نهمین کنگره نقشه برداری مغز

نمایش مقاله

Title:

Identifying EEG Signatures of Mental Arithmetic Performance Using Machine Learning

Subject:

Artificial Intelligence and Brain Mapping

Presenter Author:
Name:	Maedeh Sadat Tabatabaei Kamal

Author:
Name 1:	Maedeh Sadat Tabatabaei Kamal - Corresponding - Presenter
Affiliation 1:	University of social welfare and rehabilitation sciences

Name 2:
Affiliation 2:

Name 3:
Affiliation 3:

Name 4:
Affiliation 4:

Name 5:
Affiliation 5:

Name 6:
Affiliation 6:

Name 7:
Affiliation 7:

Name 8:
Affiliation 8:

Name 9:
Affiliation 9:

Name 10:
Affiliation 10:

Abstract

Original Research

introduction :

Mental arithmetic engages frontal-parietal networks, but which EEG signatures distinguish high from low performers remain unclear. We identified electroencephalographic signatures characterizing performance differences using machine learning classification.

Methods:

We analyzed 36 healthy participants (age 18-26 years) from PhysioNet EEGMAT performing serial subtraction: good performers (n=24, mean=21) and poor performers (n=12, mean=7). EEG was recorded from 23 channels during 60-second task and 180-second rest. Features included spectral power (delta, theta, alpha, beta, gamma), frontal-parietal coherence, and regional activity. Random Forest and Support Vector Machine classifiers were trained with 5-fold cross-validation; t-tests compared groups.

Results:

Univariate t-tests revealed no significant between-group differences in any frequency band (all p>0.05). However, Random Forest achieved 72.7% accuracy (cross-validation: mean=72.0%, SD=9.8%), significantly exceeding chance (50%). Feature importance analysis identified occipital delta (importance=0.040) and frontal-parietal coherence (importance=0.038) as top predictors. Seven of the top ten features originated from occipital, temporal, and central regions. Machine learning captured signatures not detected by univariate analyses and revealed unexpected patterns in posterior brain regions.

Conclusion:

Our findings highlight a discrepancy between univariate tests and multivariate classification. Importantly, feature importance reflects model contributions and does not imply neural causality. Within this classification framework, occipital delta showed unexpectedly high importance, possibly reflecting individual differences in cognitive strategies. Specifically, some individuals might employ visual imagery, such as visualizing number lines, thereby engaging occipital regions. This interpretation aligns with evidence that imagery and perception share occipital-frontoparietal mechanisms. Furthermore, delta oscillations, associated with attentional control, might index visual attention deployment during arithmetic tasks. More broadly, these results demonstrate that multivariate approaches can detect distributed neural interactions not captured by univariate tests. From an educational perspective, our findings suggest that visualization strategies may relate to arithmetic performance. If confirmed in future studies, incorporating visualization techniques into instruction might benefit learners, particularly those with strong visual processing abilities. Several limitations warrant mention. As a secondary analysis of publicly available data, we could not collect supplementary cognitive measures such as strategy self-reports. Additionally, findings should be validated in independent datasets to establish generalizability. In summary, machine learning successfully identified EEG signatures not evident in univariate analyses. The model's unexpected reliance on occipital features raises important questions about individual differences in mathematical cognition and the role of visual processing in mental arithmetic.

Keywords:

Brain mapping, Electroencephalography, Mental arithmetic, Machine learning, Cognitive performance