Unraveling the neural dynamics of mathematical interference in english reading: A novel approach with deep learning and fNIRS data

• Published in: Brain research bulletin Vol. 227; p. 111398
• Main Authors: Liang, Zhijie, Wang, Ling, Su, Jianyu, Sun, Bo, Wang, Daifa, Yang, Juan
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01.07.2025; Elsevier
• Subjects: Adult; Brain - diagnostic imaging; Brain - physiology; Cognition - physiology; Data Mining; Deep Learning; English Language Learners; Female; FNIRS; Humans; Language; Learning - physiology; Male; Mathematical Concepts; Mathematical Interference; Reading; Spectroscopy, Near-Infrared - methods; Young Adult; Data Mining; FNIRS; English Language Learners; Deep Learning; Mathematical Interference
• ISSN: 0361-9230; 1873-2747; 1873-2747
• DOI: 10.1016/j.brainresbull.2025.111398

English has emerged as the predominant global language, driving efforts to optimize its acquisition through interdisciplinary cognitive research. While behavioral studies suggest a link between English learning and mathematical cognition, the neural mechanisms underlying this relationship remain poorly understood. To bridge this gap, the present study employs functional near-infrared spectroscopy (fNIRS) to construct a novel dataset on mathematical interference in English acquisition. Utilizing this dataset, a novel deep learning model named AC-LSTM is proposed, amalgamating Transformer and LSTM architectures to identify residual mathematical cognition during the English learning process. The AC-LSTM model achieves an exceptional accuracy rate of 99.8 %, surpassing other machine learning and deep learning models. Moreover, a multi-class classification experiment is conducted to discern algebra, geometry, and quantitative reasoning interference, with the AC-LSTM model achieving the highest accuracy of 75.9 % in this classification task. Furthermore, crucial brain channels for interference detection are pinpointed through grid search, and alterations in vital brain regions (R-Broca and L-Broca) are unveiled via association rule analysis. By integrating fNIRS, deep learning, and data mining techniques, this study delves into cognitive interference in English learning, providing valuable insights for educational neuroscience and data mining research. •Large-Scale Neural Dynamics Dataset: Unveils brain activity patterns during English reading with math interference.•AI Decodes Brain Science: Harnesses AI to explain the neuroscience of English learning from a biomedical perspective.•99.8 % Accuracy in Detection: Our AC-LSTM model achieves exceptional accuracy in identifying math-related cognitive residues.•Brain Channel Identification: Grid search pinpoints key brain channels linked to mathematical interference in learning.•Classifying Math Interference: Model differentiates algebra, geometry impacts on English with 75.9 % accuracy.