Using EEG spectral components to assess algorithms for detecting fatigue

https://doi.org/10.1016/j.eswa.2007.12.043Get rights and content

Abstract

Fatigue is a constant occupational hazard for drivers and it greatly reduces efficiency and performance when one persists in continuing the current activity. Studies have investigated various physiological associations with fatigue to try to identify fatigue indicators. The current study assessed the four electroencephalography (EEG) activities, delta (δ), theta (θ), alpha (α) and beta (β), during a monotonous driving session in 52 subjects (36 males and 16 females). Performance of four algorithms, which were: algorithm (i) (θ + α)/β, algorithm (ii) α/β, algorithm (iii) (θ + α)/(α + β), and algorithm (iv) θ/β, were also assessed as possible indicators for fatigue detection. Results showed stable delta and theta activities over time, a slight decrease of alpha activity, and a significant decrease of beta activity (p < 0.05). All four algorithms showed an increase in the ratio of slow wave to fast wave EEG activities over time. Algorithm (i) (θ + α)/β showed a larger increase. The results have implications for detecting fatigue.

Impact on industry: The results of this research have the implication for detecting fatigue and can be used for future development of fatigue countermeasure devices.

Introduction

Fatigue is a constant occupational hazard for any long-distance or professional driver, and can affect one’s judgement of his or her suitability to continue driving (Brown, 1997). Efficiency and performance can be impaired during fatigue when an individual persists in continuing the current activity as normal (Brown, 1994). Lamond and Dawson (1999) reported that a driver who has remained without sleep for 24 h has reduced driving skills, and is comparable to driving with illegally high blood alcohol concentration of 0.10%. Fatigue is independent of energy consumption and cannot simply be measured by performance impairment (Brown, 1994). Hence, the need for physiological fatigue countermeasures arises to prevent fatigue related accidents.

In the recent years, researchers have investigated different types of fatigue countermeasure technologies, which include development of electroencephalography (EEG) algorithms to detect fatigue (Eoh et al., 2005, Lal et al., 2003), facial movement and feature detectors (Gu, Ji, & Zhu, 2002), and PERCLOS, which detects the percentage of eye closure (Wierwille, Ellsworth, Wreggit, Fairbanks, & Kim, 1994). However, Artaud et al. (1994) found that EEG is one of the most reliable indicators of fatigue, and hence, seems to be a promising fatigue countermeasure approach (Lal et al., 2003).

Fatigue countermeasure devices ought to have high reliability standard. Brown (1997) argued that factors other than fatigue also influenced the changes in driving performance, such as vehicle steering. A number of false alarms may occur if the countermeasure device has small reliability level. Therefore, reliability is one of the most important factors of future fatigue countermeasure devices (Brown, 1997). EEG has been shown to have a good test and retest reliability and high reproducibility for the delta and theta bands (Lal and Craig, 2005, Pollock et al., 1991), as well as the alpha activity (Gasser et al., 1985, Tomarken et al., 1992).

Four frequency components can be obtained from EEG recordings, which are delta (δ) (±0 to 4 Hz), theta (θ) (4–8 Hz), alpha (α) (8–13 Hz), and beta (β) (13–20 Hz), and these can be measured to detect the current state of a driver (Åkerstedt, Kecklund, & Knutsson, 1991). Delta activity is high during sleep. Early stage of drowsiness can be indicated by an increase in theta activity (Åkerstedt & Gillberg, 1990). Alpha activity reflects a relaxed wakefulness state, and decreases with concentration, stimulation or visual fixation (Stern & Engel, 2005). However, other researchers have also found an increase in alpha activity in train drivers who were sleepy enough to fall asleep while driving (Åkerstedt and Gillberg, 1990, Torsvall and Åkerstedt, 1987). Furthermore, increased beta activity has also been related to the alertness level, and decreases during drowsiness (Eoh et al., 2005). Torsvall and Åkerstedt (1987) believed that alpha activity was the most sensitive measure that could be used in detecting fatigue, followed by theta and delta activities. However, delta activity is more related to occurrence of sleep proper (Torsvall & Åkerstedt, 1987). Lal and Craig (2002) have also shown changes in brain wave activity with fatigue during driving.

A number of methods for fatigue detection using EEG have been proposed, such as detection of alpha spindles by Tietze (2000), and an algorithm that utilises the combination of all frequency components of EEG to signify level of alertness by Lal et al. (2003). Other studies have proposed two algorithms, which were (θ + α)/β and β/α, that can be used as a fatigue detection technique (Brookhuis and Waard, 1993, Eoh et al., 2005). Delta activity was excluded from and not investigated in the study by Eoh et al. (2005) since it reflected the sleeping state of a person, and was not expected to show high activity during the driving activity. Eoh et al. (2005) believed that the first algorithm, (θ + α)/β, was a more reliable fatigue indicator since it showed a clear indication of increasing fatigue as the ratio between the slow wave and fast wave activities increased.

The current study investigated the performance of different algorithms, which had the potential to function as a fatigue indicator. The two algorithms studied by Eoh et al. (2005), (θ + α)/β and β/α, were compared with another two new proposed algorithms, (θ + α)/(α + β) and θ/β. However, since the current study intended to investigate the ratio between slow and fast wave activities over time, hence, the second algorithm by Eoh et al. (2005), β/α, was denoted as α/β. The four frequency components in the EEG recording, delta, theta, alpha and beta, were also investigated in the analysis to understand each EEG band separately.

Section snippets

Materials and methods

Fifty-two non-professional drivers (36 males and 16 females), aged 20–70 years (mean: 28 ± 10 years), were recruited to perform a monotonous driving simulator task. The average body mass index (BMI) was 23 ± 7 kg/m2 (normal range: 18.50–24.99 kg/m2 (World Health Organization, 2007)). All participants provided informed consent prior to participating in the study. Lifestyle appraisal questionnaire was used as a selection criteria, which required participants to have no medical contraindications such as

Results

The total average time for the driving session was 63 min ± 12 min. From the study by Gillberg, Kecklund, and Åkerstedt (1996), 30 min of monotonous driving activity has been found to induce fatigue during driving. The average pre-study systolic blood pressure (SBP) was 118 ± 11 mmHg and diastolic blood pressure (DBP) was 75 ± 9 mmHg. The average post-study SBP was 114 ± 13 mmHg and DBP was 71 ± 17 mmHg. The average pre-study heart rate was 72 ± 10 beats/min and 65 ± 9 beats/min for the post-study heart rate. Student

Discussion

Driver sleepiness is one of the main factors associated with road crash accidents (Horne and Reyner, 1995, Horne and Reyner, 1999). Several factors have been studied to prevent fatigue or sleepiness during driving. For example, Reyner and Horne (2002) studied the effect of caffeine as fatigue countermeasure and the efficacy of caffeine in counteracting driver sleepiness. These authors found that caffeine was effective in reducing sleep-related vehicle accidents. Others studied technological

Conclusion

This study has investigated the four EEG frequency bands, delta, theta, alpha and beta, and four algorithms (algorithm (i) (θ + α)/β, algorithm (ii) α/β, algorithm (iii) (θ + α)/(α + β), and algorithm (iv) θ/β) to assess fatigue. Some significant differences were detected from the alert baseline over time, which were mostly in the temporal site, except for the delta and theta activities, which were also different in the central, frontal, parietal sites, as well as the entire brain. This study has

Acknowledgements

The research was supported by an Australian Research Council (ARC) Linkage Grant (LP0560886) and by SENSATION Integrated Project (FP6-507231) co-funded by the Sixth Framework Programme of the European Commission under the Information Society Technologies priority. National Health and Medical Research Council (NHMRC) Equipment grant and NHMRC Training Fellowship (#169309) are also acknowledged. We also acknowledge Dr. Saskia Waters, Anna Kriaris, and Laarnie Pe Benito, who assisted in some

References (33)

  • L. Torsvall et al.

    Sleepiness on the job: Continuously measured EEG changes in train drivers

    Electroencephalography and Clinical Neurophysiology

    (1987)
  • A. Vuckovic et al.

    Automatic recognition of alertness and drowsiness from EEG by an artificial neural network

    Medical Engineering & Physics

    (2002)
  • T. Åkerstedt et al.

    Subjective and objective sleepiness in the active individual

    International Journal of Neuroscience

    (1990)
  • T. Åkerstedt et al.

    Manifest sleepiness and the spectral content of the EEG during shift work

    Sleep

    (1991)
  • Artaud, P., Planque, S., Lavergne, C., Cara, H., de Lepine, P., Tarriere, C., et al. (1994). An on-board system for...
  • K.A. Brookhuis et al.

    The use of psychophysiology to assess driver status

    Ergonomics

    (1993)
  • Cited by (545)

    • BCI-AMSH: A MATLAB based open-source brain computer interface assistive application for mental stress healing

      2023, e-Prime - Advances in Electrical Engineering, Electronics and Energy
    View all citing articles on Scopus
    View full text