Spectral Entropy Can Predict Changes of Working Memory Performance Reduced by Short-Time Training in the Delayed-Match-to-Sample Task

Yin Tian, Huiling Zhang, Wei Xu, Haiyong Zhang, Yang Li, Shuxing Zheng, Yupan ShiFrontiers in Human Neuroscience 2017 · DOI

Abstract

Spectral entropy, which was generated by applying the Shannon entropy concept to the power distribution of the Fourier-transformed electroencephalograph (EEG), was utilized to measure the uniformity of power spectral density underlying EEG when subjects performed the working memory tasks twice, i.e., before and after training. According to Signed Residual Time (SRT) scores based on response speed and accuracy trade-off, 20 subjects were divided into two groups, namely high-performance and low-performance groups, to undertake working memory (WM) tasks. We found that spectral entropy derived from the retention period of WM on channel FC4 exhibited a high correlation with SRT scores. To this end, spectral entropy was used in support vector machine classifier with linear kernel to differentiate these two groups. Receiver operating characteristics analysis and leave-one out cross-validation (LOOCV) demonstrated that the averaged classification accuracy (CA) was 90.0 and 92.5% for intra-session and inter-session, respectively, indicating that spectral entropy could be used to distinguish these two different WM performance groups successfully. Furthermore, the support vector regression prediction model with radial basis function kernel and the root-mean-square error of prediction revealed that spectral entropy could be utilized to predict SRT scores on individual WM performance. After testing the changes in SRT scores and spectral entropy for each subject by short-time training, we found that 16 in 20 subjects' SRT scores were clearly promoted after training and 15 in 20 subjects' SRT scores showed consistent changes with spectral entropy before and after training. The findings revealed that spectral entropy could be a promising indicator to predict individual's WM changes by training and further provide a novel application about WM for brain-computer interfaces.

Introduction

Working memory (WM) serves as a critical "mental workspace" for the temporary maintenance and manipulation of information. Despite its fundamental role in cognitive function, WM capacity varies significantly across individuals and is susceptible to enhancement through targeted training. Identifying a robust, objective biomarker for WM ability remains a significant challenge in neuroscience. While traditional EEG frequency band analysis (e.g., alpha or theta power) provides some insight, these measures often lack the inter-subject consistency required for universal prediction.

This study proposes spectral entropy (SE) as a normalized, robust alternative. By applying Shannon entropy to the power distribution of EEG signals, SE quantifies the complexity and uniformity of neural activity. The researchers hypothesized that SE could not only distinguish between performance levels but also track cognitive improvements resulting from short-term training.

Methodology

The study involved 20 healthy, right-handed male subjects who participated in two experimental sessions: one before and one after a short-term WM training period.

Experimental Paradigm

Participants performed a Delayed-Match-to-Sample (DMTS) task. In each trial, subjects were presented with a memory array of letters (load sizes of 2, 4, or 8), followed by a 3000ms retention period. Finally, a probe letter was shown, and subjects indicated whether it was part of the original array.

The Delayed-Match-to-Sample (DMTS) task paradigm used to assess working memory performance across encoding, retention, and retrieval phases.

Feature Extraction and Performance Metrics

EEG data were recorded from 64 channels. The primary feature, Spectral Entropy, was calculated specifically during the 3000ms retention period. Behavioral performance was quantified using Signed Residual Time (SRT), a metric that accounts for the speed-accuracy trade-off, thereby providing a more nuanced measure of cognitive efficiency than raw accuracy or reaction time alone.

Modeling and Validation

The researchers employed a multi-stage analysis pipeline to correlate brain activity with performance:

  1. Correlation Analysis: Spearman's correlations identified the EEG channels most strongly associated with SRT scores.
  2. Classification: A Support Vector Machine (SVM) categorized subjects into high- and low-performance groups.
  3. Regression: Support Vector Regression (SVR) predicted individual SRT scores.
  4. Validation: Leave-One-Out Cross-Validation (LOOCV) ensured the generalizability of the results.
Methodological framework for calculating Spearman's correlations between spectral entropy and performance across 64 channels, resulting in topographic and fingerprint maps.

Results and Discussion

Identifying the Neural Hub

The analysis revealed that the strongest correlation between SE and WM performance occurred at channel FC4, located over the right frontal cortex. This finding aligns with existing literature identifying the right prefrontal cortex as a central hub for executive control and working memory maintenance.

Predictive Performance

The SVM model achieved high classification accuracy, reaching 90.0% for intra-session data and 92.5% for inter-session prediction. Furthermore, the SVR model demonstrated a strong correlation between predicted and actual SRT scores ($r = 0.814$ in session one; $r = 0.794$ in session two), suggesting that SE is a highly reliable predictor of individual performance.

Training and Plasticity

The study confirmed the efficacy of short-term training, with 16 out of 20 subjects showing improved SRT scores. Crucially, in 15 of these subjects, the change in SE mirrored the behavioral improvement. This suggests that higher spectral entropy during the retention phase—indicating a more complex or flexible neural state—is a hallmark of enhanced WM performance.

Conclusion

Spectral entropy represents a promising biomarker for monitoring and predicting working memory performance. Its ability to track training-induced changes suggests significant potential for developing adaptive Brain-Computer Interfaces (BCIs) and clinical tools for assessing cognitive impairments. Future research should expand the participant demographic and focus on real-time implementation for clinical applications in conditions such as ADHD or depression.

Summary updated

Abstract

Spectral entropy, which was generated by applying the Shannon entropy concept to the power distribution of the Fourier-transformed electroencephalograph (EEG), was utilized to measure the uniformity of power spectral density underlying EEG when subjects performed the working memory tasks twice, i.e., before and after training. According to Signed Residual Time (SRT) scores based on response speed and accuracy trade-off, 20 subjects were divided into two groups, namely high-performance and low-performance groups, to undertake working memory (WM) tasks. We found that spectral entropy derived from the retention period of WM on channel FC4 exhibited a high correlation with SRT scores. To this end, spectral entropy was used in support vector machine classifier with linear kernel to differentiate these two groups. Receiver operating characteristics analysis and leave-one out cross-validation (LOOCV) demonstrated that the averaged classification accuracy (CA) was 90.0 and 92.5% for intra-session and inter-session, respectively, indicating that spectral entropy could be used to distinguish these two different WM performance groups successfully. Furthermore, the support vector regression prediction model with radial basis function kernel and the root-mean-square error of prediction revealed that spectral entropy could be utilized to predict SRT scores on individual WM performance. After testing the changes in SRT scores and spectral entropy for each subject by short-time training, we found that 16 in 20 subjects' SRT scores were clearly promoted after training and 15 in 20 subjects' SRT scores showed consistent changes with spectral entropy before and after training. The findings revealed that spectral entropy could be a promising indicator to predict individual's WM changes by training and further provide a novel application about WM for brain-computer interfaces.
BibTeX
@inproceedings{tian2017spectral,
  author = {Yin Tian and Huiling Zhang and Wei Xu and Haiyong Zhang and Yang Li and Shuxing Zheng and Yupan Shi},
  title = {{Spectral Entropy Can Predict Changes of Working Memory Performance Reduced by Short-Time Training in the Delayed-Match-to-Sample Task}},
  booktitle = {Frontiers in Human Neuroscience},
  year = {2017},
  doi = {10.3389/fnhum.2017.00437},
}
Figure