Detection of arousal level and characterization and treatment of disorders of consciousness based on electrical signals in the brain

Abstract

A model may be generated for characterization of a fluctuation in an arousal level based on EEG data for a cohort of subjects. Changes in gamma and delta waves can be used to detect arousal units (AUs) for cortical periods. One or more metrics relating at least in part to gamma and theta waves can be determined for each AU. Centroids for the cortical periods can be used to train the model to generate an arousal index. The model can be used to determinate arousal level in one or more patients. For example, EEG data can be acquired for a subject using a set of sensors. AUs in the EEG data can be detected, and a characterization of a fluctuation in an arousal level of the subject can be generated using the model. The characterization may be used in evaluating and / or treating a health condition, such as a consciousness disorder.

Description

[0001]Atty. Dkt. No.: 093873-1450 DETECTION OF AROUSAL LEVEL AND CHARACTERIZATION AND TREATMENT OF DISORDERS OF CONSCIOUSNESS BASED ON ELECTRICAL SIGNALS IN THE BRAIN CROSS-REFERENCE TO RELATED PATENT APPLICATIONS This application claims priority to U.S. Provisional Patent Application No. 63 / 598,734 filed November 14, 2023, the contents of which are incorporated herein by reference in its entirety. BACKGROUND The present application relates generally to detection of brain electrical activity (e.g., fluctuations in electrical activity) in detection of arousal level, and to generation of predictive biomarkers for evaluation and treatment of disorders of consciousness. SUMMARY Various embodiments of the present disclosure relate to a system. The system may include at least one neural sensor disposed on a subject, the neural sensor configured to record neural activity of the subject over a period of time and output neural data. The system may also include at least one visual sensor configured to record motor activity of the subject over the period of time and output motor data. The system may also include at least one autonomic data sensor configured to record autonomic system activity of the subject over the period of time and output autonomic data. The system also includes one or more processors configured to determine a predictive biomarker for disorders of consciousness via an algorithm. The algorithm may include the step of receiving the neural data from the at least one neural sensor. The algorithm may include the step of receiving the motor data from the at least one visual sensor. The algorithm may include the step of receiving the autonomic data from the at least one autonomic data sensor. The algorithm may include the step of analyzing the neural data to identify the dominant frequency ranges over the period of time. The algorithm may include the step of correlating the motor data that co-occurred with the identified dominant frequency ranges. The algorithm may include the step of correlating the autonomic data that co-occurred with the identified dominant frequency ranges. The Atty. Dkt. No.: 093873-1450 algorithm may include the step of generating a predictive biomarker, the predictive biomarker configured to indicate a level of arousal of the subject. In various embodiments, the at least one neural sensor is configured to detect cortical neural activity and output cortical neural data. In various embodiments, the at least one visual sensor is a video camera. In various embodiments, the algorithm further includes the step of identifying a plurality of cortical states from correlated motor data that co-occurred with the identified dominant frequency ranges of the cortical neural data. In various embodiments, the algorithm further includes the step of associating the plurality of cortical states with a plurality of motor states presented in the correlated motor data that co-occurred with the identified dominant frequency ranges of the cortical neural data. In various embodiments, the plurality of motor states includes: twitches, limb and generalized body movements, weak weight-bearing movements, and organized movements. Additional embodiments of the present disclosure relate to a method of generating (e.g., training, updating, etc.) a model for the characterization of a fluctuation in an arousal level of a subject. The method includes the steps of: acquiring EEG data for a cohort of subjects; detecting, based at least in part on changes in gamma waves and delta waves in the EEG data, arousal units for a plurality of cortical periods; determining, for each arousal unit, a plurality of metrics relating at least in part to gamma and theta waves in the arousal unit; generating a centroid for each cortical period, the centroid based on the plurality of metrics for the arousal units in the corresponding cortical period; training, using the centroids of the plurality of cortical periods, the model to generate an arousal index; and providing the model for determination of arousal level in one or more subjects. In various embodiments, each cortical period corresponds to a period of consciousness. In various embodiments, the plurality of metrics comprises gamma power. In various embodiments, the plurality of metrics comprises theta frequency. Additional embodiments of the present disclosure relate to a method of using a model. The method includes: acquiring EEG data from a subject using a set of sensors; detecting arousal units in the EEG data based at least in part on changes in gamma, theta, and delta waves in the EEG data; generating a characterization of a fluctuation in an arousal level of Atty. Dkt. No.: 093873-1450 the subject, wherein generating the characterization comprises determining an arousal metric for the subject based on the arousal units; and using the characterization in evaluating and / or treating the subject for a health condition. In various embodiments, detecting the arousal units comprises detecting a span of increased gamma activity, identifying, within the span, a sub-span of decreased delta activity, and detecting a gamma burst within the sub-span. In various embodiments, the gamma burst comprises a gamma level exceeding a threshold. In various embodiments, the health condition is a disorder of consciousness. In various embodiments, the disorder of consciousness is related to a coma, a coma like state, a metabolic disorder, or a brain injury. This summary is illustrative only and is not intended to be in any way limiting. Other aspects, inventive features, and advantages of the devices or processes described herein will become apparent in the detailed description set forth herein, taken in conjunction with the accompanying figures, wherein like reference numerals refer to like elements. BRIEF DESCRIPTION OF THE DRAWINGS The disclosure will become more fully understood from the following detailed description, taken in conjunction with the accompanying figures, wherein like reference numerals refer to like elements. FIG. 1 is a schematic diagram showing the composition of arousal units (AUs), features, and applicability to the clinic, in accordance with various example embodiments. AUs represent cortical patterns that consist of interactions between three spectral variables: fractional gamma, delta, and gamma power that can be sequentially identified within specific windows over the EEG traces and that are linked to autonomic changes (increased breathing frequency) and behavioral changes (top) consistently found as rodents and humans emerge from unconscious states and reach a wakeful state. Therefore, they are reliable, reproducible, and translatable (middle). These AU elements derive a continuous and quantitative method to measure arousal levels in states with impaired consciousness, such as hypoglycemic coma, hypoxia, diffuse brain injury, and recovery from anesthesia in rodents and humans. FIGS. 2A-2D are exemplary plots, graphs, and charts showing that corticomotor regimes are distinguished in the frontal association area, in accordance with various example Atty. Dkt. No.: 093873-1450 embodiments. FIG. 2A shows representative EEG traces from the frontal association area (FAA) recorded during the emergence from controlled-concentration isoflurane ramp (pink line). FIG. 2B shows a normalized spectrogram (deviation from median) calculated from FAA EEG (color bar shows power measured in dBs). FIG. 2C shows dominant frequency ranges K-Means clustered into six cortical ranges (black: 3-5Hz, gray: 4-8Hz, light blue: 10- 20Hz, dark blue: 20-40Hz, purple: 30-100Hz and lilac: 70-130Hz), density estimation of cortical states, and segmentation of cortical periods after using the density estimation function and applying an abrupt change detection algorithm as to define cortical periods. Period 1 (P1) to Period 5 (P5). FIG. 2D shows detected movements from video during emergence: trunk twitch (gray), limb and body movement (blue), weak weight-bearing (wwb; dark green) including dragging and quivering, and organized movements (red lines). FIGS. 3A-3C are exemplary data traces, plots, and graphs showing that arousal units reflect cortical patterns associated with behavioral and autonomic changes during emergence from anesthesia, in accordance with various example embodiments. FIG. 3A shows a normalized spectrogram (deviation from the median) calculated from Frontal Association Area (FAA) EEG (color bar shows power measured in dBs) obtained during the emergence from controlled-concentration isoflurane ramp (pink line) and aligned with detected movements from video during emergence: trunk twitch (gray), limb and body movement (blue), weak weight-bearing (wwb; dark green) including dragging and quivering, and organized movements (red). FIG. 3B shows inputs to an algorithm to detect arousal unit features were raw bilateral EEGs recorded from FAA in an assessment epoch≈15min. FIG. 3C shows fractional gamma per window was concatenated to obtain a time series — magnification of a timespan with significantly increased fractional gamma through termination, and arousal unit components (red box) showing the time a cortical switch (dashed pink rectangle) was detected. Delta power per window was concatenated within the established span to obtain a time series. The gray depression signaled a significant drop in the delta power. Burst of gamma (dashed light green rectangle) was observed together with the onset of movement. Gamma power per window was concatenated to obtain a time series (right). The maximum of this series was recognized as burst (orange peak). Timepoints of breaths (gray crosses) linked to cortical changes automatically detected from the video. Bottom: normalized movement detected per frame automatically extracted from the video. Atty. Dkt. No.: 093873-1450 FIGS. 4A and 4B are an exemplary spectral analysis of FAA-EEGs epochs, in accordance with various example embodiments. FIG. 4A shows raw bilateral EEGs recorded from frontal association areas in an assessment epoch≈15min. FIG. 4B shows EEGs were processed using a sliding window with length=5s and step size=1s (left). For each sliding window, the spectrum was calculated, and power was measured at 100 log-spaced frequencies between 2-150Hz. Power was measured in watts to characterize non-gamma frequencies and was normalized by its order of magnitude (middle). And spectra across both channels were averaged and spectral variables were calculated, including fractional gamma, total delta, and total gamma power per sliding window (right). FIGS. 5A and 5B show exemplary traces, plots, and graphs of arousal indices monitoring emergence from isoflurane, in accordance with various example embodiments. Subjects 1-8 show top: representative EEG traces from Frontal association area (FAA) recorded during the emergence from controlled-concentration isoflurane ramp (pink line) and normalized spectrogram (deviation from median) calculated from FAA EEG (color bar shows power measured in dBs). Middle: detected movements from video during emergence: trunk twitch (gray), limb and body movement (blue), weak weight-bearing (wwb; dark green) including dragging and quivering, and organized movements (red). Bottom: Arousal index plot displaying the trajectory of each individual while restoring a wakeful state. FIGS. 6A-6E are exemplary plots, charts, and graphs showing arousal units distinguishing arousal events across cortical periods and predicting movement, in accordance with various example embodiments. FIG. 6A is a raster plot of arousal units (onset) distributed across cortical periods (P1-P5) in isoflurane long ramps from 8 mice. Arousal units (lilac bars) were extracted from EEGs by applying the automated detection algorithm. FIG. 6B shows the number of breaths per min before (gray) and after the cortical switch (brown; P2, n=17; P3, n=58 & P4, n=32 arousal units). FIG. 6C show the percentage of body area moving along frames measured before (gray) and at burst onset (brown; P2, n=15; P3, n=59 & P4, n=36 arousal units; n=5 animals). FIG. 6D shows accuracy and FIG. 6E shows sensitivity of arousal units to dominant movements (twitches, gray; limb movements, blue; weak weight-bearing movements, green; organized movements, red) in 15 minute-assessment epochs along the cortical periods (P2, n=8; P3, n=23; P4, n=26 & P5, n=12: n=8 animals). Atty. Dkt. No.: 093873-1450 Data shown as mean ± s.e.m. Paired sample and right tailed Wilcoxon test. **P=<0.01 and ***P=<0.001. FIGS. 7A-7H are exemplary plots, charts, and graphs showing that gamma-theta patterns within arousal units quantitatively discriminate arousal levels, in accordance with various example embodiments. FIG. 7A shows averaged spectra from arousal units at cortical switch (gray) and at burst (blue) extracted from all periods (P2, n=27, P3, n=106, P4, n=126, and P5, n=94 arousal units) of 8 mice exposed to anesthetic ramps. FIG. 7B shows theta frequency (100 log-space) measured at cortical switch(gray) and at burst(blue) from each arousal unit belonging to a period and extracted from mice (n=8). Two tailed Kolmogorov- Smirnov test showed no significance (n.s.), **P<0.01, and ***P<0.001. Black line displays the comparison of peak theta frequency shifts between cortical switch and burst per period. Gray line shows the comparison of peak theta frequency shifts at the cortical switch along the periods. Blue line depicts the comparison of peak theta frequency shifts at the burst along the periods. FIG. 7C shows arousal unit 100 log-space theta frequency at the burst of each AU along the different cortical periods (P2, n=27, P3, n=106, P4, n=126, and P5, n=94; arousal units of 8 animals). Data shown as mean ± s.e.m. (blue bars). FIG. 7D shows arousal unit gamma power measured at the burst of each AU along the different cortical periods (P2, n=27, P3, n=106, P4, n=126, and P5, n=94; arousal units of 8 animals). Data shown as mean ± s.e.m. (orange bars). *P<0.05, One-tailed paired sample Wilcoxon Signed test. FIGS. 7E- 7H show an automated algorithm that transforms arousal unit’s gamma-theta patterns to arousal indices. Algorithm consists of a learning phase and a tracking phase. FIG. 7E is a scatterplot showing gamma power-theta frequency from arousal units in FIG. 3A and derived centroids (red triangles) from averaging gamma-theta in each period. FIG. 7F shows Z-score normalized centroids to the mean and standard deviation of the maximal arousal level period (P5). FIG. 7G shows transformation of Z-scored centroid pairs, gamma( Z_γ) and theta (Z_θ) into a Z-scored scaler by applying a non-linear activation function d_c (Z_γ,Z_θ) where if both Z-scored centroids were positive or negative, the scaler equals their sum. Otherwise, the scaler equals the Z-score of gamma or theta, whichever is negative. FIG. 7H shows logistic curves output arousal indices by setting distance between centroids. FIGS. 8A-8E are exemplary plots, charts, and graphs showing the tracking phase of an automated algorithm that transforms arousal unit’s gamma-theta patterns to arousal Atty. Dkt. No.: 093873-1450 indices, in accordance with various example embodiments. FIG. 8A shows detected arousal units from a single animal distributed in the gamma-theta space (gray dots). FIG. 8B shows each gamma-theta pair is Z-score normalized, FIG. 8C shows them transformed to a Z-scored scaler, and FIG. 8D shows them fit into a pre-fit logistic curve to obtain an arousal index. FIG. 8E shows concatenated arousal indices (orange lines) along the cortical periods. The time series was smoothed to obtain a continuous measurement of arousal levels from a single subject (black line). Arousal indices quantify arousal levels, detect fluctuations within periods, and track transitions between cortical periods. FIGS. 9A-9I are exemplary plots, charts, and graphs showing an arousal index quantitatively tracts arousal levels over reversing a hypoglycemic comatose state, in accordance with various example embodiments. FIG. 9A shows an example trace of normalized cortical spectrogram (deviation from median) of an animal emerging from hypoglycemic coma (HC) after glucagon injection (magenta arrow) to reverse the hypoglycemic state. Color bars represent power measured in decibels (dB). Motor recovery extracted from video while mice recover from hypoglycemic coma. FIG. 9B shows (red box) arousal unit components showing at the top the cortical switch (pink dashed rectangle) and a burst of gamma (green dashed rectangle) linked to the onset of movement. Timepoints of breaths (gray crosses) linked to cortical changes. Percentage of moving body parts extracted from the video. FIG. 9C shows calculated arousal index per arousal unit (orange lines) in pharmacologically induced reversal of HC. FIG. 9D is a bar plot that shows number of breaths before and after the cortical switch. One-tailed paired sample Wilcoxon signed rank test (P2, n=9, P3, n=19 & P4, n=34 arousal units). Data shown as mean±s.e.m. FIG. 9E is a bar plot that shows the percentage of body parts moving before and at burst onset. One-tailed Paired Sample Wilcoxon Signed Rank test (P2, n=9, P3, n=49 & P4, n=55 arousal units). Data shown as mean±s.e.m. FIG. 9F shows accuracy and FIG. 9G shows sensitivity of detected arousal units to dominant movements along the cortical periods (P2, n=14; P3, n=48; P4, n=57 & P5, n=6) in all hypoglycemic mice (n=8). Data shown as mean±s.e.m. FIG. 9H shows arousal indices measured in mice spontaneously recovering from HC (gray; n=4 animals) and in those injected with glucagon (magenta) (n=4 animals). Data shown as mean±s.e.m. (shaded area). FIG. 9I shows latency of recovery from HC(P2) to P3 and P4 during spontaneous emergence (gray; n=4 animals), and after glucagon induced emergence Atty. Dkt. No.: 093873-1450 (magenta; n=4 animals). No significantly different (n.s.), *P<0.05 and **P<0.01, and ***P<0.001. FIGS. 10A and 10B depict exemplary plots, charts, and graphs showing that arousal indices track arousal levels in individuals recovering from a hypoglycemic coma, in accordance with various example embodiments. Top: example trace of normalized cortical spectrogram (deviation from median) from mice spontaneously emerging (subjects 1-4) from hypoglycemic coma (HC) and after injection with glucagon (subjects 5-8 magenta dotted line) to reverse the hypoglycemic state (color bar shows power measured in dBs). Middle: detected movements from video during emergence: trunk twitch (gray), limb and body movement (blue), weak weight-bearing (wwb; dark green) including dragging and quivering, and organized movements (red). Bottom: Arousal index plot displaying the trajectory of each individual while restoring a wakeful state. FIGS. 11A-11L are exemplary plots, charts, and graphs showing that an arousal index monitors arousal levels after brain injury, in accordance with various example embodiments. FIG. 11A shows a normalized cortical spectrogram and detected movement of a sham mouse. FIG. 11B shows a normalized cortical spectrogram and detected movement of a mouse recovering from a comatose state secondary to trauma. Color bars represent power measured in decibels (dB). Automated detection of delta waves and burst suppression (black rhomboids) for a sham, shown in FIG. 11C, and brain injured mouse, shown in FIG. 11D. Calculated arousal index per arousal unit (orange lines) using automated algorithm in a sham, shown in FIG. 11E, and brain injured mouse, shown in FIG. 11F. FIG. 11G shows a number of breaths per minute before and after the cortical switch on detected arousal units during recovery from a comatose state after trauma (n=11 animals, one-tailed paired sample Wilcoxon signed-rank test). FIG. 11H shows a proportion of moving body parts before and at burst onset (n=25 arousal units, one-tailed paired sample Wilcoxon signed-rank test). FIG. 11I is a row normalized confusion matrix showing accuracy of arousal units predicting movement (n=10, SBI animals). FIG. 11J shows sensitivity to detected arousal units to different types of movements characterizing motor recovery in SBI mice (n=10). FIGS. 11K and 11L show an arousal index plot displaying the trajectory of sham (n=5) and SBI mice (n=10) while restoring a wakeful state. Data shown as mean ± s.e.m. *P<0.05, **P<0.01 and ***P<0.001. Atty. Dkt. No.: 093873-1450 FIG. 12 depicts exemplary plots, charts, and graphs showing that arousal indices track arousal levels in sham mice, in accordance with various example embodiments. Top: example trace of normalized cortical spectrogram (deviation from median) from sham mice emerging (subjects 1-5) after a short anesthetic exposure. Color bar shows power measured in dBs. Middle: detected movements from video during emergence: trunk twitch (gray), limb and body movement (blue), weak weight-bearing (wwb; dark green) including dragging and quivering, and organized movements (red). Bottom: automated detection of delta waves and burst suppression (black rhomboids) in sham mice and arousal index plot displaying the trajectory of each individual while restoring a wakeful state. FIGS. 13A and 13B depict exemplary plots, charts, and graphs showing arousal indices track arousal levels in brain-injured mice, in accordance with various example embodiments. Top: example trace of normalized cortical spectrogram (deviation from median) from mice exposed to blast injury (subjects 1-10). Color bar shows power measured in dBs. Middle: detected movements from video during emergence: trunk twitch (gray), limb and body movement (blue), weak weight-bearing (wwb; dark green) including dragging and quivering, and organized movements (red). Bottom: automated detection of delta waves and burst suppression (black rhomboids) and arousal index plot displaying the trajectory of each individual while restoring a wakeful state. FIGS. 14A-14G are exemplary plots, charts, and graphs showing that an arousal index measures arousal levels in humans, in accordance with various example embodiments. FIGS. 14A and 14B depict a normalized spectrogram (deviation from median) calculated from EEGs in C3 & C4 electrodes (primary motor cortex) of two neonates monitored to establish signs of encephalopathy secondary to hypoxia. Color bars represent power measured in decibels (dB). Bottom: motor recovery extracted from video. Movement was classified as jerky and repetitive movements (dark blue lines), organized movements including full flexion and extension of limbs and body arching (dark green lines), and movements performed while the individual was fully awake (eyes were widely and constantly open, active crying and coordinated extension and flexion of limbs, red lines). FIGS. 14C and 14D show, at top, an automated estimation of arousal index per arousal unit (orange lines), and at bottom, time points signaling behavioral responses to stimuli (dark purple) and unresponsive (light purple). Black crosses indicate timepoints of low oxygen saturation <90% and the black arrow Atty. Dkt. No.: 093873-1450 represents an interval in which one of the neonates received multiple sensory stimuli due to an ultrasound procedure. FIG. 14E shows neonate raw EEG data displaying the cortical switch (top trace; dashed pink rectangle) and the burst component (dashed green box) of the arousal unit (red box) linked to movement (increased signal amplitude after detrending the signal; black arrow) together with changes in breathing frequency detected with impedance pneumography (bottom trace). Gray crosses indicate the timepoints of breaths. FIG. 14F is a bar plot showing number of breaths before (gray) and after the cortical switch (brown) in each subject (detected rhythmic changes representing breaths in subject 1, n=90 and subject 2, n=68). One-tailed paired sample Wilcoxon signed-rank test. FIG.14G shows movement strength was indirectly measured in neonates using the amplitude changes of the pneumograph signal before the burst (gray) and at burst onset (brown). One-tailed paired sample Wilcoxon signed-rank test (subject 1, n=97 and subject 2, n=135 arousal units). Data shown as mean ± s.e.m. ***P<0.001. FIGS. 15A-15D are exemplary plots, charts, and graphs showing that arousal indices track arousal levels in elderly patients emerging from anesthesia, in accordance with various example embodiments. FIGS.15A and 15B show normalized cortical spectrogram (deviation from the median) of EEGs measured from FC1&FC2 electrodes in senior individuals emerging from general anesthesia during elective surgery. Color bars represent power measured in decibels (dB), and black rhomboids represent acoustic noise. Bottom: movement was documented by a practicing anesthesiologist with a clinical assistant during the emergence from general anesthesia. FIGS. 15C and 15D show an automated calculation of arousal index in subjects 3 and 4 while emerging from general anesthesia. Orange lines represent arousal units. FIGS. 16A and 16B are exemplary plots, charts, and graphs showing human data inputs to an arousal unit detection algorithm, in accordance with various example embodiments. FIG. 16A shows inputs to the algorithm were preprocessed bilateral frontal association area (FAA) EEGs recorded within an epoch (≈30 minutes). FIG. 16B, at top left, shows EEGs were processed using a sliding window with duration=5s and step size=1s. At top middle, the sliding window is annotated if it contains inter-channel artifacts. Inter-channel artifacts refer to artifacts occurring in both channels. At top right, the spectrum for each channel was calculated by measuring power at 100 log-spaced frequencies between 2-150Hz Atty. Dkt. No.: 093873-1450 and normalized spectra by order of magnitude. The channel containing artifacts was removed when detecting excessive delta or gamma power compared to the other channel. At bottom left, spectra of the remaining channels were averaged. At bottom middle, the spectral variables were computed from the combined spectrum. At bottom right, the spectral limits were estimated from spectral variables over sliding windows within the epoch. Windows containing intra-channel artifacts, were excluded. FIGS. 17A-17C are exemplary plots, charts, and graphs showing theta frequency calculation, in accordance with various example embodiments. Example traces of calculating theta frequency from log-spaced spectrum when theta power (15-35 in log space) was prominently higher, shown in FIG. 17A; lower, shown in FIG. 17B; and significantly lower than the nearby delta power, shown in FIG. 17C. The 100 log-spaced spectrum were denotedas ^^ ൌ ^^^^,^^ଶ … ,^^^^^^. The theta frequency is the frequency that occupies power within thetheta band. To calculate theta frequency, a right-tailed Z-test was applied to datapoints^^^ହ,^^^^, … ^^ଷହ that represent power in the theta band (gray rectangles). For data points thatrejected the null hypothesis at significance level=0.1, their frequency indexes wereconstituted into a set (red dots). The null distribution was estimated from ^^^ହ,^^^^, … ^^ଷହ. Thehighest frequency within this set was detected as theta frequency (black line). FIGS. 18A-18D are exemplary plots showing filtered frequencies to reduce noise in EEG human data, in accordance with various example embodiments. The spectra in four subjects from 500-second EEG signals were calculated. The montage (upper right corner) displays the placement of the electrodes. The text in parenthesis specifies the frontal association areas where electrodes are located, including the prefrontal cortex, supplementary motor area, and primary motor cortex. In FIGS. 18A and 18B, black lines represent notch widths when implementing a notch filter to suppress the 60 Hz power line artifacts in subjects 1-2. FIG. 19 is a block diagram of a system for generating a predictive biomarker indicative of a level of arousal of a subject, in accordance with various example embodiments. FIG. 20 is a flow chart of an algorithm for determining a predictive biomarker for disorders of consciousness, in accordance with various example embodiments. Atty. Dkt. No.: 093873-1450 FIG. 21 is a flow chart of a method for training a model for the characterization of a fluctuation in an arousal level of a subject, in accordance with various example embodiments. Fig. 22 is a flow chart of a method of generating a characterization of a fluctuation in an arousal level of a subject, in accordance with various example embodiments. DETAILED DESCRIPTION Before turning to the figures, which illustrate certain example embodiments in detail, it should be understood that the present disclosure is not limited to the details or methodology set forth in the description or illustrated in the figures. It should also be understood that the terminology used herein is for the purpose of description only and should not be regarded as limiting. Recovery to a conscious state following coma (or a disorder of consciousness) begins with cortical desynchronization and mild movements, transitions to a state in which the individual shows receptivity to stimuli (arousal) with purposeful movements, and eventually becomes aware, following commands and having functional motor execution and communication. However, severe brain injury, delayed recovery from anesthesia, and more recently, Covid-19 infections can obscure this process. Accurate diagnostic tools are critical, both to monitor recovery as well as to predict functional outcomes. Nonetheless, early clinical assessment of patients is difficult because it relies on clinical findings obtained post- injury, and presents an acute time frame with drastic ongoing metabolic and functional changes. Further, while clinical assessments of behavior in altered conscious states can be supported by clinical neuroimaging such as functional MRI (fMRI), electroencephalography (EEG), imaging, and behavioral diagnostic scales, these measurements are often assessed separately and sporadically in the clinical setting, limiting their utility and precision due to the fluctuating course of arousal levels. Therefore, a reliable and continuously tracked biomarker that reflects recovery is in high demand. Restoration of full mobility is associated with specific patterns within increased cortical activity in cingulate, prelimbic, and motor cortical areas, despite exposure to high anesthetic concentrations. This suggests that key aspects of cortical dynamics are critical to regain movement. Similarly, animals emerging from diverse coma-like states share a common dynamic process of cortical and motor arousal that can be consistently sequenced Atty. Dkt. No.: 093873-1450 from low to high arousal levels. There are five cortical periods characterized by increased cortical activity that track restored motor behavior in a hypoglycemic coma and a range of anesthetics, whether inhaled or injected, alongside conventional righting reflex assays. Accordingly, restoring waking is a generalizable, progressive process in which cortical patterns contain features consistent with the initiation of defined behaviors suitable to establish metrics of consciousness. In some implementations of the present disclosure, EEG activity was recorded from the frontal association area (FAA). The FAA is an area in rodents that is active during motor planning-related neural activities. Its activation occurs prior to the activation of motor cortical pyramidal neurons and precedes muscle contraction, suggesting it reflects the intention to initiate a motor sequence. While recording from the FAA, the close relationship between cortical periods and dominant motor behaviors seen before while restoring a wakeful state was shown. The FAA-EEG activity during each motor behavior, demonstrated cortical patterns that consist of interactions between three spectral variables: fractional gamma, delta, and gamma power that can be sequentially identified within specific windows and that are linked to breathing frequency and behavioral changes. The combination of these components is defined as arousal units (AUs). In other implementations of the present disclosure, the ability of AUs derived from EEG activity from the supplementary motor area (SMA) to quantitatively describe and discriminate arousal levels in encephalopathic neonates and seniors emerging from anesthesia was shown in humans. As shown in FIG. 1, pairings of z transformed theta and gamma power obtained from each AU can be fitted to a logistic curve converting the arousal unit into an arousal score that can be used as a tool to discriminate and quantitatively measure the level of arousal in both rodent and human unconscious states. This novel approach integrates specific measures that constitute the arousal process and may improve monitoring progression and establish coma recovery trajectories. Arousal units represent cortical patterns linked to behavioral and autonomic changes In some implementations of the present disclosure, the relationship between cortical periods (Table 1) and dominant behaviors while restoring a wakeful state was shown to occur in the frontal association area (FAA). Bilateral EEG activity was wirelessly recorded and Atty. Dkt. No.: 093873-1450 motor recovery was visually tracked using a video camera synchronized with the DSI software in mice exposed to isoflurane anesthesia ramps. As shown in FIG. 2A, recordings began at a concentration of 1.50% (>1 MAC) and gradually reduced anesthetic concentration (0.25% steps) every half hour until reaching 0% isoflurane. Then, to visualize the cortical dynamics, the spectrogram of EEGs was calculated, as shown in FIG. 2B. To define cortical periods linked to co-occurring predominant motor behaviors (Table 1), dominant frequencies were extracted from the spectrogram, as shown in FIG. 2C. These frequencies were clustered in the dominant frequency ranges at each time instant and included: 3-5 Hz (black), 10-20 Hz (light blue), 20-40 Hz (dark blue), 30-100 Hz (purple), and 70-130 Hz (lilac). As shown in FIG. 2C, the density of these dominant frequency ranges over time was segmented to obtain cortical periods. Cortical periods (P1-P5) can vary in length and order in FAA, as shown in FIG. 2C, and were progressively associated with co-occurring behaviors, starting with twitches, followed by limb and generalized body movements, and then weak weight-bearing movements and organized movements, as shown in FIG. 2D. Thus, cortico-motor regimens are also distinguishable in motor planning areas. TABLE 1: Regimens of cortico-motor integration observed during emergence from multiple pharmacologically induced coma models. As shown in FIGS. 3A and 4A, activation of cortical motor areas was directly linked to single motor behaviors. The EEG activity during emergence from anesthesia, see FIGS. 3A and 4A, was examined to uncover changes in cortical activity surrounding each motor behavior. As shown in FIG. 4B, the normalized spectrum at 100 log-spaced frequencies from 2-150 Hz for each 5-second window (step size 1s) was calculated using the mtspectrumc Atty. Dkt. No.: 093873-1450 function in the Chronux Toolbox in Matlab (Mathworks). Next, as shown in FIGS.3B and 4B, this window was moved through 15-minute epochs continuously segmented from a total of ~3.5 hours of recordings per animal (e.g., the long anesthesia ramp shown in FIG.3A). Since the accuracy of detecting characterized signals is higher when bilateral EEG signals are used, the spectra from bilateral channels were averaged into one, as shown in FIG.4B. Also, given that gamma oscillations have been closely linked to cortical activation during movement preparation and movement initiation, this band was analyzed first. Fractional gamma ^^௧was calculated which represents the power of the gamma band as a percentage of the total power (fractional power) in each window. This is a spectral variable often associated with cortical excitation in generalized arousal, which is able to integrate events among separated cortical areas. Next, as shown in FIG.3C, the calculated fractional gamma wasconcatenated into a time series ^^^,^^ଶ, … ^^். As shown in FIG. 3C, periods of ongoing EEG(“spans”) with increased fractional gamma power were detected. As shown in FIG.3C, within these EEG spans (denoted t1-t2), and preceding each movement, a sudden reduction in delta power was found. This change is defined as a “cortical switch.” To detect the cortical switch across the detected spans, the delta power of each window within the established spanwas concatenated into a time series … , ^^௧మ. As shown in FIG. 3C, a sub-span with areduced delta power was detected. As shown in FIG.3C, a significant increase in breathing frequency aligned to each cortical switch and preceding movement. Moreover, within these EEG spans, a peak in total gamma power or gamma burst directly linked to the movement, inwhich bursts of gamma were detected in the concatenated gamma time series ^^௧య , ^^௧యା^, … ^^௧ర,see FIG.3C. Thus, collectively, these observed cortical dynamic patterns represent the interaction of cortical, autonomic, and behavioral changes. The combination of these patterns is defined as arousal units, see FIG.3C. Therefore, arousal units were found in periods of increased fractional gamma in the EEG which contains: 1) a single cortical switch (reduced delta power) preceding movement, 2) a gamma burst linked to movement and 3) increased breathing frequency aligned to cortical changes preceding movement, see FIG.3. Since arousal units were reproducible and were commonly detected among the analyzed EEG data spans (approximately 40 per animal), an automated detection algorithm was generated for these spectral variables that detects windows with arousal units suitable for analysis. Arousal units represent arousal events across cortical periods that predict movement Atty. Dkt. No.: 093873-1450 In some implementations of the present disclosure, as shown in FIGS. 5A and 5B, the automated detection algorithm was applied to eight mice emerging from long ramps of isoflurane anesthesia to demonstrate that detected arousal units represent arousal events across the estimated cortical periods. Given the features of the arousal units, no arousal units were expected in Period 1 (P1), when movement is absent and low brain oscillations predominate (Table 1). As expected, and shown in FIG. 6, no arousal units were detected. As shown in FIG 6., arousal units first arose during period 2 (P2). Then, they increased in number from P3 to P5. These units were linked to autonomic and behavioral changes. As shown in FIG. 6B, within each period, the number of breaths per minute (BPM) increased when comparing the breathing frequency before and after the cortical switch (P2 before108.74േ7.81 vs. after 130.01േ6.86 BPM (P= 1.92ൈ 10ିସ); P3 before 111.23േ2.82 vs. after154.18േ5.40 BPM (P=1.79ൈ 10ି^^); P4 before 98.33േ3.65 vs. after 126.48േ4.52 BPM(P=6.70ൈ 10ି^; n=6 animals). As shown in FIG. 6C, motor behavior increased whencomparing movement before the burst and at burst onset (P2 before 0.3േ0.1 vs. burst onset2.9േ1.7 (P=0.02); P3 before 5.8േ1.0 vs. burst onset 45.8േ4.3 (P=9.1ൈ 10ି^^); P4 before10.3േ1.7 vs. burst onset 53.2േ5.9 (P=1.3ൈ 10ି^). As shown in FIGS. 6D and 6E, arousalunits as a predictor of movement within 15-minute-assessment epochs yielded an accuracy (percentage of arousal units truly associated with movements observed in 15-minute epochs) across the periods of ~80% and a sensitivity for dominant behaviors of ~70% per period (accuracy: P20.66±0.12, P30.91±0.04, P40.86±0.04, P50.95±0.01; sensitivity: P2-trunk twitches (0.67±0.13), P3-limbs movements (0.86±0.04), P4-wwb movements 0.72±0.07, P5- organized movements (0.73±0.05). In contrast, sub-spans between arousal units were not related to movement, and showed low accuracy: P20.07±0.04, P30.20±0.04, P40.40±0.05, P50.25±0.14). Therefore, these data suggest that arousal units represent arousal events during emergence from a comatose-like state that predict movement. Gamma-theta patterns within arousal units quantitatively discriminate arousal levels In some implementations of the present disclosure, features and power of multiple frequency bands within arousal units belonging to each cortical period were examined to identify whether arousal units in each cortical period differ in oscillatory properties. As shown in FIG. 7A, when the spectra of arousal units of each period was averaged, the power Atty. Dkt. No.: 093873-1450 of the theta band predominated across periods, and its frequency peak shifted to higher frequencies from P2 to P5. Given that theta rhythms form a temporal structure that organizes gamma activity, it was determined that shifts in the theta frequency peak occurred at the burst of each arousal unit along the periods. Peak theta frequency shifts were significantly higher at burst compared to the cortical switch, suggesting a direct effect of theta on gamma linked to motor activity. As shown in FIG. 7B, this event was evident across arousal units found at periods P3-P5 when movements become more organized: P3100 log-spaced theta frequency23.50±0.34 vs. 24.26±0.30 (P=0.004); P423.47±0.29 vs. 25.25±0.31 (P=5.41ൈ 10ି଼); P525.13±0.37 vs. 27.79±0.38 (P=5.47ൈ 10ି^^. In contrast, the P2 period did not showsignificant differences (P219.43±0.40 vs. 20.00±0.42; P=0.09). Unlike the peak theta frequency measured at the burst, the peak theta frequency measured at the cortical switch remained stable from P3 through P4: P324.26±0.30 vs. P4 25.25±0.31 (P=0.6). As shown in FIGS. 7C and 7D, this peak theta-frequency shift at the arousal unit burst corresponded to a monotonic increase in the gamma power extracted from the burst of each arousal unit across the cortical periods in each subject. Theta frequency in 100 log-spaced frequencies: P220.05±0.86 vs. P323.87±0.95 (P=0.029); P323.87±0.95 vs. P425.45±0.90 (P=0.011); P425.45±0.90 vs. P527.70±0.54 (P=0.029). Gamma power: P2 0.80± 0.22 dB vs. P32.22± 0.24 dB (P=0.029); P32.22± 0.24 dB vs. P43.93± 0.19 dB (P=0.011); P43.93± 0.19 dB vs. 5.39±0.34 dB; (P=0.018). Taken together, this data reveals that gamma-theta oscillatory patterns within arousal units significantly changed as cortical periods evolved towards a wakeful state. Since gamma-theta oscillatory patterns within arousal units are distinguishable across cortical periods, an automated approach was designed to transform these oscillatory patterns collected from eight subjects into a quantitative measurement of arousal levels, termed the arousal index. In the learning phase, log-spaced theta frequency and gamma power from arousal units per period were measured and centroids were derived by averaging gamma- theta per period, as shown in FIG. 7E. Then, to allocate gamma-theta changes in the same scale, centroids were normalized using Z-scores to the mean and standard deviation of the maximal arousal level period (P5), as shown in FIG. 7F, and each Z-scored centroids (gamma & theta) was transformed to a Z-scored scaler, as shown in FIG. 7G. Next, to establish the level of arousal paired with each arousal unit, these data were fit to a logistic curve. The Atty. Dkt. No.: 093873-1450 logistic curve was fit such that the P2 centroid and P5 centroid equaled 0.1 and 0.9, respectively, as shown in FIG. 7H. During the tracking phase, the established steps in the learning phase, and the pre-fit logistic function were applied to each individual while emerging from a comatose state, as shown in FIGS. 8A-8E. As shown in FIG. 8E, the arousal level of the subject as a function of time was tracked by concatenating arousal indices according to the chronological order of arousal units’ appearance. As a result, it was verified that the arousal index quantified and continuously monitored arousal levels when mice emerged from anesthesia, see FIGS. 5A and 5B, including detecting fluctuations within periods and transitions from one period to another, as shown in FIG. 8E. Arousal indices track arousal levels during recovery from hypoglycemic coma In some implementations of the present disclosure, the automated algorithm was used in hypoglycemic coma (HC) mice to examine their recovery trajectory. Mice overdosed with insulin resulted in a HC, in which EEG activity was significantly suppressed and mice lost their righting reflex, but spontaneous breathing was preserved, see FIGS. 9A, 10A, and 10B. Arousal units were detected using the automated algorithm. As shown in FIG. 9B, arousal units were preserved and displayed the typical cortical switch-burst structure found in anesthetized mice. As shown in FIGS. 10A and 10B, indices in mice spontaneously emerging from this comatose state were calculated and those injected with glucagon to pharmacologically reverse the hypoglycemic state. As in anesthesia, and shown in FIG. 9D, cortical changes were linked to increased breathing rate when comparing breaths per minute before and after the cortical switch (P2 before 90.4±7.2 vs. after 122.5±9.0 (P=0.008); P3 before 98.7±4.9 vs. after 122.8.0±8.2 (P=0.003); P4 before 110.79±1.2 vs. after 130.9±3.5. (P=1.80×10-6)). Similarly, gamma bursts aligned with movement onset, as shown in FIG. 9B, and, as shown in FIG. 9E, motor behavior increased when comparing movement before and at burst onset (P2 before 1.0േ0.0 vs. burst onset 4.0±0.0 (P=0.04; n=9); P3 before 3.0±0.00 vs. burst onset 15.0േ0.03 (P=2.64×10-8; n=49); P4 before 2.0±0.0 vs. burst onset 17.00±0.02 (P=4.91×10-10, n=55)). Moreover, as shown in FIG. 9F, arousal units yielded an accuracy of approximately ~89% (P20.83±0.07, P30.96±0.01, P40.85±0.03, P50.98±0.01) and, as shown in FIG. 9G, a sensitivity for predicting dominant movements linked to the cortical period ~75% (P3-limb movements 0.85±0.03, P4-wwb movements 0.81±0.04, P5-organized movements 0.66±0.05). Note that the sensitivity in Period 2-trunk twitches (0.16±0.06), was Atty. Dkt. No.: 093873-1450 particularly low compared to the P2 sensitivity found in anesthetized mice, see FIG. 6E). Interestingly, and as shown in FIG. 9H, pharmacologically-induced emergence using glucagon provoked an instant rise of the arousal index after injection (before injection 0.06±0.01 vs. 2000 s after 0.20±0.00). As shown in FIGS. 9H and 9I, accompanying this change, the cortical period quickly transitioned from P2 to P3 (35.98± 1.5 minutes post- injection). In contrast, as shown in FIG. 9I, spontaneous recovery showed a prolonged transition (51.77± 19.4 minutes). Although, as shown in FIG. 9H, both groups reached a similar comatose state after insulin overdose (arousal index between 0-30 minutes post HC; p>0.05, two-tailed Mann-Whitney test), the arousal indices in mice injected with glucagon were higher between 90-137 minutes (P<0.05, one-tailed Mann-Whitney test). Likewise, as shown in FIG. 9I, there was a significantly faster transition from P2 to P4 (spontaneous 176.30±41.44 min vs. glucagon 91.72±5.05 min; P=0.03, one-tailed Mann-Whitney test). Interestingly, even after pharmacologically reversing the hypoglycemic state, arousal indices were lower than those observed during emergence from anesthesia. This demonstrates that the established arousal index could quantify arousal levels in mice restoring to a wakeful state from HC and tracked the effect of medication on recovering from this comatose state. Arousal index effectively monitors arousal levels in brain injury In some implementations of the present disclosure, a secondary blast injury (SBI) model that causes long-term cognitive and motor sequelae was used to identify arousal units and measure levels of arousal in mice acutely exposed to SBI exhibiting axotomized neurons within minutes post injury. A secondary blast was inflicted in mice briefly anesthetized with isoflurane and quickly transferred (~10s) to monitor EEG and electromyographic (EMG) activity in their home cages, as shown in FIGS. 11A and 11B. Unlike control subjects, see FIG. 11C, mice exposed to trauma, as shown in FIG. 11D, initially showed a burst suppression state (SBI 2.91± 1.04 min) and intermittent delta wave oscillations during recovery of a wakeful state (sham 1.05±0.32 min & SBI 6.03±1.74 min; P=0.005). The automated detection algorithm was applied to extract arousal units and calculated arousal indices to monitor the emergence to a wakeful state in sham mice, shown in FIGS. 11E and 12, and after SBI, shown in FIGS. 11F, 13A, and 13B. Atty. Dkt. No.: 093873-1450 Overall, and as shown in FIG. 11G, mice exposed to SBI showed increased breathing frequency together with each arousal unit measured while they returned to a wakeful state (before 178.89±5.6 vs. after the cortical switch 214.3±9.65; P=0.004). In addition, as shown in FIG. 11H, the proportion of body parts moving increased at burst onset as SBI miceprogressively recovered (before 2.7േ8 % vs. at burst onset 24.2േ5.4%; P=1.72ൈ 10ିସ). Inbrain-injured animals, arousal units yielded an accuracy of ~ 87%, shown in FIG. 11I, and a sensitivity for dominant movements of ~76% (P3-limb mov. 0.78±0.07, P4-wwb mov. 0.82±0.07 and organized mov. 0.90±0.05), shown in FIG. 11J. Compared to sham mice arousal indices during recovery of a wakeful state after trauma were notably more heterogeneous with profound fluctuations and lower indices, see FIG. 11L, than sham mice, see FIG. 11K. Altogether, these data indicate that arousal units were preserved after brain injury and quantitatively outline individual recovery trajectories while restoring awakening. Arousal indices quantitatively measure and track arousal levels In some implementations of the present disclosure, arousal units were detected in human EEG data. EEG data was previously collected from patients admitted to the neonatal intensive care unit (NICU) at the New York-Presbyterian Hospital. Compared to the rodent datasets, clinical recordings were significantly contaminated with artifacts like line power noise and occasional pollution (e.g., muscle contractions) that appeared across all or some channels. The first type of artifact prevented visualization of the arousal unit, while the second confounded its detection and quantification. Thus, raw EEGs were preprocessed to detrend and remove intra- and inter-channel artifacts before using the automated detection algorithm. For this analysis, C3 and C4 channels (motor cortex area) were chosen. These channels showed fewer artifacts than the temporal and frontal channels. The accompanying video recordings included neonates’ spontaneous activities and reactions to the environment while being monitored for several hours at the NICU. As shown in FIGS. 14A and 14B, a normalized spectrogram was calculated for the recorded time. As shown in FIGS. 14A and 14B, movements were manually identified and classified from the provided video as jerky and repetitive movements, organized movements (full flexion and extension of limbs, body arching), and awake (eyes were widely and constantly open, or active crying). The automated adapted algorithm was applied to the preprocessed EEG data to extract arousal units. As shown in FIGS. 14C and 14D, the arousal index was also calculated as a function of time. Atty. Dkt. No.: 093873-1450 The results indicated that both individuals displayed an arousal index lower than 0.5 during a large proportion of the recorded time for subject 1 (61.11%) and subject 2 (68.94%), suggesting a predominant low arousal state. At time points in which indices were particularly low (<0.2), other variables were examined that may instigate this state, and (when data were available) periods of low oxygen saturation (SpO2<90%), see FIG. 14D, were uncovered. These results contrasted with short events, in which the arousal index went above 0.5. In those cases, the neonates were awake, shown in FIGS. 14A and 14B, reactive and responsive to stimuli caused by NICU personnel or parents’ manipulation, show in FIGS. 14C and 14D. As in the rodent data, arousal units extracted from humans, shown in FIG. 14E, aligned with increased breathing frequency, as shown in FIGS. 14E and 14F, (subject 1: before 71.10±1.70 vs. after cortical switch 80.25.±1.53; P=6.33×10-6& subject 2 before 51.48±1.91 vs. 71.62±2.13, P=3.40×10-9) and changes in motor behavior were linked to burst, as shown in FIGS. 14E and 14G (subject 1: before 0.017.0±0.000 mV vs. at burst onset 0.082±0.006 mV, P=6.18×10-18; subject 2: before 0.010±0.000 mV vs. at burst onset 0.035±0.002 mV, P=3.39×10-24). Therefore, the arousal index granularly captured arousal fluctuations in neonates ranging from low states (desaturation states) to high arousal states (response to manipulation). Considering that EEG activity varies in complexity across the lifespan, as the elderly show multiple changes in EEG activity compared to younger individuals, it was assessed whether arousal units were detected and quantifiable in an older population. Deidentified EEG data from patients emerging from general anesthesia after completing elective surgical procedures was used to assess the presence of arousal units in this population. As shown in FIGS. 15A and 15B, a spectrogram was computed, including the time the anesthetic was withdrawn to when the patient reached a wakeful state. As shown in FIGS. 15A and 15B, movements were identified and documented by a practicing anesthesiologist with a clinical assistant, as video monitoring of patients under anesthesia was not permitted under the Health Insurance Portability and Accountability Act (HIPAA). EEG signals were extracted from FC1 and FC2 channels. These channels were the closest to the SMA and showed the least artifact contamination. Once arousal units were detected, an arousal index was calculated. Once again, the data was detrended and the power line noise removed. Then, as shown in FIGS. 15C and 15D, the preprocessed EEG traces were fed to the automated algorithm to Atty. Dkt. No.: 093873-1450 extract arousal units and calculate indices. When arousal indices were close to zero, subjects were in an unconscious state and did not move. Then, the index quickly increased and was linked first to throat and face movements before transitioning to arm movements. Once the arousal index was higher than 0.5, the individuals opened their eyes and responded to verbal commands. Thus, arousal units are detectable in human EEG data and can discriminate between different trajectories. More importantly, this quantitative method of assessing arousal can translate to clinical care. DISCUSSION OF EXAMPLE EMBODIMENTS Arousal units (AU) identified from cortical EEG patterns represent the interaction of cortical, autonomic, and behavioral changes (see, e.g., FIG. 1) that lawfully grade across phases of recovery from coma in multiple paradigms. AUs can be used as a novel tool to discriminate and quantitatively measure the level of arousal across conditions and species. By transforming the intrinsic gamma-theta patterns from AU into continuous values through a logistic function (arousal indices), a continuous and quantitative method is provided to measure arousal levels in states with impaired consciousness, such as recovery from anesthesia, hypoglycemic coma, and brain injury in human and non-human subjects. This measure can be used to study, for example, coma recovery in preclinical and clinical models and guide clinical decision-making for accurate diagnoses and therapeutic solutions. Unlike other biomarkers, in various embodiments of the present disclosure, AUs are measurable in several states of awareness in rodents and humans, and their features may be independent of the etiology of altered consciousness. Aligning cortical, autonomic, and behavioral signals indicates a concurrent activation of the arousal circuits that elicit varying somatic and autonomic responses. Contrary to stacked multimodal features to measure arousal, AU demonstrates that arousal is an evolving process rather than a single event. Notably, the evolution of the intrinsic features of AU as a function of the cortical periods predicts behavioral responses. Indeed, the high-frequency dynamics were mainly driven by gamma power, and its increase was directly associated with the execution of more organized movements. This aligns with current perspectives demonstrating the critical role of gamma oscillations in the integration of cortical events, planning and motor execution. In contrast to this view, gamma power changes alone are insufficient to distinguish different arousal levels, Atty. Dkt. No.: 093873-1450 including awareness. Intriguingly, theta frequency within the arousal units shifted to higher frequencies as subjects emerged into a wakeful state. However, the total theta power did not change. This finding supports the idea that a shift in theta to higher frequencies, but not the power of the theta band, is an indicator of high cognitive task demands. Interestingly, this shift mainly occurred at the burst, but not the cortical switch, indicating cross-frequency coupling dynamics between theta and gamma oscillations. Increased theta-gamma phase- amplitude coupling has been associated with an increment in motor performance. Modulating gamma activity by theta oscillations increases movement and motor skill acquisition in humans and rodents, suggesting that theta rhythm enables better processing and transmission of neural computations. Here, gamma-theta patterns were captured from the FAA in rodents and SMA in humans. These are planning-associated premotor areas that display planning-related neural activities and are implicated in goal-directed movements across species, distinct from the primary motor cortex. Indeed, the FAA and SMA are active during the mental rehearsal of a motor-sequence test but are inactive during simple repetitive actions or somatosensory discrimination, suggesting its involvement in the intention of a voluntary act. Since AUs were identified in FAA and SMA, an AU precedes voluntary movement. Also, their sensitivity for organized motor behavior (Period 3-5) was significantly higher (>80%) than for short, erratic and repetitive behavior (Period 2 <60%). Thus, AUs can potentially detect the execution of voluntary behaviors rather than reflexes and are likely to be present even if an individual is unable to move. Identification of AUs may be critical to reducing the misclassification rate of covert awareness. Regarding the respiratory rate changes linked to the AU, multiple studies in humans and rodents have shown that respiration influences neuronal excitability and engages widespread circuits to achieve an active behavior. A recent report suggests that brainstem ascending circuits coordinate through the breathing rhythm neural firing and network dynamics across cortical and subcortical networks. The detected change in breathing within the AU may act as an oscillatory pacemaker that promotes increased neuronal excitability across multiple neuronal circuits to overcome neural inertia, resulting in motor behavior during the transition from a comatose to a wakeful state. Atty. Dkt. No.: 093873-1450 Using anesthesia as the basis for discovering AUs is valid, as cortical patterns are transferable to other conditions. Similarly, brain injury and anesthesia domains have been used to improve the classification of EEG components during consciousness using deep learning. Moreover, since arousal indices are extracted from exclusive epochs expressing the integration of cortical desynchronization, increased breathing rate, and movement onset, arousal indices are more specific and interpretable than other methods. While others use a global search of cortical signatures through convolutional neural networks or evaluate features via cross-validation, the present disclosure focused on a local search of cortical features containing breathing frequency changes and movement as a visible output. The advantage of this approach exponentially reduced the search space of features, which facilitated the distinction of AUs from background oscillations and excluded factors such as subject heterogeneity and sample size, which can underpower this type of analysis. Several studies have attempted to index arousal levels by applying multiple approaches, including spectral dynamics, Lempel-Ziv complexity, spatiotemporal dynamics and EEG entropy derived from resting EEG. They have also derived biomarkers from EEG responses to stimulation that determine the state of consciousness. Although all these methods may discriminate altered and physiological steady states of consciousness, they cannot display the transition from one state to another in evolving near real-time data. Since arousal indices are generated from AUs emerging in minutes and correlated with single motor behavior, the method disclosed herein reveals arousal level fluctuations. It can also quantitatively track the transition from one state to another. This temporal resolution is unfeasible for current behavioral scales and functional studies due to labor-intensive implementation, costly evaluation, or fatigue after consecutive stimulation. Thus, AU and the derived indices may objectively inform prognosis, improve monitoring, and determine treatment efficacy. There are several clinical applications of arousal indices and AUs, in two primary domains. First, they may be valuable to monitor the level of consciousness in individuals with brain dysfunction or brain injury from trauma, hemorrhage, or hypoxia / ischemia (e.g., drowning, cardiac arrest, perinatal asphyxia). Deviations from typical trajectories of recovery could serve as an early warning of new injury (e.g., vasospasm in subarachnoid hemorrhage) or new comorbidities (e.g., early sepsis). Second, they may provide early indication of Atty. Dkt. No.: 093873-1450 recovery for individuals with persistent or fluctuating impairment of consciousness, and help clinicians prognosticate the eventual clinical outcome. Although current biomarkers and behavioral scales widely evaluate arousal levels, there is still a large proportion of patients in which arousal levels are underestimated. Cortical EEG patterns (AU) discriminate and quantitatively measure the level of arousal across conditions and species. These findings provide novel insights into understanding the arousal circuit dynamics linked to the recovery of a wakeful state and the potential of using these biomarkers to continuously quantify and track arousal levels. AU can improve monitoring progression, establish arousal fluctuations, and may provide actionable insights to clinicians of individuals with impairments of consciousness. MATERIALS AND METHODS Rodent Data All use of laboratory animals was consistent with the Guide for the Care and Use of Laboratory Animals and approved by the Weill Cornell IACUC (Protocol No.2016-0055 and 2016-0054). Ten to nineteen-week-old male and female C57BL / 6 mice (Jackson labs) were maintained on a reverse cycle, with food and water provided ad libitum. All experiments were achieved during the dark cycle (9:00 to 21:00 h). A total of 39 animals were used to perform the isoflurane long ramps(n=8), isoflurane short ramps(n=8), hypoglycemic coma(n=9), and traumatic brain injury experiments(n=14). In 22 animals, a random number generator using random.choices in Python was used such that each animal has a 2 / 3 probability of receiving trauma and 1 / 3 chance of being a sham. A higher probability was assigned for mice experiencing brain injury because of a higher mortality risk in this population. The rest of the animals were randomly assigned to either isoflurane long ramp or hypoglycemic coma via a coin toss. Eight mice were excluded because mice were euthanized or died after trauma (n=4) or hypoglycemic coma (n=1) and EEG recordings were noise contaminated (n=3). Human Data This disclosure includes New York-Presbyterian Hospital / Weill Cornell Medical College (WCMC) data. Four human subjects were included in this study. Patients 1(male) Atty. Dkt. No.: 093873-1450 and 2 (female) were newborns monitored to establish signs of encephalopathy secondary to hypoxia. Patients 3 and 4 were older individuals who underwent elective surgical procedures with general anesthesia. Patient 3 was a 68-year-old female who underwent a laparoscopic partial small bowel resection with a total anesthetic time of 2 hours and 1 minute, and patient 4 was a 69-year-old female who underwent a robotic ventral rectopexy with a total anesthetic time of 4 hours and 5 minutes. The analyzed data presented here correspond to EEG monitoring during the emergence phase from anesthesia. Deidentified EEG data that were originally collected for other purposes and were approved by the Institutional Review Board at Weill Cornell Medicine (IRB # 1801018908 and IRB # 0309006330) were used. Individuals provided informed consent prior to participating in the study. The shared data was used in accordance with the terms agreed to upon their receipt. Transmitter Implantation Mice included in the study were weighed and anesthetized using isoflurane (4%) in an induction chamber. Then, animals were transferred to a stereotaxic frame, and anesthetic concentration was maintained using a nose cone at 1.25%. The temperature during the surgery was maintained at approximately 37°C using a temperature regulator (CWE Inc). The transmitter was placed as described before. The first lead was placed in the frontal association area (AP: 2.6 mm; ML:-1 mm from bregma). The reference lead was placed within the posterior parietal region. More specifically, mice included in the study were weighed and anesthetized using isoflurane (4%) in an induction chamber. Then, animals were transferred to a stereotaxic frame, and anesthetic concentration was maintained using a nose cone at 1.25%. The temperature during the surgery was maintained at approximately 37°C using a temperature regulator (CWE Inc). Briefly, the skull was fixed to the stereotaxic frame, the surgical area was cleaned, and the wound area was infiltrated with local anesthetic (bupivacaine 0.5%). After opening an incision in the scalp, a subcutaneous pocket along the animal’s dorsal flank was opened and the body of the transmitter (model HD-X02; DSI) placed into the pocket. The first lead was placed in a craniotomy at stereotaxic coordinates targeting the frontal association area (AP: 2.6 mm; ML:-1 mm from bregma). The reference lead was placed within the posterior parietal region. The leads were secured using dental acrylic. Then, the Atty. Dkt. No.: 093873-1450 second lead was placed either at (AP: 2.6 mm; ML:+1 mm from bregma) or sutured on the trapezoid muscle in the animal’s neck when EMG monitoring was necessary. Animals received analgesics (Flunixin 5.0 mg / Kg) and saline postoperatively. Weight, temperature, and signs of distress were monitored for seven days. Monitoring of Animals During Anesthetic Ramps Animals were placed into a gastight chamber and induced with isoflurane (3%) one week after the transmitter was implanted. The anesthetic was delivered using a calibrated vaporizer in 50% air / 50% O2, and concentration was monitored using a gas analyzer (Riken Fi-I). EEG was recorded for 30 minutes at a concentration of 1.25% (~1 MAC) and gradually reduced the anesthetic at intervals of 0.25% every 30 minutes until reaching 0%. The animal’s temperature was maintained at approximately 36±05 °C. More specifically, animals were placed into a gastight chamber and induced with isoflurane (3%) one week after the transmitter was implanted. The anesthetic was delivered using a calibrated vaporizer in 50% air / 50% O2, and concentration was monitored using a gas analyzer (Riken Fi-I). EEG was recorded for 30 minutes at a concentration of 1.25% (~1 MAC) and gradually reduced the anesthetic at intervals of 0.25% every 30 minutes until reaching 0%. For short anesthetic ramps, animals were induced and kept for 15 minutes in 1.25% isoflurane. Then, the gas was turned off in a single step. The chamber was placed on a receiver, and each animal transmitted data from the implanted telemetry device to the Ponemah V6.5 software from Data Science International (DSI). This system wirelessly recorded EEG, activity counts, temperature, and video. EEG recordings were synchronized with video captured via a Noldus Media recorder linked to the DSI system. The animal’s temperature was maintained at approximately 36±05 °C using a temperature regulator (CWE Inc). EEG recordings were sampled at 1000 Hz. Secondary Blast Injury (SBI) An SBI mouse model was previously described that resulted in the degeneration of multiple areas, including the brainstem, midbrain, and forebrain regions. This animal model mimics injuries caused by flying debris generated by an explosion. Shortly, an air blast was generated using a 6-gallon portable electric air compressor attached to a nail gun (no nails were present). The air blast propelled a piston (shrapnel) toward an approximately 2x2 mm Atty. Dkt. No.: 093873-1450 brain area including the frontal / prefrontal cortical-pallidal thalamocortical loops. Before the impact, animals (n=14) were anesthetized and temperature was maintained at 36±05 °C using a CWE temperature controller. If distress was detected, humane treatment was provided. More specifically, an SBI mouse model was previously described that resulted in the degeneration of multiple areas, including the brainstem, midbrain, and forebrain regions. This animal model mimics injuries caused by flying debris generated by an explosion. Shortly, an air blast was generated using a 6-gallon portable electric air compressor attached to a nail gun (no nails were present). The air blast propelled a piston (shrapnel) toward an approximately 2x2 mm brain area including the frontal / prefrontal cortical-pallidal thalamocortical loops. Before the impact, animals (n=14) were exposed to isoflurane (1.50% vol), ophthalmic ointment was applied, and Flunixin (5.0 mg / Kg) was injected. The temperature of each animal was monitored and maintained at 36±05 °C using a CWE temperature controller. The head and half of the body was inserted into a plastic tube covered with molded foam. This tube was placed perpendicular to the SBI device toward the right parietal region of the mouse. A micro-switch sensor controlled by a motor stepper held the head in position. An air blast pressure of 50 psi was provided. Sham mice (n=8) experienced a similar procedure, except that the blast occurred several centimeters away from the head, and the piston was deflected by an object placed in the apparatus. Animals were quickly transferred (within 10 s) to the home cage for EEG / EMG monitoring. A space gel warmer was placed in the cage during recovery. After the induced-blast injury, animals underwent EEG / EMG monitoring, and recordings were sampled at 500 Hz for 20 minutes. Blast-injured mice re-righted themselves and restored a wakeful state within an average period of 8^0.76 min compared to 2.5^0.18 min in control mice. If distress was detected, humane treatment was provided following IACUC guidelines. Three control and four SBI mice were removed from the study because of transmitter lead detachments or sudden death after trauma. Hypoglycemic Coma A week after transmitter implantation, hypoglycemic coma was induced using an intraperitoneal injection of insulin (Humulin) diluted in sterile saline (n=9). Concentration was titrated according to body weight and mice were monitored for any epileptiform activity triggered during the hypoglycemic coma induction. None of the mice included in the study Atty. Dkt. No.: 093873-1450 showed abnormal epileptiform activity, and all mice maintained spontaneous breathing. In addition, to examine arousal levels during pharmacologically induced hypoglycemic coma recovery (n=4), the condition was reversed by injecting glucagon (0.5 mg / Kg). More specifically, a week after transmitter implantation, hypoglycemic coma was induced using an intraperitoneal injection of insulin (Humulin) diluted in sterile saline (n=9). Concentration was titrated according to body weight as described and mice were monitored for any epileptiform activity triggered during the hypoglycemic coma induction. None of the mice included in the study showed abnormal epileptiform activity, and all mice maintained spontaneous breathing. Recordings began once animals showed prominent slow oscillations / burst suppression state, loss of righting reflex, and were unresponsive to a noxious stimulus. Given the prolonged time for spontaneous coma recovery (n=5), EEG data were sampled at 1000 (n=4) and 500 Hz (n=1). In addition, to examine arousal levels during pharmacologically induced hypoglycemic coma recovery (n=4), the condition was reversed by injecting glucagon (0.5 mg / Kg). The EEG / EMG record started 30 minutes before the injection. Given the faster pace of recovery compared to spontaneous hypoglycemic coma recovery, EEG data were sampled at 500 Hz. Human EEG Recordings Data from neonates were recorded using the Cool-Cap System (Natus Medical Inc.), and brain activity was monitored using the CFM6000 system (Natus Medical Inc.) in the Neonatal Intensive Care Unit (NICU) at the New York-Presbyterian Hospital / Weill Cornell Medical College. Electrodes were placed over the frontotemporal regions for assessment and were maintained for several hours to establish signs of encephalopathy secondary to hypoxia. All channels were visually inspected to find and discard poor-quality signals due to electrode detachment or time points in which the EEG record was stopped. In adults, intraoperative EEG data were collected using a 32-channel actiCHamp system (Brain Vision) during elective surgical procedures at the New York-Presbyterian Hospital / Weill Cornell Medical College Only electrodes placed over frontotemporal regions (C3, C4, FC1, FC2) during emergence were included in analysis. Breathing Detection and Quantification Atty. Dkt. No.: 093873-1450 An automated approach was used to quantify breaths before and after the onset of arousal units using the videos recorded from animals emerging from comatose states. CV2, scikit-learn and SciPy packages in Python were used. Once arousal units were detected, a 30- second video was obtained by concatenating frames 10 seconds before to 20 seconds after the onset of the arousal unit. A region of interest (ROI) was manually chosen, using cv2.selectROI. ROI was centered between the thoracic and abdominal cavities to isolate the diaphragmatic muscle contraction observed in each breath. Next, ROI pixels were concatenated across frames and an ROI matrix was obtained. Pixels were transformed into grayscale using cv2.COLOR_BGR2GRAY. Then, the ROI matrix was normalized using normalize function in sklearn.preprocessing package and performed principal component analysis (PCA) transformation (number of components=3, sklearn.decomposition package) to extract the respiratory rhythmic component. More specifically, To detect breaths from the video, cv2, sklearn, and scipy packages in Python were leveraged. Detection roughly followed procedures that estimate video-based respiration signals in neonates or adults. Given an arousal unit, a 30-second video was obtained by concatenating frames 10 seconds before to 20 seconds after the onset of the arousal unit. The video was cropped with a region of interest (ROI) using cv2.selectROI function centered between the thoracic and abdominal cavities to isolate the diaphragmatic muscle contraction observed in each breath. For each frame within the cropped video, pixels were grayscale transformed using cv2.COLOR_BGR2GRAY and flattened into a vector. Vectors across frames were then concatenated into a ROI matrix. If the cropped video contained movement, concatenation was performed over frames prior to the movement. Next, ROI matrix were normalized using sklearn.preprocessing.normalize function and applied PCA transformation (number of components=3, sklearn.decomposition.PCA) to extract respiratory component. Respiratory component was identified as the PCA dimensionality that showed the clearest rhythm compared to the other two dimensionalities. Lastly, scipy.signal.find_peaks function were employed to detect peaks of the rhythm. Peaks with prominence>0.01 (scipy.signal.peak_prominences) represented breaths. Quantification of Motor Behavior Atty. Dkt. No.: 093873-1450 Motor behavior linked to the arousal units was automatically quantified from the video following similar steps as the breathing quantification. The amount of motion between frames detected in different body parts was used as an indirect measure of motor behavior. For this purpose, Python, OpenCV, and scikit-learn packages were used. Once the arousal units were detected, a 30-second video was obtained by concatenating frames 10 seconds before to 20 seconds after the onset of the arousal unit. Here, the video was cropped with a rectangular region of interest (ROI) covering the total animal’s body area. ROIs were chosen manually using the cv2.selectROI function. Next, frames in the cropped video were looped over to detect and quantify movement. The amount of movement was quantified by establishing a proportion between the area of pixels showing differences along the frames (the percentage of body area showing movement) and the total body area. Accuracy was defined as the percentage of arousal units truly associated with movement, whereas sensitivity was defined as the percentage of moving body parts associated with an arousal unit. The percentage of body area moving per frame was an indirect measure of movement calculated as the area of pixels where the movement was occurring divided by the area of pixels occupied by the animal. To quantify it, cv2 and sklearn packages in Python were leveraged. Given an arousal unit, a 30-second video was obtained by concatenating frames 10 seconds before to 20 seconds after the onset of the arousal unit. To identify the area of pixels occupied by the animal (denoted as ROI_animal), a rectangular region of interest exclusive for the animal was manually chosen in the first frame using cv2.selectROI function. Next, the video was cropped by ROI_animal to allow the detection of movement restricted to theanimal. For each frame ^^^^^, ^^ ൌ 1,2, …^^ within the cropped video (K represented totalnumber of frames), the following steps were performed. First, the motion difference between adjacent frames was calculated. Absolute difference between ^^^^^and its previous ^^^^^ି^was grayscale transformed using cv2.COLOR_BGR2GRAY function. The obtained image was denoted as img_diff. For videotapes in a dark environment, cv2.convertScaleAbs(α,β) was implemented to increase contrast of the frame(8) with α=1.2 and β=0 prior to grayscale. Atty. Dkt. No.: 093873-1450 Second, pixels are segmented. Pixels with significant changes in intensity values (denoted as pixel_seg) between adjacent frames were likely to indicate a body part moving. Binary thresholding cv2.threshold (threshold=20) was applied to img_diff to segment the frames. A threshold=20 detects subtle movement-like twitches while maintaining invulnerability to tiny variations between frames. The image after thresholding was denoted as img_thres. Since the movements in the hypoglycemic coma model were milder, the cv2.dilate function was applied to img_thres. This function used a morphological filter to expand the boundary of pixel_seg. Third, segmented pixel areas are approximated to rectangles. The area of img_thres was approximated with a set of rectangles by applying cv2.findContours and cv2.boundingRect. Each rectangle localized a body region that showed movement. Ratios between the area of each rectangle and the area of ROI_animal were summed up to obtain the percentage of body area moving per frame. Automated Quantification of Breaths and Strength of Movement in Hypoxic Newborns Given the insufficient video resolution, signals measured from the pneumography channel to quantify breaths and strength of movement associated with arousal units in hypoxic neonates were relied on. Both quantifications were done in Matlab. ^^rawdenotes the raw measurement (sampling frequency=256Hz) from the pneumography channel. x_"detrend" was derived after removing offset from ^^rawby applying locdetrend.m function in Chronux toolbox. A moving window=[1,0.5] seconds was used. Then, to measure the strength of movement, ^^ampwas calculated by taking the absolute of ^^detrend. To detect breaths, the envelope.m function was applied to compute the ^^envelopeand obtain the upper peak envelopes of ^^detrendwith minimal peak separation=50 datapoints. Next, peaks in ^^envelopewere identified by adopting findpeaks.m function. Peaks with empirical prominence=0.01 in subject 1 and 0.005 in subject 2 were defined as breaths. Sample Size To determine the sample sizes of experimental groups, pilot experiments were performed with three mice for each comatose model, including isoflurane long ramp, hypoglycemic coma, and blast brain injury. The strength of the effect and the variance across Atty. Dkt. No.: 093873-1450 groups were considered to determine this sample size (number of subjects). Sample sizes were estimated using power analysis to determine autonomic and motor changes associated with arousal units, Pearson’s correlation, and ANOVA analysis. Sample sizes exceeded the minimum sample size needed to establish statistical significance. Replication 2-3 replicates were generated for each experimental rodent model. The data for all replicates were pooled for statistical analysis. Blinding An investigator not involved in the experimental procedure blindly ran the algorithm for arousal unit detection and quantification of arousal levels using the arousal indices in rodent and human data. In addition, the quantification of breathing frequency and motor behavior was quantified by a trained researcher unaware of the experimental conditions in rodent and human data. Analyses The Lilliefors test was run to determine if the data were parametric in the following conditions: (1) statistical comparison within a period, including the number of breaths before and after the cortical switch (FIG. 6B, FIG. 9D, FIG. 11G, FIG. 14F). (2) the percentage of body area moving per frame before burst and at burst onset (FIG. 6C, FIG. 9E, FIG. 11H, FIG. 14G). (3) theta frequency at the cortical switch and burst (FIG. 7B). Since the data were nonparametric, the one-tailed Paired Sample Wilcoxon Signed Rank Test was applied. For the statistical analysis that related to comparison across periods, the two-tailed Kolmogorov-Smirnov Test was applied to group level comparison (FIG. 7B) and one-tailed Paired Sample Wilcoxon Signed Rank Test was applied to individual level comparison (FIGS. 7C and 7D). EXAMPLE AUTOMATED DETECTION ALGORITHMS OF AROUSAL UNITS Inputs to the Algorithm Inputs to the automated detection algorithm were the bilateral EEGs measured from frontal association areas within an assessment epoch. EEGs were either raw measurements Atty. Dkt. No.: 093873-1450 (rodent models) or after preprocessing (human data). Table 2 outlines the acquisition platform, sampling rate, and time length of assessment epoch in each animal and clinical model. In brief, the detection algorithm leveraged the interactions between three spectral variables: fractional gamma, delta, and gamma power. As a result, the algorithm can sequentially identify every component (cortical switch and burst) of the arousal unit within specific time windows. TABLE 2: Inputs to the automated detection algorithm. Calculation of Combined and Log-Spaced Spectrum Over EEG Channels Atty. Dkt. No.: 093873-1450 Raw (rodent) or preprocessed (human) two-channel bilateral EEGs measured from frontal association area were processed using a sliding window with length=5s and stepsize=1s. For each time window ^^ ൌ 1,2, … , ^^ and for each channel, its spectrum wascalculated by applying mtspectrumc.m function in Chronux toolbox in Matlab. As shown in FIGS. 16 and 17, power was measured in watts to characterize non-gamma frequencies. Parameters were specified as tapers = [3,5], pad=1 and frequency ranges were listed in Table 3. Next, the mean power of each 100 log-spaced frequency between 2 and 150 Hz wasmeasured and a vector ^^ ൌ ^^^^, ^^ଶ, … ^^^^^^ was obtained. Here, ^^^ ,  ^^ ൌ 1,2, … 100represented power of the ^^ th frequency. 100 log-spaced frequencies were chosen so that frequency bands of interest in linear-space fall into ranges of multiples of five in log-space to assist interpretation. For example, ^^^:^ହ(2-3.6Hz) represents delta band; ^^^ହ:ଷହ(3.6-8.6Hz) represents theta band, and ^^^^:^^^(39-150Hz) represents gamma band. Given the variability of EEGs across models, spectrum was divided by its magnitude for the purpose ofnormalization (Table 3). Next, for each sliding window ^^ ൌ 1,2, …^^ , spectra were averagedacross channels to obtain a single combined spectrum. For convenience, superscripts denoted time windows and subscripts denoted frequency indices. For example, representedfractional gamma at window ^^ and ^^^ହ represented 15th entry of spectrum ^^ ൌ ^^^^,^^ଶ, … ^^^^^൧.TABLE 3: Parameters to calculate combined spectrum Calculation of Spectral Variables Atty. Dkt. No.: 093873-1450 For each sliding window ^^ ൌ 1,2, …^^ within the assessment epoch, spectral variableswere calculated from the combined spectrum ^^ ൌ ^^^^, ^^ଶ, …^^^^^൧. Spectral variables includedfractional gamma ^^௧, delta ^^௧and gamma ^^௧power. Their formulas are shown in Table 4. To accommodate various gamma power-baseline across models, gamma offsets were introduced (measured in decibels to facilitate interpretation & denoted as “offset” in Table 4). To obtain gamma offset, midpoint gamma=^^଼ହ(in dB) was averaged over sliding windows within an epoch. Since the spectrum in the gamma band after dB transformation approached linearity, midpoint gamma could reflect shift in gamma power between epochs and between models. Midpoint gamma did not differ significantly across epochs under the same model. However, it differed significantly across models. It centered at four values: 20dB, 25dB, 30dB, and 35dB. Thus, model-dependent gamma offsets was chosen from these four values (Table 4).When the fractional gamma was calculated, the ratio mean^^^^^:^^^^ / mean^^^^:^^^^wasbrought one magnitude up to ease interpretation. To eliminate outliers or noise, gamma power was smoothed over time by a moving average filter (filter span=3). TABLE 4. Definitions of Spectral Variables and Spectral Limits Atty. Dkt. No.: 093873-1450 Spectral Limits For each epoch, spectral limits were estimated for segmenting time series in the next steps. Spectral limits were derived from spectral variables through all windows within the epoch (Table 4). When the delta limit was calculated (Table 4), an upper bound=1.5 (watt after normalization) was set, the maximum delta power considered a cortical switch. Detection of Spans with Increased Fractional Gama The fractional gamma of each window was concatenated into a time series^^^, ^^ଶ, … ^^். Here, ^^௧, ^^ ൌ 1,2, …^^ denoted the fractional gamma at window ^^ and ^^ denotedthe total number of windows within the epoch (Table 2). A span was initiated at time ^^^whenthe fractional gamma started to rise: ^^௧భ ^ ^^^^^ , ^^௧భ Here, ^^^^^represents the spectral limit defined in Table 4. The constant 0.1 is an empirical value representing a lift in the fractional gamma series in potential arousal units. This spanwas terminated at ^^ ௧ଶ when fractional gamma dropped: ^^ మା^ ^ ^^^^^. The longest span waspreserved and the others in overlapping span cases removed. Spans with a maximum fractional gamma lower than ^^୪୧୫ଶwere also removed (fractional gamma second limit defined in Table 4). In hypoxic newborns, intrusive delta-dominated oscillations, “delta brushes”, led to premature termination of a span. Therefore, spans less than 15 s apart were concatenated into one. This concatenation procedure was not implemented in other models. Detection of a Single Sub-Span of Reduced Delta within Each Span Given a span from ^^ ௧^ to ^^ଶ, a time series δ భ , δ௧భା^, … , δ௧మ was obtained byconcatenating delta per window within the span. A sub-span was initiated at ^^ଷwhen deltadropped: ^^௧య ^ ^^௧యି^, ^^௧య ^ ^^^^^ (^^^^^ was the delta spectral limit defined in Table 4) andterminated it at ^^ ௧ ^^ସ when delta rose back: ^^ >^^ ^ for all ^^ ൌ ^^ସ ^ 1,  ^^ସ ^ 2, … ^^ଶ.Detection of Bursts within the Sub-Span Atty. Dkt. No.: 093873-1450 Given a sub-span from ^^ଷto ^^ସ, gamma power was concatenated per window withinthe sub-span into a time series ^^௧య ,  ^^௧యା^, … ^^௧ర. The maximum of this time series wasdenoted as ^^^^௫. The burst was recognized if ^^^^௫> gamma spectral limit defined in Table 4. Sub-spans without burst were not included in analysis. Postprocessing In this step, short sub-spans, sub-spans containing burst suppression patterns, or acoustic contaminations were removed (Table 5). Sub-spans with time latencies of < 10 seconds were removed in isoflurane long ramps and hypoglycemic coma EEG traces and that were not containing a cortical switch and burst. Table 5. Postprocessing in automated detection algorithms Burst Suppression Detection Sub-spans of EEGs were extracted and detrended by applying the locdetrend.m function in the Chronux toolbox. Moving windows=[1,0.5] seconds were used. Then for every channel, a series of mean absolute EEG voltage was calculated using nonoverlapping window=0.5 seconds. If, in at least one channel, more than 10% of the values of these series were less than 0.02 mV, this sub-span was calculated as a sub-span containing burst suppression. Acoustic Contamination Atty. Dkt. No.: 093873-1450 In cases where acoustic contamination overlapping with gamma band were observed, the spectrum associated with the burst of the sub-span was dB transformed. It was denoted as^^ ൌ ^^^^,^^ଶ, … ^^^^^^. Then, a robust fit to ^^^^,^^^^, …^^^^^ belonging to the gamma band wasapplied. If the sigma after fit was >5 dB or the maximum residual after fit was >10dB, this sub-span was classified as containing acoustic contaminations. The remaining sub-spans after postprocessing were termed arousal units. Automated Detection of Delta Waves in a Blast Injury Model To determine if a time interval has delta waves, a series of delta power and a series of fractional gamma was calculated over windows (length=5s, step size=1s) within the interval. Intervals with median delta power>1.5 and median fractional gamma<0.4 were classified as delta waves. ADAPTION OF AROUSAL UNIT-DETECTION ALGORITHMS TO HUMAN DATA EEGs measured from optimal electrode pairs were preprocessed and fed them into an adapted automated detection algorithm to extract arousal units from the clinical datasets. Detrending Signals To detrend the signals local linear regression was used by applying the locdetrend.m function in the Chronux toolbox. Moving window=[1,0.5] seconds. Lowpass Filtering Signals The signals were low pass filtered by implementing the mne.filter function in the MNE package. Removing the Power Line Noise A notch filter mne.notch_filter from the MNE package was applied to suppress power line interference at specific frequencies. By default, notch widths=1 Hz. As shown in FIG. 18, the 60 Hz power line noise in subjects 1-2 spanned wide frequency ranges. To remove them while averting severe signal distortions, notch widths=3 Hz was set in subject 1 and 10 Hz in subject 2. Electrode Selection Atty. Dkt. No.: 093873-1450 Bilateral electrodes placed in the areas of interest included C3&C4, F3&F4, FC1&FC2. To determine the optimal pair to detect arousal units, preprocessed EEGs were visually scanned and compared every 30 seconds across electrodes. The electrode pair with fewer electrical interference, fewer movement artifacts, and contrastive EEG changes related to movement was chosen. Table 2 lists the results of electrode selection per individual. Spectral analysis (intra-channel and inter-channel modules) EEGs were preprocessed (assessment epoch≈ 30 minutes; Table 2) and treated using a sliding window with time length=5s and step size=1s. To minimize the effect of artifacts in estimating spectral variables and limits, intra-channel and inter-channel modules were introduced to tackle artifacts that appeared in the channels. Inter-channel artifacts The intra-channel artifacts referred to artifacts occurring in one channel. To recognize them, the normalized and 100 log-spaced spectra for each channel was calculated, denoted as^^ ൌ ^^^^,  ^^ଶ, … ^^^^^^ and ^^ ൌ ^^^^,  ^^ଶ, … ^^^^^^. If the upper-frequency limit was lower than150 Hz, the spectrum was padded with NaN values to match the size of 100. Channel ^^ was removed if it contained excessive power in the delta or gamma band compared to channel ^^ or vice versa. Excessive delta power was detected by measuring paired distance:mean^x1:15 െ ^^^:^ହ^ >0 and mean൫abs^^^^:^ହ െ ^^^:^ହ^൯ >3dB. Here, abs represented takingabsolute value. Similar equations were applied to detect excessive gamma power. For the remaining channels, their spectra were averaged to one. Based on the combined spectrum, spectral variables, including fractional gamma, delta and gamma (Table 4) were derived and spectral limits per epoch (≈ 30 minutes; Table 4) estimated. When calculating spectral limits, windows containing intra-channel artifacts were excluded because they did not measure authentic brain signals. Detecting Spans, Sub-Spans, and Bursts. Detections of spans, sub-spans and bursts were the same as those in the animal models Steps 2-4. The only exception was that when bursts within a sub-span were detected, sliding windows containing intra-channel artifacts were excluded because they did not reflect real cortical gamma activity. Atty. Dkt. No.: 093873-1450 Theta frequency peak calculation. Theta frequency refers to the frequency that occupies power (in watts) in the theta band from a log-spaced spectrum between 2-150 Hz. To calculate theta frequency, uncontaminated from the presence of prominent delta, a right-tailed Z-test was applied todatapoints ^^^ହ,^^^^, … ^^ଷହ representing power in theta band (3.6-8.6 Hz). Datapoints wherechosen that rejected the null hypothesis at significance level=0.1 (outliers), and their frequency indices constituted into a set (red dots). The highest frequency within this set was considered as theta frequency (black line). If no datapoint with P-value<0.1 was found, by default, theta frequency=15 (in 100 log space) was set. To justify significance level=0.1, theta frequency time series and the mean spectrum were calculated from all arousal unit spectra in each period (eight isoflurane long ramps). It was observed that the prominent theta peak in the mean spectrum overlapped with the mean theta time series at 0.1 and was contaminated by the prominent presence of delta power at 0.05 and 0.001, as shown in FIG. 17. Thus, significant level=0.1 was superior in reflecting theta frequency dynamics. Theta frequency was smoothed over time by a moving average filter with span=3 windows to eliminate outliers or noise. Gamma Power Calculation Given a 5s EEG trace, its power at 100 log-spaced frequencies between 2-150Hz wasmeasured and a vector ^^ ൌ ^^^^, ^^ଶ, … ^^^^^^ obtained. ^^ was normalized (Table 3) and dBtransformed and average power of ^^^^, ^^^^, … ^^^^^ calculated. These datapoints correspondedto 39-150 Hz, which stored power in the gamma band. Finally, this mean value was added with gamma offsets (Table 4), which are condition-dependent constants, to allocate gamma changes in the same range. EXAMPLE ALGORITHMS TO OBTAIN AROUSAL INDEX FROM AROUSAL UNITS Here, an automated algorithm that transforms arousal units into an arousal index is described. This index is a quantitative and trackable measure of arousal levels, as shown in FIG. 7. The algorithm consists of learning and implementation phases. In the learning phase, the algorithm takes arousal units detected from 8 isoflurane long ramps as inputs. The algorithm learns a logistic curve that generates arousal indices, and which are continuous Atty. Dkt. No.: 093873-1450 values between 0 and 1 by comparing the gamma-theta distance of arousal units across cortical periods. Zero represents no arousal, and 1 illustrates full wakefulness. In the implementation stage, the pre-fit logistic curve is applied to arousal units detected in a single animal and concatenated the obtained arousal indices over time, thus, quantitatively tracking arousal levels. Learning Phase The first step of the learning phase is detecting arousal units per period, quantifying gamma power and theta frequency, and deriving centroids. To detect arousal units, the automated algorithm shown in FIG. 3 was implemented on 8 isoflurane long ramps and arousal units were extracted. The gamma power and theta frequency were measured for each arousal unit, as shown in FIG. 7E. Centroids were derived by averaging gamma-theta belonging to each cortical period in P2-P5. Cortical periods were time intervals of characterized cortical patterns that calibrate ordinal levels of wakefulness. The second step of the learning phase is to fit a logistic curve by measuring the distance between centroids of a period. Centroids were z-scored-normalized to bring gamma and theta features down to a similar scale. Mean and standard deviation used in the normalization were calculated from all arousal units in P5, which is the maximal arousal level. Then, each z scored pair was transformed into a Z-scored scaler by applying a non- linear activation function ^^^^^^ఊ,^^ఏ^ where if both Z-scored gamma (^^ఊ) and Z-scored theta(^^ఏ) were either positive or negative, they were summed to obtain the Z-scored scaler; otherwise, the Z-scored scaler equaled Z-scored gamma or Z-scored theta, whichever was negative. This quantity was called a Z-scored scaler because it quantifies total number of standard deviations in both gamma and theta to reach / exceed the P5 centroid. Ultimately, a logistic curve was learned to generate arousal index from ^^^by setting distance between thecentroids. The logistic function was written as 1 / ^1 ^ െ ^^^^൯^. Here, ^^ wasthe logistic growth rate and ^^^was the midpoint. The logistic curve was fit such that output of P2 centroid and P5 centroid equaled 0.1 and 0.9, respectively. The third step of the learning phase is to concatenate arousal indices as a function of time. Arousal index were spaced along the time axis. To acquire a trackable measure of arousal levels from occasional arousal indices, zero-padding, interpolation and smoothing Atty. Dkt. No.: 093873-1450 was performed in Matlab. In detail, zeros were added every second from the start of recording to the first arousal unit and every T∕c seconds between adjacent arousal units. Here, T represents the time length (measured in seconds) from the start to the end of the recording. c was a normalization constant and its specification is described later in this paragraph. 1D interpolation was applied to the series using the interp1.m function. The query points were spaced every second from the start of the recording to the last arousal unit. Finally, the curve was smoothed using a moving average filter by implementing smooth.m function. The span of the moving average= T∕c. To find the value of c that optimized the curve fitting, c was iterated from 10 to 100 until R-squared metric (coefficient of determination) was higher than 0.8. The obtained curve showed quantitative and trackable measure of arousal levels. EXAMPLE IMPLEMENTATIONS In example implementations of the present disclosure, depicted in FIG. 19, a system 1910 is provided. The system 1910 includes one or more sensors 1920. The one or more sensors 1920 may include one or more neural sensors (e.g., probes, electrodes, etc.) to be placed on a subject or patient (e.g., on their head). The one or more neural sensors are configured to detect and / or record neural activity (e.g., cortical neural activity, etc.) over a period of time. The one or more neural sensors output the recorded neural activity as neural data. The one or more sensors 1920 may also include one or more visual sensors (e.g., a camera, a video camera, an accelerometer, a pressure sensor, etc.) configured and located to record, for example, motor activity of the subject or patient over a period of time. The one or more visual sensors may provide or output motor data. The system 1910 further includes one or more processors 1930 configured to determine a predictive biomarker (e.g., arousal units, arousal index, etc.) for disorders of consciousness by executing one or more algorithms (e.g., one of the automated detection algorithms for detecting arousal units or arousal indices, described herein). As shown in FIG. 20, the algorithm performed by the one or more processors 1930 of the system 1910 may be an algorithm 2010. Algorithm 2010 may include the step of receiving sensor data 2020. This may be neural data output by one or more neural sensor comprising the sensors 1920 of system 1910. This also may include motion data output by one or more visual sensor comprising the sensors 1920 of system 1910. This also may include Atty. Dkt. No.: 093873-1450 autonomic data output by one or more autonomic data sensors comprising the sensors 1920 of system 1910. When the data received includes neural data, the algorithm 2010 may include the step of analyzing the neural data 2030 to identify the dominant frequency ranges over the period of time the data was collected. These dominant frequencies may relate to the frequencies identified as dominant during the various cortical periods P1 through P5, shown in Table 1. When the data received includes both neural and motor data, the algorithm 2010 may include the step of correlating the motor data 2040. When co-occurring, the motor data may be correlated to the neural data. The motor data may be correlated to the dominant frequency ranges identified in the analyzing neural data step 2030. In some implementations, cortical states (e.g., P1-P5 shown in Table 1) may be identified in the motor data. These identified states may be correlated to the dominant frequency ranges. In some implementations, the identified cortical states may be associated with motor states presented in the motor data. These motor states may include twitches, limb and generalized body movements, weak weight-bearing movements, and organized movements. When the data received includes both neural and autonomic data, the algorithm 2010 may include the step of correlating the autonomic data 2050. When co-occurring, the autonomic data may be correlated to the neural data. The autonomic data may be correlated to the dominant frequency ranges identified in the analyzing neural data step 2030. Finally, algorithm 2010 may include the step of generating a predictive biomarker 2060. The predictive biomarker is configured to indicate the arousal level of the subject or patient. The predictive biomarker may be one or more arousal units or one or more arousal indices, as described herein. The biomarker may be used to characterize fluctuations in the arousal level of the subject or patient. In some implementations of the present disclosure a model may be employed in the step of predicting a biomarker 2060. The model may be trained according to the method 2110 shown in FIG. 21. The method 2110 may include the step of acquiring EEG data 2120 for a cohort of subjects (e.g., mice, humans, etc.). The EEG data may be acquired as the subjects are in varying states of consciousness or arousal (e.g., recovering from a disorder of consciousness, awaking from anesthesia, etc.). Atty. Dkt. No.: 093873-1450 The method 2110 may include the step of detecting arousal units 2130. Arousal units are detected within the EEG data based at least in part on changes in gamma and delta wave activity. For example, an arousal unit may be characterized by a decrease in delta wave activity and a peak or burst in gamma wave activity. These arousal units may be present or occur over a plurality of cortical periods. These cortical periods may correspond to periods of consciousness or wakefulness (e.g., P1-P5 shown in Table 1). The method 2110 may include the step of characterizing arousal units 2140. This step includes calculating a plurality of metrics, for each arousal unit, relating at least in part to the gamma and theta waves present in the arousal units. A plurality of metrics relating to gamma wave and theta wave activity in the arousal units may be used to characterize the arousal units. These metrics may include a calculation of gamma wave power (gamma power) within the arousal units. These metrics may also include the dominant frequency present in the theta wave activity of the arousal units (theta frequency). The method 2110 may include the step of generating centroids 2150. The centroids are generated for each cortical period and are based on the metrics calculated in the step of characterizing the arousal units 2140 for each arousal unit in the corresponding cortical period. For example, the centroids may be generated based on the theta frequency and gamma power of the arousal units in a cortical period. The method 2110 may include the step of training the model 2160. Here, the model is trained using the centroids of the plurality of cortical periods and configured to generate an arousal index. The arousal index may be a value between zero and one, where zero represents complete unconsciousness and one represents total consciousness. In other implementations of the present disclosure, a different range of values or configuration may be used to distinguish between states of arousal. Finally, the method 2110 may include the step of providing the model 2170. The model provided may be employed in an algorithm to produce arousal indices for patients under monitoring. The indices produced may be used in treating or adjusting the treatment of those patients. In example implementations of the present disclosure, and as shown in FIG. 22, a method for characterizing fluctuations in an arousal level of a subject or patient 2210 is Atty. Dkt. No.: 093873-1450 provided. The method 2210 may include the step of acquiring EEG data from a subject using a set of sensors. The sensors may be probes, electrodes, etc. disposed on the subject. The subject may be a patient receiving or in need of treatment for a health condition. The method 2210 may include the step of detecting arousal units 2230. Here, arousal units within the EEG data are detected based at least in part on changes in gamma waves and delta waves in the EEG data. In example implementations, arousal units may be detected in the EEG data by detecting spans of the EEG data with increased gamma activity, identify within those spans of EEG data a sub-span with decreased delta activity, and designating the sub-span an arousal unit if a gamma burst or peak in gamma power is detected within the sub-span. In some implementations, the gamma burst comprises a gamma level exceeding a threshold value. The method 2210 may include the step of generating a characterization 2240. This characterization is a characterization of the fluctuation in the arousal level of the subject. The characterization includes determining an arousal metric (e.g., an arousal index as described herein) for the subject based on the detected arousal units. Finally, the method 2210 may include the step of using the characterization 2250. This characterization is that produced in the step of generating a characterization 2240 and may be employed in treating the subject for a health condition. The health condition may be a disorder of consciousness. The disorder of consciousness may be related to a coma, a coma- like state such as anesthesia, a metabolic disorder, or a brain injury. Various example non-limiting embodiments and aspects of the disclosure include the following: Embodiment A1: A system comprising: at least one neural sensor disposed on a subject, the neural sensor configured to record neural activity of the subject over a period of time and output neural data; at least one visual sensor configured to record motor activity of the subject over the period of time and output motor data; one or more processors configured to determine a predictive biomarker for disorders of consciousness via an algorithm, the algorithm including the steps: receiving the neural data from the at least one neural sensor; receiving the motor data from the at least one visual sensor; analyzing the neural data to identify the dominant frequency ranges over the period of time; correlating the motor data Atty. Dkt. No.: 093873-1450 that co-occurred with the identified dominant frequency ranges; and generating a predictive biomarker, the predictive biomarker configured to indicate a level of arousal of the subject. Embodiment A2: The system of Embodiment A1, further comprising at least one autonomic data sensor configured to record autonomic system activity of the subject over the period time and output autonomic data. Embodiment A3: The system of Embodiment A2, wherein the algorithm further includes the step of: receiving the autonomic data from the at least one autonomic sensor. Embodiment A4: The system of either Embodiment A2 or A3, wherein the algorithm further includes the step of: correlating the autonomic data that co-occurred with the identified dominant frequency. Embodiment A5. The system of any of Embodiments A1-A4, wherein the at least one neural sensor is configured to detect cortical neural activity and output cortical neural data. Embodiment A6: The system of any of Embodiments A1-A5, wherein the at least one visual sensor is a video camera. Embodiment A7: The system of any of Embodiments A1-A6, wherein the algorithm further includes the step of: identifying a plurality of cortical states from correlated motor data that co-occurred with the identified dominant frequency ranges of the cortical neural data. Embodiment A8: The system of Embodiment A7, wherein the algorithm further includes the step of: associating the plurality of cortical states with a plurality of motor states presented in the correlated motor data that co-occurred with the identified dominant frequency ranges of the cortical neural data. Embodiment A9: The system of Embodiment A8, wherein the plurality of motor states includes: twitches, limb and generalized body movements, weak weight-bearing movements, and organized movements. Embodiment B1: A method of generating a model for characterization of a fluctuation in an arousal level of a subject, the method comprising: acquiring EEG data for a cohort of subjects; detecting, based at least in part on changes in gamma waves and delta waves in the EEG data, arousal units for a plurality of cortical periods; determining, for each arousal unit, Atty. Dkt. No.: 093873-1450 a plurality of metrics relating at least in part to gamma and theta waves in the arousal unit; generating a centroid for each cortical period, the centroid based on the plurality of metrics for the arousal units in the corresponding cortical period; training, using the centroids of the plurality of cortical periods, the model to generate arousal index; and providing the model for determination of arousal level in one or more patients. Embodiment B2: The method of Embodiment B1, wherein each cortical period corresponds to a period of consciousness. Embodiment B3: The method of either Embodiment B1 or B2, wherein the plurality of metrics comprises gamma power. Embodiment B4: The method of any of Embodiments B1 – B3, wherein the plurality of metrics comprises theta frequency. Embodiment C1: A method comprising: acquiring EEG data from a subject using a set of sensors; detecting arousal units in the EEG data based at least in part on changes in gamma waves and delta waves in the EEG data; generating a characterization of a fluctuation in an arousal level of the subject, wherein generating the characterization comprises determining an arousal metric for the subject based on the arousal units; and using the characterization in treating the subject for a health condition. Embodiment C2: The method of Embodiment C1, wherein detecting the arousal units comprises: detecting a span of increased gamma activity; identifying, within the span, a sub- span of decreased delta activity; and detecting a gamma burst within the sub-span. Embodiment C3: The method of Embodiment C2, wherein the gamma burst comprises a gamma level exceeding a threshold. Embodiment C4: The method of any of Embodiments C1 – C3, wherein the health condition is a disorder of consciousness. Embodiment C5: The method of Embodiment C4, wherein the disorder of consciousness is related to a coma or a brain injury. Embodiment D1: One or more computing systems and / or one or more computing devices comprising one or more processors to perform any of the methods of Embodiments B1 – C5. Atty. Dkt. No.: 093873-1450 Embodiment E1: One or more non-transitory computer-readable storage media storing instructions executable by one or more processors to perform any of the methods of Embodiments B1 – C5. It should be noted that although method steps may be described in a specific order, it is understood that the order of these steps may differ from what is described. In a non- limiting example, two or more steps may be performed concurrently or with partial concurrence. Also, some method steps that are performed as discrete steps may be combined, steps being performed as a combined step may be separated into discrete steps, the sequence of certain processes may be reversed or otherwise varied, and the nature or number of discrete processes may be altered or varied. The order or sequence of any element or apparatus may be varied or substituted according to alternative embodiments. Accordingly, all such modifications are intended to be included within the scope of the present disclosure as defined in the appended claims. It is understood that all such variations are within the scope of the disclosure. While this specification contains many specific embodiment details, these should not be construed as limitations on the scope of any inventions or of what may be claimed, but rather as descriptions of features specific to particular embodiments of the systems and methods described herein. Certain features that are described in this specification in the context of separate embodiments may also be embodied in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment may also be embodied in multiple embodiments separately or in any suitable subcombination. Moreover, although features may be described above as acting in certain combinations and even initially claimed as such, one or more features from a claimed combination may in some cases be excised from the combination, and the claimed combination may be directed to a subcombination or variation of a subcombination. Having now described some illustrative embodiments and embodiments, it is apparent that the foregoing is illustrative and not limiting, having been presented by way of example. In particular, although many of the examples presented herein involve specific combinations of method acts or system elements, those acts and those elements may be combined in other ways to accomplish the same objectives. Acts, elements, and features discussed only in Atty. Dkt. No.: 093873-1450 connection with one embodiment are not intended to be excluded from a similar role in other embodiments. The phraseology and terminology used herein is for the purpose of description and should not be regarded as limiting. The use of “including,” “comprising,” “having,” “containing,” “involving,” “characterized by,” “characterized in that,” and variations thereof herein, is meant to encompass the items listed thereafter, equivalents thereof, and additional items, as well as alternate embodiments consisting of the items listed thereafter exclusively. In one embodiment, the systems and methods described herein consist of one, each combination of more than one, or all of the described elements, acts, or components. As utilized herein with respect to numerical ranges, the terms “approximately,” “about,” “substantially,” “essentially,” and similar terms generally mean + / - 10% of the disclosed values. When the terms “approximately,” “about,” “substantially,” “essentially,” and similar terms are applied to a structural feature (e.g., to describe its shape, size, orientation, direction, etc.), these terms are meant to cover minor variations in structure that may result from, for example, the manufacturing or assembly process and are intended to have a broad meaning in harmony with the common and accepted usage by those of ordinary skill in the art to which the subject matter of this disclosure pertains. Accordingly, these terms should be interpreted as indicating that insubstantial or inconsequential modifications or alterations of the subject matter described and claimed are considered to be within the scope of the disclosure as recited in the appended claims. Any references to embodiments or elements or acts of the systems and methods herein referred to in the singular may also embrace embodiments including a plurality of these elements, and any references in plural to any embodiment or element or act herein may also embrace embodiments including only a single element. References in the singular or plural form are not intended to limit the presently disclosed systems or methods, their components, acts, or elements to single or plural configurations. References to any act or element being based on any information, act, or element may include embodiments where the act or element is based at least in part on any information, act, or element. Any embodiment disclosed herein may be combined with any other embodiment, and references to “an embodiment,” “some embodiments,” “an alternate embodiment,” “various Atty. Dkt. No.: 093873-1450 embodiments,” “one embodiment,” or the like are not necessarily mutually exclusive and are intended to indicate that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment. Such terms as used herein are not necessarily all referring to the same embodiment. Any embodiment may be combined with any other embodiment, inclusively or exclusively, in any manner consistent with the aspects and embodiment disclosed herein. References to “or” may be construed as inclusive so that any terms described using “or” may indicate any of a single, more than one, and all of the described terms. Where technical features in the drawings, detailed description or any claim are followed by reference signs, the reference signs have been included for the sole purpose of increasing the intelligibility of the drawings, detailed description, and claims. Accordingly, neither the reference signs nor their absence have any limiting effect on the scope of any claim elements. The foregoing description of embodiments has been presented for purposes of illustration and description. It is not intended to be exhaustive or to limit the disclosure to the precise form disclosed, and modifications and variations are possible in light of the above teachings or may be acquired from this disclosure. The embodiments were chosen and described in order to explain the principals of the disclosure and its practical application to enable one skilled in the art to utilize the various embodiments and with various modifications as are suited to the particular use contemplated. Other substitutions, modifications, changes, and omissions may be made in the design, operating conditions and embodiment of the embodiments without departing from the scope of the present disclosure as expressed in the appended claims.

Claims

Atty. Dkt. No.: 093873-1450 WHAT IS CLAIMED IS:

1. A system comprising: at least one neural sensor disposed on a subject, the neural sensor configured to record neural activity of the subject over a period of time and output neural data; at least one visual sensor configured to record motor activity of the subject over the period of time and output motor data; at least one autonomic data sensor configured to record autonomic system activity of the subject over the period of time and output autonomic data; and one or more processors configured to determine a predictive biomarker for disorders of consciousness via an algorithm, the algorithm including the steps: receiving the neural data from the at least one neural sensor; receiving the motor data from the at least one visual sensor; receiving the autonomic data from the at least one autonomic sensor; analyzing the neural data to identify the dominant frequency ranges over the period of time; correlating the motor data that co-occurred with the identified dominant frequency ranges; correlating the autonomic data that co-occurred with the identified dominant frequency ranges; and generating a predictive biomarker, the predictive biomarker configured to indicate a level of arousal of the subject.

2. The system of claim 1, wherein the at least one neural sensor is configured to detect cortical neural activity and output cortical neural data.

3. The system of claim 2, wherein the at least one visual sensor is a video camera.

4. The system of claim 2, wherein the algorithm further includes the step of: identifying a plurality of cortical states from correlated motor data that co-occurred with the identified dominant frequency ranges of the cortical neural data.

5. The system of claim 4, wherein the algorithm further includes the step of:Atty. Dkt. No.: 093873-1450 associating the plurality of cortical states with a plurality of motor states presented in the correlated motor data that co-occurred with the identified dominant frequency ranges of the cortical neural data.

6. The system of claim 5, wherein the plurality of motor states includes: twitches, limb and generalized body movements, weak weight-bearing movements, and organized movements.

7. A method of generating a model for characterization of a fluctuation in an arousal level of a subject, the method comprising: acquiring EEG data for a cohort of subjects; detecting, based at least in part on changes in gamma waves, theta waves, and delta waves in the EEG data, arousal units for a plurality of cortical periods; determining, for each arousal unit, a plurality of metrics relating at least in part to gamma and theta waves in the arousal unit; generating a centroid for each cortical period, the centroid based on the plurality of metrics for the arousal units in the corresponding cortical period; training, using the centroids of the plurality of cortical periods, the model to generate arousal index; and providing the model for determination of arousal level in one or more subjects.

8. The method of claim 7, wherein each cortical period corresponds to a period of consciousness.

9. The method of claim 7, wherein the plurality of metrics comprises gamma power.

10. The method of claim 7, wherein the plurality of metrics comprises theta frequency.

11. A method comprising: acquiring EEG data from a subject using a set of sensors; detecting arousal units in the EEG data based at least in part on changes in gamma waves and delta waves in the EEG data;Atty. Dkt. No.: 093873-1450 generating a characterization of a fluctuation in an arousal level of the subject, wherein generating the characterization comprises determining an arousal metric for the subject based on the arousal units; and using the characterization in treating the subject for a health condition.

12. The method of claim 11, wherein detecting the arousal units comprises: detecting a span of increased gamma activity; identifying, within the span, a sub-span of decreased delta activity; and detecting a gamma burst within the sub-span.

13. The method of claim 12, wherein the gamma burst comprises a gamma level exceeding a threshold.

14. The method of claim 11, wherein the health condition is a disorder of consciousness.

15. The method of claim 14, wherein the disorder of consciousness is related to a coma, a coma-like state, a metabolic disorder, or a brain injury.

Images