A Predictive Model of Reading Comprehension Based on MEG Imaginary Coherence Functional Connections

Abstract

Purposes Reading comprehension is one of the most important cognitive abilities of human beings. Objective indicators should be provided in order to evaluate human reading comprehension ability. Methods A prediction model based on magnetoencephalogram (MEG) imaginary coherent brain functional connections is proposed in this paper. The imaginary coherence algorithm is used to construct the whole brain MEG functional connections, and the features are selected by univariate feature selection algorithm. The reading comprehension ability is predicted with partial least squares (PLS) prediction model. Findings The partial least squares regression model based on MEG imaginary coherent functional connections can successfully predict reading comprehension scores. For univariate feature selection model forecasting, higher performance and accuracy can be obtained (R2=0.524[PVT-Language], MSE[PVT-Language]=5.042; R2 [ORRT-Language]=0.536, MSE[ORRT-Language]=5.142). The task state data set related to reading comprehension is more suitable than the resting state data set to predict reading comprehension ability, and the functional connection of feature selection is more accurate. Conclusions The PLS prediction model based on MEG virtual coherence functional connection can be used to objectively evaluate human reading comprehension.

Keywords

• reading comprehension ability
• prediction model
• pls
• ts-meg
• univariate feature selection
• meg