Brain Circuits During Arithmetic Are Predictive of Real-Time Trial-Based Math Anxiety Measurement

721 

Abstract Submission 

Chan-Tat Ng¹, Yi-Cheng Cho¹, Xin-Yu Chen¹, Ting-Ting Chang^1,2

¹Department of Psychology, National Chengchi University, Taipei, Taiwan, ²Research Center for Mind, Brain, and Learning, National Chengchi University, Taipei, Taiwan

Chan-Tat Ng  
Department of Psychology, National Chengchi University
Taipei, Taiwan

Yi-Cheng Cho  
Department of Psychology, National Chengchi University
Taipei, Taiwan

Xin-Yu Chen  
Department of Psychology, National Chengchi University
Taipei, Taiwan

Ting-Ting Chang  
Department of Psychology, National Chengchi University|Research Center for Mind, Brain, and Learning, National Chengchi University
Taipei, Taiwan|Taipei, Taiwan

Math anxiety (MA) poses a global educational challenge, affecting learners of different ages (Barroso et al., 2021). While neuroimaging studies have provided insights into brain mechanisms of MA, they often focused on trait anxiety measure outside the scanner. This study addresses this gap by investigating the neural predictors associated with real-time MA during math problem-solving in adults using functional MRI and machine-learning techniques.

Fig 1 depicts the overall methodology. We recruited 43 adults (M = 22.6, SD = 2.12; 26 females) in this study. During fMRI scanning, participants performed math problems of varying complexity (Lyons et al., 2012), and real-time MA was evaluated via post-trial ratings, averaging standardized differences in emotional responses and perceived difficulty between complex and simple problems. Generalized Anxiety Disorder-7, Chinese-adapted Math Anxiety Questionnaire for Adults (MAQA; Szczygiel, 2022), WAIS-IV working memory, and arithmetic fluency (Chang et al., 2018) were assessed in separate sessions and used as control variables. Brain activity was analyzed using SPM12's general linear model. After a standard preprocessing pipeline, voxel-wise t-maps contrasting complex and simple problems were generated for each participant. Whole-brain group-level regression was performed to explore the differences in the relative brain activity between median-split groups based on real-time MA, identifying activations at a voxel-wise threshold of p < .001 and a cluster threshold of p < .01. Further analyses were conducted based on 21 predefined regions of interest (ROIs), identified via Neurosynth's term-based meta-analysis. We extracted clusters based on search terms "anxiety", "arithmetic", and "DMN" (default mode network), with a cluster size of k >= 50. LASSO regression with a logistic function was conducted 1000 times with 10-fold cross-validation for feature selection, identifying the most relevant ROIs based on a 75% frequency threshold of nonzero selection. Moderation analyses within the selected ROIs were conducted to assess the impact of MA on the brain-performance relationship.

Spearman correlation analyses revealed that real-time MA was moderately correlated with MAQA (r = .438, p = .005) and task error rate (r = .321, p = .043), controlling for other variables. Whole-brain analyses revealed that participants with higher real-time MA showed increased brain activity within the medial superior frontal gyrus (mSFG) and left middle frontal gyrus (MFG; Fig 2A). ROI analyses identified 6 brain regions as stably contributing to the cross-validated LASSO model: the right intraparietal sulcus (IPS) from the arithmetic network, the subcallosal cortex (SCC) from the anxiety network, and the left angular gyrus (AG), mSFG, right SFG, and right crus II from the DMN (Fig 2B). All these predictors showed positive relative beta coefficients, except for IPS being negative (Fig 2C). Critically, interactive effects between relative task error rate and MA group on relative brain activity were observed in the SCC (p = .016) and right SFG (p = .005) after accounting for the control variables (Fig 2D). Particularly, marginally significant connections between brain and behavior were observed in the low-level MA group (p = .080 and .081, respectively) but not in the high-level MA group (p > .10).

This study advances our understanding of the neural correlates of real-time MA, highlighting its impact on task performance and brain activity within relevant networks. The whole-brain analysis and LASSO feature selection together demonstrated real-time MA to be associated with heightened activity within the DMN and prefrontal cortex, as well as reduced activity in right IPS. Notably, real-time MA also moderated the connection between brain activation and task performance during math problem-solving. These findings provide valuable perspectives for strategies targeting MA and potential applications in educational contexts.

Emotion and Motivation Other ¹

Reasoning and Problem Solving ²

Activation (eg. BOLD task-fMRI)

Classification and Predictive Modeling

ADULTS

Anxiety

Emotions

FUNCTIONAL MRI

Other - Math Anxiety, Arithmetic, Default Mode Network