3.1.

NINscan 4 Multimodality Monitoring and Recording

In our NINscan system designs, we have attempted to maximize the range of monitoring environment and minimize the perturbation to subject’s normal activities. Considering the broad potential applications, we extended NINscan’s monitoring and recording time to increase the opportunities to catch unpredictable transient events, and implemented several auxiliary physiological monitoring channels so that for a specific application, NINscan data collection can be as fully self-contained, complete, and independent as possible. Besides the small form factor and motion resistance, our NINscan series includes two key features. First, among these is long term and independent data collection capabilities. With its internal memory, NINscan by itself can collect data continuously and independently for more than 24 h. NINscan data collection does not need a central receiver (computer) to support the data collection, such as those required by wireless communication-based aNIRS devices. This way the total weight of the devices required for data collection is reduced (NINscan 4 with batteries weighs only 170 g), and there is less restriction to the range of activities (without distance restrictions such as needed for reliable wireless communication-based aNIRS monitoring). This is especially important for monitoring in extreme environments, such as in the study of mountain sickness and space applications. Second, NINscan 4 provides multimodality monitoring capabilities, including three actigraphy channels ($x$, $y$, and $z$ accelerations) and one ECG channel are recorded together with the systemic and cerebral hemodynamic channels; all channels are strictly synchronized.

In this article, using battery-powered NINscan 4, we performed simultaneous and continuous monitoring of cerebral hemodynamics, systemic hemodynamics, ECG, and actigraphy for 24 h, at 250-Hz sampling rate. Subjects were asked to keep a voice or written log of their activities throughout the day, and the two integrated event buttons were utilized by subjects to mark special events. The timing of the event button press was recorded to help precisely identify the events of interest.

To monitor cerebral and systemic hemodynamics, four aNIRS channels were implemented. A dual wavelength (785 and 830 nm) laser diode was used as the light source (Axcel Photonics, Inc., Marlborough, Massachusetts), and two OPT101 photodiodes (Texas Instruments, Inc., Dallas, Texas) were used as detectors. Laser power at each wavelength was about 3.5 mW, under the maximum permissible exposure (MPE) standards for skin illumination (The skin MPE is about $300\mathrm{mW}/{\mathrm{cm}}^2$ for NIR light, depending on the wavelengths; and NINscan illumination intensity is about $30\mathrm{mW}/{\mathrm{cm}}^2$). An optical diffuser was used to expand the laser beam profile and reduce spatial density of the light. The systemic hemodynamics was acquired from a source–detector pair with 1.5-cm separation, whereas the cerebral hemodynamics was acquired from source–detector pair at 4.0-cm separation. We will refer to measurements from short source–detector pair or the scalp/skull hemodynamics as “systemic” because it is not regulated, as a comparison to the regulated cerebral hemodynamics. Precisely speaking, it is only a local estimate of the systemic hemodynamics. The four aNIRS channels were implemented using frequency division multiplexing and lock-in demodulation scheme, with a signal detection bandwidth of 20 Hz. A three-dimensional (3-D) accelerometer, ADXL325 (Analog Devices, Inc., Norwood, Massachusetts), was used as the sensor for NINscan 4 body position/orientation and actigraphy measurements. The accelerometer was sensitive to both static acceleration of gravity (e.g., tilting), as well as dynamic acceleration, resulting from motion, shock, or vibration. The accelerometry channels measured acceleration in $x$, $y$, and $z$ directions up to $+/-5$ $g$, with a bandwidth of 40 Hz. The single ECG channel utilized the standards for regular ECG Holter monitoring, with a bandwidth of 0.1 to 40 Hz; and disposable Ag-AgCl ECG electrodes were used for ECG monitoring.

Successful long-term aNIRS recording in people’s real life activities depends critically on the system’s resistance to motion artifact; therefore, it is also worth mentioning the motion resistance features of NINscan system. In NINscan 4, many measures to improve the systems resistant to motion artifact were included in the design. Motion recording-based adaptive filtering algorithm have been shown to reduce motion artifacts;³⁹ our tests suggest that both simultaneous recording of subjects’ motion signal and the multidistance optical probe helps with the identification, reduction, and management of motion artifacts during NINscan monitoring. In our NINscan probe, there are four optical channels, the short source–detector distance channels are used for monitoring of systemic hemodynamics, and the far source–detector channels are used to monitor cerebral hemodynamics. Usually motion artifacts appear in both near and far optical channels, and this “common mode” feature of motion artifact can be used for motion artifact identification. With a multidistance optical probe, we can also apply scalp/skull interference correction in our data analysis to acquire higher sensitivity and specificity in brain function measurements.^40–44 Other details of motion artifact resistant probe and system design have previously been reported.²⁴ These include soft and light weight probe design, with flexible curvature adapted to the of the shape of the head; adhesive probe surface to secure the probe’s position relative to the skin, and reduce the possibility of probe shifting; elimination of any sharp edges or pressure-points and appropriate attachment of the probe to the head to ensure subjects’ comfort. Since the laser diode is integrated directly on the probe, it is important that the diode has efficient heat dispensing structure. With prolonged monitoring time, what are minor issues for short-term measures, such as the pressure of the probe or temperature of laser diodes, become critical and need to be carefully managed to ensure subjects’ safety and comfort. For long-term probe attachment, it is also important that the probe be made of medical grade and skin friendly materials to reduce possible skin irritation.

3.2.

Human Subject Data Collection

The experimental protocol was approved by the MGH IRB and data were collected from healthy subjects. The optical probe was secured via its adhesive surface and the head band over the right prefrontal cortex (between 10/20 position ${\mathrm{F}}_4$ and ${\mathrm{F}}_{\mathrm{p}2}$, or approximately Brodmann areas 10 and 46). The reasons we choose to monitor prefrontal cortex in our feasibility tests is because the area is important in attention, working memory,⁴⁵ decision making,⁴⁶ and sleep-affected neurophysiologic processes,⁴⁷ and that it is easy to attach the probe and acquire good signal quality on the forehead. ECG was collected from chest leads. The motion sensor was attached to the optical probe on the head. The recorder was carried in the subject’s pocket or a separate small bag, as was convenient.

At the beginning of the recording, an event button integrated in the NINscan 4 box was pressed at a specific time on the subjects’ watch, which was used to keep the activity log. This way the NINscan recording time and the diary time could be synchronized. Subjects were asked to press event buttons to record the beginnings and ends of consecutive activities during the entire data collection period. Event marker presses were accompanied by verbal entries recorded through a voice recorder, which were subsequently compiled into an auditory data activities log used in data preparation and analysis.

Few restrictions were placed on the subjects’ activities during the 24-h recording. The most significant one was to avoid getting the NINscan device wet (e.g., bathing). Two data collection examples are shown in Fig. 1.

Fig. 1

Examples of 24-h ambulatory recording of cerebral hemodynamics, systemic hemodynamics, ECG, and actigraphy during subjects’ daily activities. (A) Monitoring of a subject before he went on the subway. For this subject, the optical ambulatory near infrared spectroscopy (aNIRS) probe and the accelerometer were located on the right side of the forehead. This subject used a voice recorder (together with the event buttons) to log events during his daily activities. The NINscan 4 recorder was in his pocket (not shown in the picture). (B) Monitoring of a subject while he was playing baseball in the field. The aNIRS probe and the accelerometer were again placed on the right side of the subjects forehead, and the NINscan 4 device was conveniently put in his shirt pocket. For both subjects, ambulatory ECG was simultaneously recorded using standard chest leads.

3.3.

Data Analysis

For aNIRS channels, the raw 250-Hz optical data were corrected for dark current and digitally low pass filtered at 5 Hz (in addition to the instrument filter). To acquire the total hemoglobin (HbT) and oxygen saturation (Sat) levels, we first calculated the optical density changes (delta OD) from raw aNIRS measurements. This was done by normalizing the 24-h raw datasets with the averaged data collected during earliest morning 0:00 to 0:30 a.m. in that dataset (defined as the “baseline” for this test). Combining the multidistance aNIRS measurements, the interference from the superficial layers (e.g., scalp and skull) were estimated and removed from cerebral hemodynamics calculation similar to previous approaches.^40–44 The deoxy-hemoglobin (Hb) and oxy-hemoglobin (${\mathrm{HbO}}_2$) concentration changes (compared with the 0:00 to 0:30 a.m. baseline section) were calculated using modified Beer–Lambert law^48,49 and layered tissue model;^40,44 and then the total hemoglobin concentration (HbT) change and tissue oxygen saturation (Sat) are calculated:

Here, [$\delta \mathrm{Hb}$] and [$\delta {\mathrm{HbO}}_2$] are the concentration changes of Hb and ${\mathrm{HbO}}_2$, [$\delta \mathrm{HbT}$] is the concentration change of the total hemoglobin, Sat is the oxygenation level, and [${\mathrm{Hb}}_0$] and [${\mathrm{HbO}2}_0$] are the averaged (or baseline) Hb and ${\mathrm{HbO}}_2$ concentrations. Since NINscan uses continuous wave NIRS techniques, it provides measurements of the change of ${\mathrm{HbO}}_2$ and Hb concentrations only, therefore the oxygen saturation level was calculated based on estimated tissue concentrations: $[{\mathrm{Hb}}_0]=18\mu \mathrm{Mol}$ and $[{\mathrm{HbO}2}_0]=41\mu \mathrm{Mol}$. These baseline concentrations were estimated based on the literatures^43,50 as well as our radio frequency NIRS measurements.

For ECG analysis, the QRS detection and heart rate estimation were performed based on the traditional Hamilton algorithm,⁵¹ available from Physionet.org. Median filtering over 5 heart beats was used to smooth the final heart rate.

For actigraphy analysis, the raw accelerometer readings were converted to gravity units ($g$ values). The orientation and position of the accelerometer were recorded, so the correct $x$ (left–right), $y$ (posterior–anterior), and $z$ (inferior–superior) directions could be identified. Therefore, both body position and movement of the subject’s head could be measured. To estimate the activity level, we use the digital integration method to calculate an “activity score:”²⁶ the amplitude of the total acceleration (combined acceleration from $x$, $y$, and $z$ axes) was band-pass filtered to 0.25 to 3 Hz, then integrated using a time window of 2 s. An actigraphy-based algorithm can be used to automatically distinguish sleep episodes from wakefulness. Previous work^27,28 shows that this algorithm can correctly distinguish sleep from wakefulness $\sim 90\%$ of the time, as compared with the gold standard polysomnography. Since our accelerometer was placed on the head instead of the wrist (as most commercial actigraphy meters do), and that the coefficients and thresholds used in commercial products to distinguish sleep and wakefulness are proprietary, we set up our own parameters. In the equation to calculate $D$ and identify sleep and wakefulness,^27,28 we set $P=0.13$; ${\mathrm{W}}_{-4}\sim {\mathrm{W}}_{+2}$ to be 0.010, 0.015, 0.028, 0.031, 0.085, 0.015, and 0.010, respectively; and $D<0.00055$ indicates sleep.

As a simultaneously monitored auxiliary parameter, accelerometer data (from which actigraphy is derived) provides the following information: (1) quantitation of activity and movement; (2) identification of motion artifacts for data analysis and potential correction;³⁹ (3) subjects’ body position (e.g., stand, lying down supine), as blood distribution and even tissue status may vary as body position changes, therefore, it is important to use body position as a reference while explore tissue hemodynamics and activity; and (4) help substantiate a log of activities. Actigraphy, together with NINscan 4 event buttons and manual/verbal logs, serves as a good “diary” keeping and checking tool, allowing us to easily identify different types and timings of real life daily activities, such as sleep, work, sports, riding elevators, and even driving.

To reduce the temporal correlation of the measurements in our time series, we down sampled the time series before the statistical analysis. For example, in our long-term circadian study, where the signal was filtered to 0.005 Hz, before linear regression we reduced the number of measurements to $200\mathrm{s}/\mathrm{sample}$, or 438 data points per one 24-h time series. Linear regression analysis was conducted, with significance identified by $p<0.05$.

4.1.

Raw 24-h Recording Data

An example 24-h raw dataset is shown in Fig. 2. In this recording, data collection was initiated at 11:26 a.m. on day 1, and completed at 11:46 a.m. on day 2. During the 24-h data collection period, the subject participated in his normal daily activities, including miscellaneous static and dynamic tasks at home, preparing food, eating, daily hygiene, socialization with family members and coworkers, commute to and from work using public transport (subway and shuttle bus), resting, yawning, falling asleep, and sleeping. During the night, the subject slept with minor interruptions. All signals in Fig. 2 are plotted as raw digitization units ($A/D$ $\text{range}=0$ to 4096), and the rows in each subfigure are motion (actigraphy), ECG, and aNIRS recordings, respectively.

Fig. 2

An example 24-h multimodality NINscan 4 monitoring dataset. (A) 24-h overview of actigraphy, ECG, and aNIRS raw data. This dataset demonstrates good signal quality, and we can clearly see (from actigraphy) the activity level change between awake and sleep periods. The timings of event button press were all logged to the millisecond. (B) View of a 1-h raw data before and after pressing an event button. The diary indicated the subject was working on the computer after pressing the event button. (C) View of 2-min raw data to demonstrate in detail the bandwidth and quality of the raw recordings.

Figure 2(a) is an overview of the whole 24-h recording. The 24-h overview is usually used to ensure the overall signal quality, and to identify possible low quality segments such as those from signal saturation or sensor detachment (neither of these were detected). From the accelerometer recording, we can clearly identify the body position and activity level change during the day and during the night. From the amplitude range, we can tell that the ECG recordings are high quality, with occasional artifacts due to motion. The aNIRS signals exhibited no saturation, signal dropout, or other unexpected features. During the 24 h recording, the subject pressed the two event buttons 80 times, the timings of which, labeled as E1 or E2, were all precisely logged into the memory and shown in this figure. Together with the subjects’ detailed diary, different events, and activities can be accurately identified, and their associated body positions, movements, ECG, systemic, and cerebral responses can be calculated.

Figure 2(b) expands 1 h of the recording [the segment in yellow in Fig. 2(a)] following one of the event button presses in the evening. The associated diary for this event button press read: “Got up from the low arm chair, walked to the kitchen with the computer, sat in the high-back chair, continued working on the computer”. The three-axis accelerometry recordings demonstrated the body position change associated with this “sat in the high-back chair” event, and the subject’s body position remained stable after that, with occasional small movements. The ECG recordings and aNIRS recordings were stable through this period as well.

Given the high-resolution recorded data, even the zoomed Fig. 2(b) only shows the amplitude range and low-frequency trend in signal variations. In Fig. 2(c), the curves were further zoomed to show 2 min of data [the segment in pink in Fig. 2(b)] in detail. In this subfigure, we can clearly see the individual heart beats in ECG and aNIRS, which demonstrate the bandwidth and high quality of the raw recordings. From these data, aNIRS channels in NINscan 4 appear more motion resistant than ambulatory ECG recordings. This can be verified by comparing the motion, ECG and aNIRS signal in Fig. 2(c).

4.2.

Motion Resistance of NINscan 4 Recordings

To further demonstrate the motion resistant features of NINscan 4, we examined the influence of head motion in our aNIRS recordings, and compared it with the subject’s spontaneous and physiological signal variation levels.

During one of our 24-h ambulatory recordings, the subject was asked to rapidly shake his head “no” (moving from left to right, and then back). After completing the recording, we compared the subject’s systemic hemodynamic measurements (calculated from aNIRS channels with source–detector separation of 1.5 cm) during the head movement with the spontaneous signal variations during his motionless sleep in the night.

Figure 3 shows the result. Column A is the episode of the recordings when the subject was sleeping. In this $\sim 2\text{-}\min$ recording, the subject was in a laying-down position, facing left. In the first row in column A, the total acceleration (combined acceleration from all three directions) indicates that the subject was very still, with no obvious movement ($<0.02g$). From the systemic hemodynamic recording (second row), we can see that the amplitude of cardiac activity in [$\delta \mathrm{Hb}\mathrm{T}$] was about 3 μM. A spontaneous low-frequency fluctuation can be identified, with a valley at 03:35:20 a.m. and a magnitude of $-12\mu \mathrm{M}$. Column B shows the result when the subject rapidly shook his head. The first row of column B demonstrates that this intentional head movement introduced a total acceleration of $\sim 1g$ (on top of the $1\text{-}g$ gravity level). The aNIRS recordings, however, were essentially unaffected by this motion. The major hemodynamic variation is still the cardiac activity, which is totally out-of-phase with the head movement; and the magnitude of the cardiac activity (about 3 μM) did not change during the activity as compared with no head movement at the beginning and end of this period.

Fig. 3

Tests for NINscan 4 motion resistance. Column A: One segment of NINscan 4 recordings during sleep, which demonstrated the level of physiological signal variation. The total acceleration recordings (row 1) showed that subject had little movement during sleep; the systemic hemodynamics (row 2) showed the subject’s physiological variations (cardiac about 1 mM, slow wave about 6 mM). The linear regression analysis showed no correlation between hemodynamic variations and motion. Column B: NINscan 4 recordings while the subject was shaking his head. The adhesive probe was secured on the right side of the subject’s forehead. The accelerometry recordings showed the subject’s total acceleration was up to 1 $g$, the systemic hemodynamics measurements did not show obvious motion artifacts (row 1) corresponding to the head movement; the major variation in the signal was from cardiac activity. Again linear regression analysis showed no correlation between hemodynamic variations and motion.

For both the sleep and the head movement data segments, we further plot the hemodynamic variation as a function of the head acceleration (third row in Fig. 3). For linear regression analysis, the range of the [$\delta \mathrm{HbT}$] and the total acceleration were first normalized to range of 0 to 1, and the signals were down sampled to 5-Hz sampling rate to reduce temporal autocorrelation and improve the independence of separate measurements. For the sleep data section, the linear regression of [$\delta \mathrm{HbT}$] versus total acceleration gave Pearson correlation coefficient of $r=-0.07$ ($\text{slope}=-0.09$; $p>0.05$); for the head shaking data section the regression gave Pearson correlation coefficient of $r=0.01$ ($\text{slope}=0.01$; $p>0.5$). Therefore, acceleration from both vigorous head movement ($\sim 1g$) or sleep ($\sim 0.02g$) demonstrated no significant change in the hemoglobin measurements in aNIRS; compared with the cardiac and other spontaneous physiological hemodynamic variations in the long-term recordings, artifacts introduced by head motion were minor and could be ignored.

5.1.

Application of NINscan 4 Recordings in Circadian Study

Comparison of physiological or pathological features during wakefulness and during sleep is key to the study of people’s circadian physiology. For example, it is well known that during sleep, due to reduced physical activity of the human body, subjects will demonstrate lower heart rate and overall metabolism level. Circadian brain physiology is of considerable interest but—while many technologies have been applied, such as ECG, electroencephalography (EEG), fMRI in the study of sleep and circadian cycles—so far no existing technology has been able to continuously monitor and compare cerebral hemodynamic features through a 24-h cycle.

Using NINscan monitoring, we are able to compare the cerebral hemodynamic features during wakefulness (normal daily activities) and during sleep. Figure 4 shows a typical result comparing the average cerebral blood volume during daily activities and during sleep (at the subject’s home). Again, the NINscan probe was located on the forehead over the right frontal cortex, and cerebral hemodynamics, systemic hemodynamics and three axis accelerometries were all performed on this region. In order to clearly demonstrate the low-frequency trend in the circadian features, we low pass filtered the signals with a cutoff frequency of 0.005 Hz. The result is shown in Fig. 4 and Table 1.

Fig. 4

Application of multimodality NINscan 4 monitoring in circadian studies: comparison of actigraphy (A), heart rate derived from ECG (B), cerebral hemodynamics including total hemoglobin concentration change (C), and oxygenation change (D) during wakefulness and sleep. The shadowed section indicates the bedtime logged on the subject’s diary; the line segments in the actigraphy indicate the sleep time detected using actigraphy-based algorithm.

Table 1

Application of multimodality NINscan 4 monitoring in circadian studies: comparison of actigraphy, heart rate, cerebral hemodynamics during wakefulness and sleep.

Figure 4(a) shows the change in activity level across a full day. It is clear that during the night the activity score was significantly reduced. From Table 1, we see that during wakefulness the average activity score was 0.0092, while during sleep it was 0.0001, more than 90-fold reduction. Here, the wakefulness and sleep were identified using the actigraphy algorithm. The variations in activity level and the heart rate [Fig. 4(b)], derived from ECG, were closely aligned during wakefulness (linear regression on data down sampled to one measurement ever 200 s; $r=0.78$, $p\ll 0.0001$). Average heart rate during sleep in this subject was 43 bpm, 20% less than the heart rate during the day, and the heart rate decrease is significant ($p\ll 0.05$). This overall low resting heart rate was confirmed by both NIRS and ECG using independent instruments.

Although reduced activity level and heart rate during sleep is expected, changes in the cerebral blood volume level [Fig. 4(c)] and oxygenation level [Fig. 4(d)] during sleep compared with wakefulness have not to our knowledge been experimentally studied. In this single case, we compared the blood volume level and oxygenation level during sleep and wakefulness to demonstrate the feasibility of this analysis based on long-term data collection. Our results show that [$\delta \mathrm{HbT}$] (HbT change compared with the baseline level measured at 12:00 to 12:30 a.m.) was $8.7+/-2.2\mu \mathrm{M}$ during wakefulness, and $6.9+/-3.3\mu \mathrm{M}$ during sleep. This represents a significant 21% reduction in [$\delta \mathrm{HbT}$] ($p\ll 0.0001$). The cerebral oxygenation level change, in contrast, was not significant, being estimated at $77\%+/-5.3\%$ during wakefulness and $78+/-3.9\%$ during sleep ($p>0.05$). Tissue oxygen saturation level should lie between the arterial oxygen saturation and venous oxygen saturation; it is an averaged value for blood contained in all arteries, arterioles, capillaries, venules, and veins in the area probed by the photons (related to the diffuse light sensitivity profile). It has been shown that sleep cerebral metabolic rates of glucose and oxygen decrease by 44% and 25%, respectively, during specific phases of sleep.^52,53 When further assuming a stable oxygen extraction rate, the static cerebral oxygenation would suggest a reduced blood volume, as measured.

NINscan provides a unique opportunity to measure and compare parameters estimated from long duration measurements and during daily activity. Sleep is widely studied using NIRS,^47,54–59 however, most of the previous studies were performed with short term tests (e.g., afternoon nap) or sleep in the lab environment. The result presented here demonstrates the long-term data acquisition and analysis capability of the NINscan monitoring, and to the best of our knowledge, is the first to extend the experimental study of cerebral hemodynamics during wakefulness and sleep during a full 24-h period of unrestricted activity. This will enable, for example, detailed investigation of circadian influences on cerebral physiology.

5.2.

Unpredictable Symptom Capture: Response to Environmental Noise during Sleep

Many physiological or pathological conditions—syncope, epilepsy, schizophrenia, hot flashes, delayed stroke—feature transient symptoms with unpredictable onsets. Detecting such conditions requires long-term ambulatory recording to increase the opportunity catching such unpredictable events. Here, we present an interesting nonclinical example NINscan 4 recording of a subject who woke up unexpectedly to environmental noise.

The unexpected awakening event happened in the early morning when the subject was in motionless sleep. The subject was suddenly awoken by environmental noise. He raised his head, determined what was happening, pressed the event button, then went back to sleep. The subject recalls the event being quite short: he felt that he “immediately” woke up at the environmental noise and from his being aware of the noise to his going back to sleep; the whole process was estimated to take 1 to 2 min.

The event as captured by NINscan 4 monitoring is shown in Fig. 5. The four rows in the figure are A, three axis acceleration (bandwidth 0 to 5 Hz); B, heart rate (derived from ECG, 5 heart beat median filtered); C, blood volume change (bandwidth 0 to 5 Hz) and D, oxygenation change (bandwidth 0 to 5 Hz), respectively.

Fig. 5

Example of NINscan 4 recording of a transient event with unpredictable onset. In this segment of data, the subject was woken up unexpectedly by environmental noise. The E1 event button was pressed by the subject upon awakening, and from the NINscan 4 recording we can clearly see the (A) actigraphy, (B) heart rate derived from ECG, (C) cerebral total hemoglobin concentration change, and (D) cerebral oxygen saturation before, during and after this unexpected event. The horizontal green bar in the actigraphy recording is the sleep section identified by the actigraphy-based sleep/wakefulness identification algorithm.

The event button was pressed at 4:30 a.m., and we can clearly see motion, heart rate and cerebral blood volume, and oxygen saturation change before, during, and after this event. From the motion channels, we can see that the subject was in his supine position (blue, green, and red indicates $x$, $y$, and $z$ directions, where $y$ is points up with magnitude of $0.75g$), and remained very still until 4:26 a.m., where he showed two slight movements. The subject then woke up at 4:30 a.m., and demonstrated substantial head motion. He lay down again, with slight body position change. After 4:34 a.m., the subject was motionless again.

What is interesting in the recordings is that the subject recalled this event being very short, and he felt that he slept very well both before and after this short event (in Fig. 5, we marked the section where the subject felt conscious with yellow background). However, the recordings of heart rate, cerebral blood volume, and oxygenation all demonstrate changes several minutes “before” the subject felt conscious. The subject’s heart rate was about 41 during sleep, at 4:26 a.m. his heart rate demonstrated significant increase (up to 60 bpm) and fluctuation, along with enhanced cerebral hemodynamic fluctuations, although he was not physically moving (as indicated by the mostly silence in the 3-D accelerometer recordings). His response to the environmental noise, shown as elevated heart rate, cerebral blood volume, and oxygenation (both in magnitude and dynamic fluctuation) continue, without him being conscious of the environmental noise, until the point when he felt that he woke up and begun to move (4:30 a.m.). All the parameters were further elevated and peaked at 4:30:50 a.m., where the heart rate went up to 78. After 4:34 a.m., all parameters quickly went back to the regular sleep level. The consequences of environmental noise on the recorded physiology were very consistent: almost all parameters including heart rate, cerebral and systemic hemodynamics show coincident variations.

The horizontal green bar in the actigraphy recording is the sleep section identified by the actigraphy-based sleep/wakefulness identification algorithm. Although not as early as the elevation of heart rate and hemodynamic level, the beginning of the episode of “wakefulness” identified using the actigraphy-based algorithm (4:28 a.m.) is also earlier than the subjective feeling described by the subject.

This example demonstrates an interesting and fairly large difference between the “objective measurements” and the “subjective feeling,” which is important in many social or psychiatric studies. It also demonstrate that with long-term ambulatory monitoring techniques it is possible to record objective physiological or pathological parameters leading, during, and following a specific and unpredictable event; the data collection procedure and the data analysis for the “unexpected sudden wakeup” example presented here is not only relevant to sleep studies, but also for catching and analyzing symptoms such as epilepsy, syncope, and others. We expect NINscan 4 to be a powerful tool to understand the conditions that lead to those symptoms, the features of the events, and their consequences measured in cerebral and systemic perfusions and oxygenation.