 4 and Multimedia Appendix 3 and Multimedia Appendix 4).

No significant predictive effect was observed when using the structural covariance matrices of the MOBA group to predict their weekly gaming time.

The classification results showed that the FPS group could be differentiated from the MOBA group based on high AUC (72.37%), accuracy (70.89%), sensitivity (85%), and specificity (56.41%). The AUC, accuracy, sensitivity, and specificity were 82.95%, 73.68%, 64.10%, and 83.78%, respectively, for the classification between the FPS and control groups. The AUC, accuracy, sensitivity, and specificity were 74.26%, 67.53%, 57.50%, and 78.38%, respectively, for the classification between the MOBA and control groups (Figure 3).

This study aimed to examine whether long-term FPS and MOBA gaming experience is associated with the covariation of CT across brain regions, thereby elucidating the potential corresponding cognitive functions. Our results corroborated the initial hypotheses, revealing that compared to the MOBA and control groups, FPS players exhibited differences in CT covariation between the auditory cortex and the frontal, occipital, and parietal regions. These networks, involved in auditory processing, visual perception, attentional control, cognitive flexibility, and sensorimotor integration, also demonstrated predictive value for weekly gaming hours. Compared to the control group, MOBA players showed differences in CT covariation between the lateral occipital cortex and the inferior parietal and supramarginal regions, as well as between the superior parietal and the lateral orbitofrontal cortex. Furthermore, significant connectivity edges differentiating the FPS and MOBA groups were used as input features for an SVM classifier, and the results indicated that these structural signatures effectively discriminated between the 2 player groups.

Previous research has found that long-term experience with RTS games can enhance WM connectivity in the occipito-parietal regions [8], and such structural changes may further drive functional reorganization in frontoparietal and parieto-occipital brain regions [73]. Classical video game genres primarily include FPS, MOBA, and RTS. Early studies have demonstrated the benefits of video games for visual attention [5,74]. In this study, we observed covariation within frontal-parietal-occipital regions among MOBA and FPS players, which may reflect the influence of shared cognitive demands associated with these game genres. Two review articles on the cognitive benefits of video game training have suggested that not all video games elicit the same cognitive effects [3,75]. Consistent with this perspective, our findings revealed that FPS players exhibited more extensive covariation patterns across auditory, visual, and sensorimotor networks. This may indicate that compared with MOBA players, whose covariation patterns were primarily confined to visual-attentional networks, FPS gameplay imposes more comprehensive cognitive demands on players.

Longitudinal intervention studies investigating video game training have reported increases in GM volume in the superior temporal cortex [45] and the dorsolateral prefrontal cortex [76]. The superior and transverse temporal gyri are key anatomical substrates for auditory processing [77,78], whereas prefrontal regions are primarily associated with cognitive flexibility [79]. These findings are consistent with the temporal-prefrontal covariance patterns observed in our study. This correspondence suggests that FPS players may rely on dynamic interactions between auditory processing networks and attentional control networks [19].

Visual stimuli, a core element of AVGs, produce stronger visual motion effects in first-person perspectives compared to third-person perspectives [80]. A cross-sectional study found that AVG experience might lead to improvements in attention observed in both the visual and auditory domains [17], and AVG training has been shown to improve phonemic decoding and visual attention in children with developmental dyslexia [81]. Long-term FPS intervention could increase GM volume in visual regions such as the middle occipital gyrus [45]. Notably, alterations in occipital networks have been linked to neurodegenerative conditions such as Lewy body dementia [10]. Moreover, auditory-perceptual game-based training has been shown to improve cognitive performance in older adults, an effect that depends on the integrity of occipito-temporal WM connections [82]. Thus, the observed auditory-visual covariance, which positively correlates with weekly gaming hours in FPS players, implies that FPS-like gaming paradigms may offer a promising approach to maintaining brain health and mitigating age-related cognitive decline.

Research has shown that long-term FPS interventions may increase CT of the sensorimotor regions [44]. The superior parietal and inferior parietal regions were associated with the execution control of visually related motor responses [83]. Moreover, the superior parietal cortex also serves as a critical region for hand movements [84]. In an auditory-motor integration task, the superior temporal and supramarginal regions were co-activated in pianists [85] who shared certain common features with AVG players in frequent hand movements. The joint engagement of superior temporal and inferior parietal cortices functions to modulate motor responses to auditory stimuli in relation to visual stimuli [86]. FPS games have been examined in motor skill research due to their distinct kinematic characteristics [13,14]. Furthermore, these games are rich in auditory cues. For example, players often rely on enemy footsteps to localize opponents and execute rapid sequences of actions to eliminate them. The observed SCEs between auditory and sensorimotor networks may imply an association with integrated audio-motor processing.

We found that the negative SCEs between the entorhinal cortex and the frontal regions as well as the superior temporal gyrus in FPS players could effectively predict weekly gaming hours. This is a noteworthy phenomenon that may be linked to negative cognitive effects associated with FPS gaming. West et al [43] found that spatial navigation strategies can increase GM volume in the entorhinal cortex. In contrast, the use of in-game GPS guidance in FPS games—directing players to subsequent locations or events—may encourage greater reliance on response learning strategies, thereby indirectly contributing to reductions in hippocampal and entorhinal regions involved in spatial navigation. Additionally, the observed negative edge between the entorhinal cortex and superior temporal gyrus resembles the reduced hippocampal-superior temporal functional connectivity reported by Benady-Chorney et al [87] in AVG players, which may further indicate potential memory impairment. With advancing age, humans tend to shift toward response learning strategies [88], and FPS gaming may exacerbate such aging-related cognitive risks. Therefore, future game design should aim to balance elements related to response learning and spatial navigation strategies to safeguard the memory system.

This study has several limitations. First, the cross-sectional design precludes causal conclusions about the relationship between game type and brain structure. It is also possible that inherent brain structure specificity leads players to perform better in FPS games rather than MOBA games. Second, although we observed covariation of CT between the auditory cortex and several other regions in FPS players, no auditory tasks were designed and conducted, which limits our ability to directly correlate auditory behavior with structural network changes. Future research should use a longitudinal intervention design, coupled with task-based functional MRI, to explore the effects of FPS and MOBA games on players’ auditory attention. Finally, while IDSCN has been shown to resemble functional networks in several studies [31-33] and offers substantial advantages by overcoming the limitations of group-level structural covariance analysis through individual-level assessments, it serves only as an indirect marker of neuroplasticity. Its biological basis remains incompletely understood, and interpretation should be approached with caution.

This study characterizes the distinct structural network architectures associated with long-term FPS and MOBA gaming. The findings indicate that while both genres engage visual-attentional networks, FPS gaming is uniquely associated with extensive structural covariance between the primary auditory cortex and regions governing visual attention, sensorimotor processing, and multisensory integration. These distinct patterns likely reflect the broader cognitive demands of FPS gaming, supporting the potential utility of FPS-based paradigms in auditory and visual rehabilitation [89-92]. Additionally, the identified structural features offer a theoretical foundation for the assessment and selection of professional esports players. However, the observed negative edges involving the entorhinal cortex warrant caution. This may indicate that the heavy reliance on response learning strategies in FPS gaming comes at the expense of spatial navigation systems dependent on the entorhinal cortex. This potential trade-off is particularly relevant for designing game-based interventions aimed at mitigating age-related cognitive decline. Given the cross-sectional nature of this study, these structural associations cannot confirm causality. Future research should prioritize longitudinal designs combined with auditory behavioral tasks to verify whether these gaming habits actively induce neuroplasticity or reflect preexisting traits and to further elucidate the balance between cognitive benefits and potential risks to the spatial memory system.