Thirty-five young, healthy subjects (14 males, 21 females), aged 23.3 ± 3.6 years (mean ± standard deviation), participated in our study that extended from November 2014 to June 2015. Our study protocol—approved by the Ethics Committee of the University of Liège—led each subject to perform three 10-minutes PVTs over two consecutive days, under conditions of increasing sleep deprivation conditions induced by acute, prolonged waking.
The protocol required subjects without any alcohol dependencies, drug addictions, or sleep disorders. Each was asked to maintain a normal sleep pattern for the week prior to taking the first PVT, and to have a full night sleep (of 7–8 hours at least) just before this PVT. Each was also asked to maintain a sleep diary during that week, to allow us to verify that the sleep requirements were met. Once a subject took the first PVT, he/she was not allowed to sleep until after the third and last PVT, thereby inducing a total sleep deprivation of 28–30 hours. We organized several sessions of three PVTs, each with a few subjects (typically 2–3 subjects) successively taking each PVT.
The details of the tests follow. Day 1. At 8:30 (in 24 hour time), the scheduled subjects arrived at the laboratory, and were equipped with the PSG electrodes. Between 10:00 and 11:00, they (successively) carried out the first PVT, called PVT1. Afterwards, they were equipped with wrist actigraphs to verify that they would not sleep, and were allowed to leave the laboratory. From 12:00 on, they were not allowed to consume any coffee, tea, energy drinks, or other stimulants. At 20:30, they returned to the laboratory, and were equipped with the PSG electrodes. They stayed overnight in the laboratory, and until the end of the tests. During the night, they were allowed to use multimedia devices, to play card and board games, to interact with the laboratory staff, and to consume the soft drinks and biscuits that we provided. Day 2. Between 3:30 and 4:00, they carried out the second PVT, called PVT2. At 8:30, we provided breakfast. Between 12:00 and 12:30, they
Chapter 3. Sleep-deprivation dataset 29
Subject free Subject at the lab.
PVT1 PVT2 PVT3
No stimulant
Normal sleep Sleep deprivation
Subject free + actigraph Subject at the lab.
7:00 8:30 10:00 11:00 12:00 20:30 3:30 4:00 12:00 12:30
DAY 1 DAY 2
Figure 3.1 – Pictorial summary of the data acquisition schedule.
carried out the third PVT, called PVT3. This concluded the tests. We strongly advised the participants not to drive home by themselves, and we offered alternative transportation
solutions when necessary. Figure3.1depicts the data acquisition schedule for each subject.
The PVTs were all performed in a quiet, isolated laboratory environment without any temporal cues (e.g., watch or smartphone). The room lights were turned off for PVT2 and PVT3. For a 15-minute period before each PVT, we instructed the subjects to part with their phones, computers, and any other screen devices. At the beginning of each PVT, we asked the participant to self-estimate their own level of drowsiness in terms of the KSS. During each PVT, we also recorded the PSG signals, the RTs (in milliseconds), and the face images of the subject, all in a perfectly time-synchronized manner. All data were collected anonymously.
3.2.2 Psychomotor Vigilance Task (PVT)
The PVT has become one of the most widely used tools to measure performance impair- ments induced by drowsiness. Multiple studies have shown its validity, reliability, and
extreme sensitivity to sleep deprivation [13,44], and by extension to drowsiness. The PVT
gives the RTs to visual or auditory stimuli that occur at random inter-stimulus interval. Compared to other tests, the PVT has the advantage of being almost independent of ap-
titude (low inter-subject variability) and learning (high intra-subject reproducibility) [44].
In our study, we implemented our own version of the 10-minute PVT, adapted from the
one proposed by Basner and Dinges [13]. The subjects were instructed to monitor a red
rectangular box over a black background on a computer screen, and to press a physical, response button as soon as they noticed the appearance within the box of a yellow stimulus counter (expressed in milliseconds). When the button was pressed, the counter stopped and the achieved RT remained displayed for 1 second. RTs below 100 milliseconds were discarded as false starts (errors of commission). After 30 seconds without any response, the counter timed out and displayed a yellow “overrun” message inside the box for a few seconds. The inter-stimulus interval, defined as the time interval between the last response and the appearance of the next stimulus, was varied randomly between 2 and 10 seconds. Furthermore, each time the achieved RT was stopped being displayed, the red box position was randomly varied among five positions on the computer screen, i.e., at its center and at its four corners. In this way, the face images contain more variability in head pose and eye gaze direction, in a similar manner to what would be found in real-life settings.
3.2.3 Polysomnography (PSG) signals
The PSG signals, i.e., the EEG, EOG, ECG, and EMG, are regarded as the “Gold Standard”
Chapter 3. Sleep-deprivation dataset 30
such PSG signals are also useful to characterize drowsiness since the activity in the alpha (8–12 Hz) and theta (4–8 Hz) bands in the EEG signal, and the slow eye movements in the
EOG signal are strong, reliable indicators of drowsiness when performing a task [4,54]. In
our study, we recorded the following channels via electrodes and the portable, laboratory Embla Titanium system, all sampled at a frequency of 512 Hz:
• EEG: Fz, Pz, Cz, C3, and C4 channels, all referenced to the A1 channel, via electrodes
positioned on the scalp following the international 10–20 system [65];
• EOG: two channels for the vertical EOG, via electrodes positioned above and below the right eye; and two channels for the horizontal EOG, via electrodes positioned at the right and the left of the eyes;
• EMC: two channels, via electrodes positioned below the chin;
• ECG: two channels, via electrodes positioned on the chest;
• PGND: one ground channel, used for common mode rejection, via an electrode posi- tioned on the scalp.
3.2.4 Face images
The face images were acquired with the Microsoft Kinect v2 sensor, which provides—for each video frame—a color image, and a pair of aligned, near-infrared intensity and range images. Since drowsiness characterization systems must generally operate in all lighting conditions, including in total darkness, we only retained the intensity and range images, which are both related to active near-infrared illumination. The near-infrared intensity and range images are of size 512×424 pixels, have 16-bit values, and are recorded at 30 FPS. Note that a framerate of 30 FPS corresponds to a temporal resolution of ∼ 33ms, which is sufficient as (1) the duration of a blink is on average greater than 100ms and (2) drowsiness is characterized by long blinks. The camera was positioned just below the computer screen
used for the PVTs, at a distance of about 0.7 m from the subject. Figure3.2shows pairs
of example of near-infrared intensity and range images.
3.2.5 Loss of data
Due to some technical issues, only 88 PVTs (out of 105) from 32 subjects (12 males, 20 females) turned out to be usable. In particular, the PVT1 data was lost for subjects 9, 11, 31, and 32, and never occurred for subjects 7 and 24; the PVT2 data was lost for subjects 9, 12, 14, 31, and 32; and the PVT3 data was lost for subjects 9, 14, 15, 16, 31, and 32.