Real-Time Assessment of Mental Workload with Near-Infrared Spectroscopy: Potential for Human-Computer Interaction

Fantini, Sergio; Sassaroli, Angelo; Tong, Yunjie; Hirshfield, Leanne M.; Girouard, Audrey; Solovey, Erin Treacy; Jacob, Robert J. K. · 2008 · Crossref

DOI: 10.1364/biomed.2008.bmd14

archive: archived pipeline: cataloged verified

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

Summary

This study addresses the challenge of collecting quantitative neurophysiological data to assess human cognitive workload in real-time, a critical requirement for adaptive human-computer interaction (HCI). While electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) have been proposed for this purpose, the authors aim to develop a tool that provides continuous feedback to dynamic interfaces by applying machine learning algorithms to fNIRS data. The research specifically investigates whether these algorithms can distinguish between different levels of mental workload based on hemodynamic responses recorded from the forehead. The experimental protocol involved five subjects performing tasks with varying cognitive demands, defined by the number of colors on a rotating physical cube or a computer interface. Workload levels ranged from zero (rest/no cube) to four colors, with each workload cycle lasting approximately 45 seconds followed by 40 seconds of rest. Data were collected using an OxiplexTS fNIRS system with laser sources at 690 nm and 830 nm, and source-detector distances of 1.5, 2.0, 2.5, and 3.0 cm. The raw intensity data were processed using the modified Beer-Lambert law to calculate changes in oxygenated hemoglobin ([HbO]) and deoxygenated hemoglobin ([Hb]). The resulting time series were filtered and analyzed using two machine learning classification algorithms: Dynamic Time Warping (DTW) and Symbolic Aggregate Approximation (SAX). The physiological results showed clear activations across all source-detector distances. Notably, the trends for Δ[HbO] and Δ[Hb] were inverted compared to typical motor task responses, a phenomenon previously reported in literature for mental tasks. The classification results demonstrated the efficacy of the machine learning approach. Using DTW, the system achieved classification accuracies ranging from 77.8% to 94.4% across the five subjects for distinguishing workload levels 0, 2, and 4. Similarly, the SAX algorithm yielded accuracies between 72.2% and 94.4%. The optimal channels and parameters varied by subject, indicating individual differences in hemodynamic response patterns. The study concludes that machine learning algorithms applied to fNIRS data hold significant potential for distinguishing different levels of mental workload. This capability represents a foundational step toward developing interactive human-computer interfaces that can adapt to user cognitive states. The authors suggest that future work should focus on systematically selecting the most informative data instances and channels to improve classification efficiency and robustness, thereby enhancing the practical applicability of real-time workload assessment in HCI systems.

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 qwen3.6-27b-nvidia on 2026-08-10; verification: verified.

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