Optimized EEG–fNIRS Based Mental Workload Detection Method for Practical Applications

Chu, Hongzuo; Cao, Yong; Jiang, Jin; Yang, Jiehong; Huang, Mengyin; Li, Qijie; Jiang, Changhua; Jiao, Xuejun · 2021 · Crossref

DOI: 10.21203/rs.3.rs-683529/v1

archive: archived pipeline: cataloged

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the limitations of existing EEG–fNIRS-based mental workload detection methods, which often suffer from complex signal acquisition configurations and suboptimal detection accuracy, hindering their practical application in high-stakes environments like military and aviation. The research aims to optimize the signal acquisition configuration to create a more accurate and convenient detection method for complex man–machine systems. The study employed a Multi-Attribute Task Battery (MATB) task to simulate realistic operator cognitive demands, involving 20 male volunteers from the China Astronaut Research and Training Center. Participants performed four blocks of tasks at four difficulty levels (1, 3, 5, and 7), with each block lasting three minutes. Data collection included 64-channel EEG signals (sampled at 500 Hz) and two-channel frontal fNIRS signals (sampled at 50 Hz), alongside subjective NASA-TLX scale ratings and task performance metrics. EEG preprocessing involved re-referencing, band-pass filtering, and Independent Component Analysis to remove artifacts, followed by extraction of Power Spectral Density (PSD) in theta, alpha, beta1, and beta2 bands. fNIRS data were processed to extract statistical features of oxygenated and deoxygenated hemoglobin concentrations. Channel selection was performed by ranking channel importance, leading to the optimization of the EEG setup from 64 to 26 channels. Classification models using Support Vector Machine (SVM), Random Forest (RF), and Decision Tree (DT) were trained on EEG-only, fNIRS-only, and fused EEG-fNIRS feature sets. Results indicated that subjective workload scores increased while task performance decreased as difficulty rose. Physiological analysis confirmed that theta power in the prefrontal region and beta1/beta2 power in the occipital region increased with task difficulty, while alpha power decreased. Notably, the study observed for the first time that EEG band energy in the occipital lobe varied significantly with load. In terms of classification accuracy, the fused EEG-fNIRS feature set outperformed unimodal sets. The Random Forest classifier achieved the highest accuracy with the fused features, reaching a mean four-level detection accuracy of 78.25% ± 4.71%, compared to 55.47% for EEG-only and 69.20% for fNIRS-only. The fused model also demonstrated lower standard deviation, indicating greater robustness. The significance of this work lies in the establishment of an optimized configuration comprising 26 EEG channels and two frontal fNIRS channels, which balances portability with high detection accuracy. This approach overcomes the spatial resolution limitations of EEG and the temporal resolution limitations of fNIRS through feature fusion. The findings support the deployment of such multimodal systems in real-time monitoring for military, driving, and other complex human-computer interaction scenarios, where accurate and stable mental workload assessment is critical for safety and performance.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.8-27b-gittensor on 2026-08-23; verification: pending re-verification.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• neuro workload indices
• workload measurement

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: physiological data