The data used in this study are part of the project ‘The temporal dynamics of neurocognitive changes induced by complex task training in the form of strategic computer game’ conducted at the NeuroCognitive Research Center (SWPS University in Warsaw) founded by the National Science Center. The research was approved by the Ethics Committee of SWPS University (number: 38/2018).
2.1 Participants
In the whole study participants took part in the structural magnetic resonance imaging (sMRI) acquisition, cognitive evaluation, and EEG (electroencephalography) session at four time points: before VG training (T0), after 10 hours of training (T1), after 30 hours of training (T2) and after 60 hours of training (T3). The presented study is focused on pre-training (TP0), diffusion magnetic resonance imaging (dMRI sessions) and the first 30 hours of VG training (Fig. 1).
A total of 23 subjects that performed approximately 30 h (± 3 h) of StarCraft II training and completed the 3rd time point measurements (T3) were included in this study. One participant was excluded due to issues with dMRI images, and the final sample consisted of 22 participants (female = 15) with a mean age = 27.45 (SD = 5.12).
2.2 Study procedure
First, subjects completed an online survey26 about VG playing experience, education, and demographics through the GEX platform (GEX Immergo, Funds Auxilium Sp. z o.o). The survey included questions about age, gender, level of education, marital status, work experience, total experience with video games in years, frequency of playing video games, devices used to play video games, game genres played by participants, subjective level of gaming expertise and the frequency of playing specific game genres. Participants' inclusion criteria were: [1] little or no previous experience with RTS video games, [2] experience with other types of video games with no more than 8 h / week during the last six months. All participants declared that they had no history of neurological illness, no use of psychoactive substances, and reported normal vision and hearing. All subjects signed an informed consent in accordance with Declaration of Helsinki before the first measurement session to participate in the study and completed health-related questions prior to MRI sessions. Participants also carried out nonverbal group test, Raven's Progressive Matrices27 in order to control higher-order intellectual functions (mean score for all participants = 24.14, SD = 5.30).
Subjects recruited for the study were randomly assigned to two training groups: the experimental ‘Variable’ group (n = 13, Mage = 27.1, SD = 3.52, 10 females) and an active control group ‘Fixed’ (n = 9, Mage = 28, SD = 7.05, 5 females). The groups differed on the basis of the opponent's race and strategy for each match. Since the main focus in our study is on the overall game skill acquisition, we have combined these two groups into one. Furthermore, the sample size was not sufficient to perform a separate analysis.
2.3 Training task - StarCraft II
Participants underwent StarCraft II gaming sessions (Fig. 2) in controlled laboratory settings with the prohibition of playing outside of the laboratory. Before the first StarCraft II game, each participant underwent an introductory training session with the StarCraft II coach. The StarCraft II training in the presented study consisted of 30 hours, lasting between 3 and 4 weeks with a minimum duration of 5 h and a maximum duration of 10 h per week.
Training was carried out using a dedicated desktop PC running Windows 7 (professional edition, 64-bit operating system) equipped with a dedicated graphics card (NVIDIA GeForceGTX 770), 8GB of RAM, and a 24′′LED display which allowed to play at the high graphic quality (1920×1080 pixels resolution, 60Hz). The participants played the game using a mouse, keyboard and headset.
2.4 Telemetric data from StarCraft II
Telemetric variables were obtained from StarCraft II Replays, using sc2reader (retrieved from https://github.com/ggtracker/sc2reader) and PACanalyzer (retrieved from https://github.com/Reithan/PACAnalyzer), which are Python libraries to extract information from various StarCraft II resources. The mentioned libraries allowed us to extract replay details, such as information on the used map, length of the match, game type, etc., but also more specific data such as (1) player details, (2) unit selection and hotkeys events, (3) resource transfers and requests, (4) unfiltered unit commands such as attack, move, train, built, etc., (4) camera movements. During the study, we collected a total of 1980 StarCraft II replays from 22 participants [each participant produces approximately 90 (SD = 17.5) matches during the training process]. After initial extraction, we filtered matches that lasted longer than 60 minutes or shorter than 1 minute, assuming that these are the results of a computer or user error (such as starting the wrong type of match). Then we have calculated the mean length of the matches for each of the participants. The next filter cut matches longer and shorter than the mean length of played matches ± 1 standard deviation. Telemetric data was analyzed from a total of 1519 matches. Then it is important to mention that game proficiency does not depend only on one specific variable. We decided to focus on basic variables, which at the same time are very sensitive to player’s skills development (Thompson et al., 2017).
(1) Action per minute (APM), which is the average number of actions performed per minute in the game and can be interpreted as cognitive, motor, and decision-making speed. APM is the most popular variable used to show how skilled the player is, but only to a certain extent. As our participants were able to play only for a 30-hour period, the increase in APM perfectly reflects the process of gaining the ability to play StarCraft II.
(2) Perception Action Cycles (PACs) Action latency, which is a switch in the focus of attention to a new location, followed by the last one. While the first matches of the participants were played only in one observed region (PAC = 1), along with gaining experience in the game, the participants were able to simultaneously take actions in many distant and independent places on the map.
(3) HS (hotkeys) usage, the average number of hotkey presses per minute in each game, where each of these actions represents an automated selection of multiple units or buildings25. Using hotkeys allows players to play in a more efficient way, to perform some actions simultaneously and much faster, and is necessary if players significantly want to improve their performance.
All three of the chosen variables are characterized by a low starting score, resulting from zero experience of playing, and a regular increase with the subsequent matches played.
It should be noted that although all players showed the same normal trends in development in the discussed variables, the intensity of those increases largely depends on the individual capabilities of the player. Selected variables allow for the observation of the acquisition of abilities at the general level, at the group level, and also at the level of individual player achievements.
2.5 Magnetic resonance imaging (MRI)
2.5.1 MRI image acquisition. All magnetic resonance (MR) images were collected using a GE Discovery MR750w 3T MRI scanner (CNS Lab, IBBE PAS, Warsaw, Poland) before RTS training. The MRI scanner was equipped with an 8-channel phased-array head coil. Foam padding was used around the head of the participants to provide comfort and minimize head movements. Subjects were asked to remain still and not fall asleep, while maintaining a comfort level as high as possible.
The spin-echo diffusion weighted echo planar imaging (DW_EPI) sequence was performed with TR = 16000 ms, TE = 100 ms, flip angle = 90, and voxel size = 2 x 2 x 2 mm3. We used AP (anterior-posterior) and PA (posterior-anterior) phase coding directions, 40 volumes with a b-value of 1000 s/mm2 for AP, 40 volumes with a b-value of 1000 s/mm2 for PA, along with 4 images with applying no diffusion gradient (b-value = 0) for both AP and PA (eight b0 images in total).
2.5.2 Diffusion data preprocessing. The entire diffusion data preprocessing procedure was performed in FSL 6.0.428 (FMRIB Software Library, http://www.fmrib.ox.ac.uk/fsl/). First, the B0 images were extracted from the NIfTI files and combined into a single-volume image. Then, a topup tool was used to estimate and correct the susceptibility-induced field distortions29. We used the FDT FMRIB Diffusion Toolbox for the analysis of diffusion weighted images30,31. Head motions and distortions produced by the gradient coils were reduced using the eddy current correction32. The Brain Extraction Tool (BET) was used to isolate and remove nonbrain tissue with a fractional intensity threshold of 0.3. The DTIFIT command was then used to calculate the diffusion tensors and fractional anisotropy (FA). The FA images for all participants were registered to the MNI standard space (FMRIB58_FA_1mm template) using FSL’S FLIRT33,34 and FNIRT35 tools.
2.6 MRI data analysis
All steps for the diffusion data analysis of both approaches were performed with FSL 6.0.428 (FMRIB Software Library, http://www.fmrib.ox.ac.uk/fsl/).
For WM regions (no tractography) the masks were created using the JHU-ICBM DTI 81 White Matter atlas (Mori et al., 2005). Then FA metrics for each person and each region of interest were calculated with a threshold of 10%10. For the probabilistic tractography approach, Bedpostx tool30,31 was run to build and estimate the distributions of diffusion parameters at each voxel and model the crossing fibers using Markov Chain Monte Carlo sampling. Probabilistic tracking was performed using the Probtrackx tool30,31, with 5000 samples taken for each input voxel with a 0.2 curvature threshold. The FA values for each extracted pathway were calculated with a threshold of 10% to exclude low-probability voxels10. Figure 3 represents the steps for processing the diffusion data.
2.7 White matter regions and tracts selected for analysis.
The white matter regions of interest were chosen based on the literature described in the Introduction section and presented in Fig. 4. For the approach based on the extraction of FA values from the WM regions (without tractography) we focussed on the bilateral anterior limb of the internal capsule (ALIC), bilateral posterior limb of the internal capsule (PLIC), bilateral retrolenticular limb of the internal capsule (RLIC), bilateral external capsule (EC), bilateral cingulum/hippocampus (CG/HIP) and bilateral fornix stria terminalis (FX/ST). The external capsule and parts of the internal capsule were chosen based on the research by Kowalczyk-Grębska et al.9. The cingulum and fornix areas are based on the study by Ray et al.15.
For the probabilistic tractography approach, motor and visual pathways were chosen: bilateral corticospinal tract (CST), bilateral superior longitudinal fasciculus (SLF), bilateral inferior longitudinal fasciculus (ILF), and bilateral inferior fronto-occipital fasciculus (IFOF). The tracts were chosen based on the study by Zhang et al.10 where the results revealed significant differences in these tracts between VGPs and amateurs.
2.8 Statistical analysis
To evaluate the association between white matter microstructure and game learning, correlation analyzes were performed using Jamovi software, version 2.3.12 (retrieved from https://www.jamovi.org, 2001) and RStudio software, version 2022.02.0 + 443, (retrieved from http://www.rstudio.com/, 2020).