Abstract
Purpose: :
Fatigue is a critical factor in performance. Many different physiological parameters have been used to try and determine if a person is alert enough to perform satisfactorily or not. An increase in pupillary oscillations has been used to identify fatigue but the pupillary unrest index clinically used requires 11 minutes of recording. We wished to determine if there were characteristic changes in wavelet entropy that would enable detection of fatigue more rapidly.
Methods: :
14 subjects were tested with binocular infra-red video-oculography at 240 Hz. 58 recordings were performed. 33 were performed on days when subjects reported having had a normal sleep pattern and on also done on 26 days when subjects reported little or no sleep the night before. Some were performed in the mornings and others in the late afternoon. We recorded for 90 seconds in a dimly lit room with a 2 second light stimulus at the initial portion of the recording. A db7 wavelet was used at 13 scales to generate multiple time series from which entropy was calculated. Entropy was evaluated in windows of varying size from 20-90 seconds. A Bayesian classifier was used to distinguish fatigued from normal tracings.
Results: :
All subjects showed a significant increase in power in the lower frequency bands during the 90 seconds of recording. The differences between the fatigued and normal tracings were most prominent when looking at smaller window sizes. The higher frequency (0.16-0.33 Hz) content was best able to distinguish between normal and fatigue between 45-90 seconds while lower (0.02-0.08) frequency content played a more significant role in classification between 0 and 45 seconds. Overall, we were able to classify with a overall accuracy of 70%. If studies done in the morning or afternoon were considered as separate groups, identification within each group increased to over 85%.
Conclusions: :
Wavelet entropy is an important new means to evaluate and compare different time series. Even without the benefit of a baseline recording for comparison, we were able to correctly categorize over 90% of the normal recordings and overall classify 70% of the recordings as either fatigued or nonfatigued independent of the time of the recording. Grouping the morning and afternoon recordings separately enabled us to classify over 85% correctly. A larger number of tracings is required to confirm the accuracy of the paradigm.
Keywords: pupil • neuro-ophthalmology: diagnosis