An MRI Study on Effects of Math Education on Brain Development Using Multi-Instance Contrastive Learning

Yupei Zhang, Shuhui Liu, Xuequn ShangFrontiers in Psychology 2021 · DOI

Abstract

This study investigates the impact of mathematical education on brain development through the analysis of brain MRIs. While previous research focused on biochemical changes in the left middle frontal gyrus, this work proposes classifying students based on structural data from the intraparietal sulcus (IPS). We introduce Multi-Instance Contrastive Learning (MiCL) to address the challenges of multi-instance representation learning in 3D MRI volumes. By extracting 2D slices from cropped IPS regions, the model learns robust representations that are subsequently used by a linear neural network to distinguish between students who continued math education and those who did not. Experimental results on 123 adolescent subjects demonstrate a classification accuracy of 90.24%, representing a significant improvement over standard contrastive learning frameworks.

Introduction

Neuroplasticity—the brain's ability to reorganize itself in response to learning—is a well-documented phenomenon. However, identifying the specific structural "signatures" left by particular academic subjects, such as mathematics, remains a complex challenge. Previous studies have linked a lack of math education to lower concentrations of the neurotransmitter GABA in the middle frontal gyrus (MFG). This research shifts the focus to the intraparietal sulcus (IPS), a region critical for numerical processing and a key component of the frontoparietal network.

From a machine learning perspective, analyzing MRI data presents a Multi-Instance Learning (MIL) problem. A single MRI scan is a 3D volume, which can be decomposed into multiple 2D "slices" (instances). The goal is to extract a unified representation from these multiple instances to characterize the subject's brain structure accurately.

Methodology: Multi-Instance Contrastive Learning (MiCL)

The proposed MiCL framework extends standard contrastive learning (such as SimCLR) to a multi-instance setting to capture the spatial relationships across the 3D brain region.

Data Acquisition and Preprocessing

The study utilized T1-weighted MRIs from 123 UK adolescents (aged 16), divided into a "math group" (n=72, A-level math students) and a "non-math group" (n=51, students who ceased math education). The researchers focused on the IPS, cropping 3D "volumes of interest" (VOIs) and extracting 20 representative 2D slices per student.

Figure 1: Positions of the Volume of Interest (VOI) in the intraparietal sulcus (IPS) across sagittal (A), coronal (B), and axial (C) MRI slices.

The MiCL Pipeline

  1. Data Augmentation: Slices undergo random cropping, resizing, and Gaussian blurring to improve model robustness.
  2. Feature Extraction: A ResNet backbone, with weights shared across all 20 slices, extracts latent features from each instance.
  3. Feature Fusion: These 20 individual representations are concatenated into a single high-dimensional vector, creating a comprehensive "bag" representation for the student.
  4. Contrastive Loss: The model is trained to minimize the distance between two differently augmented versions of the same student's collective slices while maximizing the distance between different students.
  5. Classification: A single-layer neural network uses the learned representations to predict the student's educational group.

Results

The MiCL model significantly outperformed the baseline SimCLR framework across all major metrics:

  • Accuracy: 90.24% (a >5% improvement over standard CL).
  • AUC (Area Under Curve): 0.9578, demonstrating high discriminative power.
  • Recall: 97.56% for the math group, indicating the model is highly effective at identifying students who continued math education.
  • Visualization: t-SNE plots confirmed that MiCL produces tighter, more distinct clusters for math vs. non-math students compared to traditional methods.

Discussion and Conclusion

The high classification accuracy validates the IPS as a primary site for structural changes driven by mathematical study. The success of the MiCL approach highlights the importance of joint learning; by processing slices collectively rather than independently, the model captures cumulative structural patterns across the 3D region.

While the results are promising, the study is limited by its sample size (123 subjects) and its cross-sectional nature. Future work should involve larger cohorts and longitudinal designs to establish causality—determining whether math education causes these structural changes or if pre-existing structures influence educational choices. Additionally, incorporating other brain regions like the MFG and employing attention mechanisms could further refine the model's predictive capabilities.

Summary updated

Abstract

This paper explores whether mathematical education has effects on brain development from the perspective of brain MRIs. While biochemical changes in the left middle front gyrus region of the brain have been investigated, we proposed to classify students by using MRIs from the intraparietal sulcus (IPS) region that was left untouched in the previous study. On the cropped IPS regions, the proposed model developed popular contrastive learning (CL) to solve the problem of multi-instance representation learning. The resulted data representations were then fed into a linear neural network to identify whether students were in the math group or the non-math group. Experiments were conducted on 123 adolescent students, including 72 math students and 51 non-math students. The proposed model achieved an accuracy of 90.24 % for student classification, gaining more than 5% improvements compared to the classical CL frame. Our study provides not only a multi-instance extension to CL and but also an MRI insight into the impact of mathematical studying on brain development.
BibTeX
@inproceedings{zhang2021mri,
  author = {Yupei Zhang and Shuhui Liu and Xuequn Shang},
  title = {{An MRI Study on Effects of Math Education on Brain Development Using Multi-Instance Contrastive Learning}},
  booktitle = {Frontiers in Psychology},
  year = {2021},
  doi = {10.3389/fpsyg.2021.765754},
}
Figure