RELATED APPLICATIONS

The present application claims priority from Australian provisional patent application No. 2019903911 filed 16 October 2019, the content of which is hereby incorporated in its entirety by reference.

TECHNICAL FIELD

The present invention relates to an automated method for assessment of neuropsychological impairment in subject.

BACKGROUND

Any references to methods, apparatus or documents of the prior art are not to be taken as constituting any evidence or admission that they formed, or form part of the common general knowledge.

Obstructive sleep apnea (OSA) is a serious sleep disorder characterized by breathing interruptions during sleep. Complete cessation of airflow is defined as obstructive apnea and a partial decrease is defined as obstructive hypopnea. The total number of apnea and hypopnea events per hour of sleep is known as the respiratory disturbance index (RDI) [1].

The reference standard for OSA diagnosis is attended Type 1 polysomnography (PSG) [1]. RDI is the major outcome of PSG but it also provides information related to the neurophysiological aspects of sleep such as the arousal index (Arl), sleep latency (SL), and sleep architecture.

Long-term risks of OSA includes diabetes, obesity, and cardiovascular disease [2]. The immediate daytime consequences are neuropsychological impairments [3] such as excessive daytime sleepiness, loss of attention, and impairment of memory and executive functions.

Neuropsychological impairments cause medical, economic and social costs to society. According to the national highway traffic safety administration, USA, drowsy driving causes 100,000 motor vehicle crashes per year resulting in 1,550 deaths, 71,000 injuries and $12.5 billion in damage [4]. OSA patients are 2-9% more likely to have motor vehicle and industrial accidents [4].

In the following disclosure the PVT test is primarily used as a reference for gauging neuropsychological impairment and thus a ‘vigilance’ aspect of the impairment is present. Other aspects of such impairment can relate to impairment of memory, attention and executive function.

The US Center for Disease Control and Prevention discusses cognitive impairment, such as neuropsychological impairment, in general in the following terms:

Cognitive impairment is when a person has trouble remembering, learning new things, concentrating, or making decisions that affect their everyday life. Cognitive impairment ranges from mild to severe. With mild impairment, people may begin to notice changes in cognitive functions, but still be able to do their everyday activities. Severe levels of impairment can lead to losing the ability to understand the meaning or importance of something and the ability to talk or write, resulting in the inability to live independently.

One of the major consequences of (Apnea) OSAS is an impact on neurocognitive functioning. Several studies have shown that OSAS has an adverse effect on inductive and deductive reasoning, attention, vigilance, learning, and memory. Neurocognitive impairment can be measured objectively with tests such as the Wechsler Adult Intelligence Scale-Revised, the Psychomotor Vigilance Task, the Steer Clear Performance Test, and tests of repetitive finger tapping. Neuropsychological manifestations are direct consequences of OSA but there are no efficient tools to measure these. At present, sleep physicians depend on surrogate clinical measures and subjective analysis to evaluate the neuropsychological aspects of OSA. These measures are (i) sleep latency (from a multiple sleep latency test (MSLT) [5]), arousal index, and subjective questionnaires (such as Epworth sleepiness scale (ESS) [6]) and (ii) interview with the patients.

The main outcome of MSLT is sleep latency, defined as the time required for falling asleep. It provides an indirect quantitative measure of “sleepiness”. MSLT is expensive, resource intensive, subjective (in scoring sleep onset), and is not available for routine clinical use on all OSA patients. It requires specialized equipment and several hours of expert care. The arousal index, defined as the number of EEG arousals per hour of sleep, is available from a routine in-facility PSG test as another quantitative measure. However, it is only weakly correlated with diurnal neuropsychological manifestations of OSA limiting its clinical utility.

The ESS is a questionnaire which provides a measure of the general level of daytime sleepiness based on subject self-assessment. However, the self- reporting of sleepiness is not that reliable when subjects have cognitive deficiencies clouding their capacity for judgment. ESS, despite its subjective nature and the narrow focus on sleepiness, is widely used in clinical practice.

Tests such as the psychomotor vigilance task (PVT) can also be used to assess different facets of neuropsychological impairment in OSA. These tests are time consuming, can be expensive, and may not be suitable for the elderly or people with disabilities. Neuropsychological tests, despite potential, are not available as a routine clinical tool in managing OSA.

Neuropsychological manifestations of OSA are the most conspicuous daytime symptoms in OSA [3, 7]; they can have a major impact on patients’ daily activities and occupational safety as well as long term metabolic health. Despite the importance of neuropsychological measures, currently there are no efficient and objective tools to measure them in clinical practice. At present, the decision to begin OSA treatment is not straightforward. The PSG-provided RDI alone is insufficient for the purpose. The physician has to make a judgment call on how severe the effects of OSA are on a given patient.

It would be advantageous if solution were provided which addressed at least one of the above described problems.

SUMMARY OF THE INVENTION

According to a first aspect of the present invention there is provided a method of assessing neuropsychological impairment of a subject, comprising: processing first and second EEG signals to determine a measure of asynchrony therebetween, the first and second EEG signals having been recorded from electrodes placed to monitor brain activity from different regions of the subject’s brain; forming a test vector based on the measure of asynchrony and/or one or more values for features computed therefrom; applying the test vector to a pre-trained pattern classifier; classifying the subject as being neuropsychologically impaired based on an output from the classifier; and generating a display corresponding to the output from the pre-trained pattern classifier to indicate whether or not the subject suffers from a neuropsychological impairment.

In an embodiment the first and second EEG signals are recorded from electrodes placed on the left and right hemispheres of the brain of the subject.

In an embodiment the first and second EEG signals are recorded from electrodes placed on the same hemisphere of the brain of the subject.

In an embodiment the first and second EEG signals are acquired as part of a polysomnography (PSG) test. In an embodiment the method includes segmenting each of the first and second EEG signals into respective epochs.

In an embodiment determining the measure of asynchrony between the first and second EEG signals comprises processing the first and second EEG signals to calculate a spectral correlation between corresponding epochs of the respective epochs of the first and second EEG signals.

In an embodiment the method includes calculating the spectral correlation between the corresponding epochs in each of a number of frequency bands.

In an embodiment the frequency bands include one or more of the delta, theta, alpha, beta, high beta and gamma, frequency bands.

In an embodiment the spectral correlation between the corresponding epochs for each of said frequency bands comprises an asynchrony time series for each frequency band.

In an embodiment the test vector includes values of features of the asynchrony time series for each frequency band.

In an embodiment the features of the asynchrony time series comprise one or more of: mean of the asynchrony time series for one or more of the respective frequency bands; standard deviation of the asynchrony time series for one or more of the respective frequency bands; variance of the asynchrony time series for one or more of the respective frequency bands; skewness of the asynchrony time series for one or more of the respective frequency bands; and kurtosis of the asynchrony time series for one or more of the respective frequency bands. In an embodiment the method includes incorporating features in respect of the subject into the test vector, comprising one or more of: an indication of heart rate variability; an indication of deviation from Gaussian distribution of the measure of asynchrony and/or one or more quantities computed therefrom;

Chi-Squared and/or Lilliefors test statistic and/or p-values of such tests in respect of the measure of asynchrony and/or one or more quantities computed therefrom; an indication of oxygen saturation such as peripheral oxygen saturation (SpO2); age; gender;

Epworth Sleepiness Score (ESS);

Body Mass Index (BMI);

Depression anxiety / stress scale result;

Respiratory Disturbance Index (RDI); and Arousal Index (Arl).

In an embodiment the method includes applying Principal Component Analysis (PCA) to transform a dimensional input space into an orthogonal feature space of lower dimension than the input space when forming the test vector.

In an embodiment the test vector includes features from one or more of the feature classes set out in Table 22 herein.

In an embodiment the method includes training a classifier to produce the pretrained pattern classifier, wherein the classifier is trained with reference to a population of subjects including a first group of subjects suffering from neuropsychological impairment and a second group of subjects suffering from less said impairment than the first group.

In an embodiment the method includes testing each of the subjects of the population for neuropsychological impairment by applying a Psychomotor Vigilance Test (PVT) to the subject. In an embodiment the subjects belonging to the first group and subjects belonging to the second group are determined by clustering results of the PVT test.

In an embodiment the PVT is applied to each subject for determining a training vector associated with the subject.

In an embodiment the PVT is applied in the evening prior to the subject taking a PSG test during which the first and second EEG signals are acquired.

In an embodiment the method is implemented by one or more electronic processors configured with a software product comprised of machine-readable instructions for the one or more electronic processors to operate according to any one of the preceding claims.

According to a further aspect of the present invention there is provided an apparatus for neuropsychological impairment assessment comprising: at least one electronic processor; a data communication port arranged to interface with an output side of an analog-to-digital converter of an EEG acquisition device for receiving first and second EEG signals from different regions of a brain of a subject; an electronic memory in communication with the at least one processor, the electronic memory storing executable instructions configuring the at least one processor to: acquire the first and second EEG signals; determine a measure of asynchrony of first and second EEG signals relative to each other; form a test vector based on the measure of asynchrony and/or one or more values for features computed therefrom apply the test vector to a pre-trained pattern classifier; and present an assessment of the neuropsychological impairment based on an output signal from the pre-trained pattern classifier in response to the test vector. In an embodiment the electronic memory further stores executable instructions configuring the at least one processor to form the test vector with values for features of an asynchrony time series, said features comprising one or more of: mean of the asynchrony time series for one or more of the respective frequency bands; standard deviation of the asynchrony time series for one or more of the respective frequency bands; variance of the asynchrony time series for one or more of the respective frequency bands; skewness of the asynchrony time series for one or more of the respective frequency bands; and kurtosis of the asynchrony time series for one or more of the respective frequency bands.

In an embodiment the electronic memory further stores executable instructions configuring the at least one processor to form the test vector with values for features in respect of the subject comprising one or more of: an indication of heart rate variability; an indication of deviation from Gaussian distribution of the measure of asynchrony and/or one or more quantities computed therefrom;

Chi-Squared and/or Lilliefors test statistic and/or p-values of such tests in respect of the measure of asynchrony and/or one or more quantities computed therefrom; an indication of oxygen saturation such as peripheral oxygen saturation (SpO2); age; gender;

Epworth Sleepiness Score (ESS);

Body Mass Index (BMI);

Depression anxiety / stress scale result;

Respiratory Disturbance Index (RDI); and Arousal Index (Arl). In an embodiment the electronic memory further stores executable instructions configuring the at least one processor to apply Principal Component Analysis (PCA) to transform a dimensional input space into an orthogonal feature space of lower dimension than the input space when forming the test vector.

In an embodiment the electronic memory further stores executable instructions configuring the at least one processor to form the test vector including values for features from one or more of the feature classes set out in Table 22 herein.

According to a further aspect of the present invention there is provided a method for assessing neuropsychological impairment of a subject based on an indication of asynchrony between first and second EEG signals from electrodes placed to monitor brain activity from different regions of the subject’s brain.

The first and second EEG signals may be from electrodes placed on the left and right hemispheres of the brain of the subject.

Alternatively, the first and second EEG signals may be from electrodes placed on the same hemisphere of the brain of the subject.

According to another aspect of the present invention there is provided an apparatus for assessing neuropsychological impairment of a subject, the apparatus being arranged to determine an indication of asynchrony between first and second EEG signals from electrodes placed to monitor brain activity from different regions of the subject’s brain.

The apparatus may receive the first and second EEG signals from electrodes placed on the left and right hemispheres of the brain of the subject.

Alternatively, the apparatus may receive the first and second EEG signals from electrodes placed on the same hemisphere of the brain of the subject. BRIEF DESCRIPTION OF THE DRAWINGS

Preferred features, embodiments and variations of the invention may be discerned from the following Detailed Description which provides sufficient information for those skilled in the art to perform the invention. The Detailed Description is not to be regarded as limiting the scope of the preceding Summary of the Invention in any way. The Detailed Description will make reference to a number of drawings as follows: Figure 1 depicts an apparatus according to an embodiment of the present invention for assessing neuropsychological impairment of a subject, shown connected to an EEG acquisition assembly which has electrodes that are coupled to a head of the subject. Figure 2 is a block diagram of the apparatus.

Figure 3 is a flowchart of a method implemented by the apparatus, the method being coded as instructions in a software product stored in a memory of the apparatus. Figure 4 depicts a typical deep sleep EEG signal from a hemisphere of the subject. Figure 5 depicts a typical NREM sleep EEG signal from a hemisphere of the subject.

Figure 6 graphically represents exemplary epochs of EEG data processed by the apparatus during performance of the method,

Figure 7 depicts typical data epochs suffering from artefacts and to be discarded. Figure 8 depicts typical waveforms of EEG signal for each of five exemplary EEG frequency bands. Figure 9 depicts ROC curves showing the different performances of each PVT measurement in separating PVT results for a number of subjects into into groups G1 (suffering from neuropsychological impairment) and G2 (suffering less neuropsychological impairment than G1, during training of a classifier. Figure 10 is an ROC curve for a feature selection combination described in Table 8. Figure 11 is an ROC curve for a further feature selection combination described in Table 8. Figure 12 is an ROC curve for a feature selection combination set out in Table 16. Figure 13 is an ROC curve for a feature selection combination set out in Table 18. Figure 14 is a flowchart of a method training a classifier and for testing the sensitivity, specificity and accuracy of the classifier once trained.

DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS

Figure 1 depicts a preferred hardware configuration for practicing a method according to a first embodiment of the present invention. The hardware configuration in this first embodiment comprises an EEG acquisition assembly 1 which is interfaced to an apparatus being an Impairment Assessment Machine 33 in the form of a specially programmed computer. Acquisition assembly 1 is a conventional arrangement for converting electrical signals between pairs of electrodes into digital signals that can be transmitted to the Impairment Assessment Machine 33. In the presently described first embodiment two pairs of electrodes are fitted to patient 3, in subsequent embodiments more than two pairs may be used and supplementary physiological signals may also be acquired in addition to EEG signals. For example, as shown in Figure 1, machine 33 may receive electrocardiograph (ECG) signals and peripheral oxygen saturation SpO2 concentration signals from respective sources 8 and 4 and supplementary data, for example demographic information about the subject from a remote source 10 via data bus 6. Commonly the EEG signals and the SpO2 signal will be received from apparatus used to perform a PSG test on the subject. It will be realized that not all of the signals that are acquired need to be used for all embodiments of the present invention as will become clear in the following discussion. The two pairs of electrodes depicted in Figure 1 include a first pair comprising electrodes C4 and A2 for monitoring EEG signals from the right hemisphere of the brain of patient 3. The second pair of electrodes comprises electrodes C3 and A1 for monitoring the left hand hemisphere of the brain of patient 3.

The method can only be carried out using a specialized EEG processing impairment assessment apparatus or “Impairment Assessment Machine”. The Impairment Assessment Machine 33 may be a dedicated assembly that includes specific circuitry to carry out each of the steps that will be discussed. Alternatively, the Impairment Assessment Machine 33 may be a specially configured desktop computer, as illustrated in Figure 1, or a portable computational device such as a smartphone that contains at least one processor in communication with an electronic memory that stores instructions that specifically configure the processor in operation to carry out the steps of the method as will be described. It will be appreciated that it is impossible to carry out the method without the specialized hardware, i.e. either a dedicated machine or a machine that is comprised of specially configured one or more processors.

Figure 2 is a block diagram of the EEG processing impairment assessment machine 33 implemented using the one or more processors and memory of a desktop computer configured according to Impairment Assessment Software 40. Impairment Assessment Machine 33 includes a main board 34 which includes circuitry for powering and interfacing to one or more onboard microprocessors 35.

The main board 34 acts as an interface between microprocessors 35 and secondary memory 47. The secondary memory 47 may comprise one or more optical or magnetic, or solid state, drives. The secondary memory 47 stores instructions for an operating system 39. The main board 3 also communicates with random access memory (RAM) 50 and read only memory (ROM) 43. The ROM 43 typically stores instructions for a startup routine, such as a Basic Input Output System (BIOS) or Unified Extensible Firmware Interface (UEFI) which the microprocessor 35 accesses upon start up and which preps the microprocessor 5 for loading of the operating system 39. The main board 34 also includes an integrated graphics adapter for driving display 47. The main board 33 will typically include a communications adapter 53, for example a LAN adaptor or a modem or a serial or parallel port, that places the server 33 in data communication with a data network or another device such as EEG acquisition device 1.

An operator 67 of Impairment Assessment Machine 33 interfaces with it by means of keyboard 49, mouse 21 and display 47. As will be explained later in the present specification, in some embodiments information about a patient, such as demographic information or scores concerning the a subject's condition, for example Arl, RDI and similar data, may need to be submitted to the machine 33 and that can be done by use of the keyboard 49, mouse 21 and display 47.

The operator 67 may operate the operating system 39 to load software product 40. The software product 40 may be provided as tangible, non-transitory, machine readable instructions 59 borne upon a computer readable media such as optical disk 57. Alternatively it might also be downloaded via port 53.

The secondary storage 47, is typically implemented by a magnetic or solid state data drive and stores the operating system, for example Microsoft Windows, and Ubuntu Linux Desktop are two examples of such an operating system.

The secondary storage 47 also includes software product 40, being an impairment assessment software product 40 according to an embodiment of the present invention.

Although in the presently described embodiment the Machine 33 is built using an architecture of a desktop computer, it might equivalently make use of a tablet laptop or smartphone as its hardware platform. For example EEG signal acquisition and processing using a smartphone running Android operating system is discussed at length in BioMed Research International Volume 2017, Article ID 3072870, EEG Recording and Online Signal Processing on Android: A Multiapp Framework for Brain-Computer Interfaces on Smartphone by Sarah Blum, Stefan Debener, Reiner Emkes, Nils Volkening, Sebastian Fudickar, and Martin G. Bleichner; the disclosure of which is hereby incorporated in its entirety by cross-reference. During operation of the Machine 33 the one or more CPUs 35 load the operating system 39 and then load the software 40. Software 40 includes a pre-trained classifier that is trained to assess, i.e. to classify, severity of a subject's neuropsychological impairment based upon a test vector associated with the subject that contains information including features of an indication of asynchrony between EEG signals from left and right hemispheres of the patient's brain.

A first example of a method according to an embodiment of the invention will now be described with reference to Figures 1 and 2 and also with reference to Figure 3 which is a flowchart of the method that is coded as machine readable instructions into Impairment Assessment Software 40.

Initially, first and second electrodes, which in the present case are the electrodes C3, A1 for the left hemisphere EEG signal and C4, A2 for the right hemisphere EEG signal are attached to the head of patient 3 as illustrated in Figure 1. In other embodiments the various electrodes may all be located on one or other of the two hemispheres. Preferably the patient is sleeping and attended by a sleep technician who is carrying out a polysomnography (PSG) test. Each of the electrode pairs C3, A1 and C4, A2 pick up EEG signals such as Deep Sleep and NREM sleep EEG signals as illustrated in Figures 4 and 5.

The EEG interface 1 is activate and converts the left and right hemisphere EEG signals into equivalent digital signals which are conveyed to Impairment Assessment Machine 33 via a data connection such as a data bus 6 and port 53.

With reference to Figure 3, at box 301, the microprocessor 35 records the left and right hemisphere EEG signals as files in secondary storage 47. At box 303 the microprocessor 35 reads the left and right hemisphere EEG signal files from the secondary storage 47 and segments each into a series of left and right signal segments as illustrated in Figure 6.

The digitized EEG data recorded from hemisphere "i" of the brain during PSG test is referred to as x _i (t), where i = Rh and i = Lh symbolize the right and left hemispheres of the brain, respectively.

Microprocessor 35 segments x _i (t) into blocks of size T = 30 seconds. Note in clinical scoring, these segments are called “epochs.” From here onward, the same notation is followed. Where there are a total of S epochs the symbol x ^s _i (t) represents the sth epoch, where s = 1, 2,..., S.

Artefact elimination is also preferably performed. Although the EEG signals will normally be a recording of the brains electrical activity, they may contain electrical activities arising from sites other than the brain. The types of artefacts that may occur are: a) movement artefacts — these are contributed by, head movement of the patient, changing sleep position, etc; b) Electro-oculogram artefacts — these are of two kinds: eye blink and eye ball movement artefacts; and c) electrode displacement artefacts — these artefacts are induced into the EEG when a sensor comes off.

Epochs that were contaminated with these kinds of artefacts are manually identified by looking at multiple physiological signals and marked as artefact epoch (AE). Microprocessor 35 excludes all AE-marked epochs from subsequent processing. Fig. 6 and 7 shows the example of the typical EEG data from a 30-s epoch included for further processing and rejected due to artefacts, respectively. Q is the total number of artefact-free epochs and is the qth artefact-free epoch at the end of Step 303, where q = 1, 2, 3,...,Q. At box 305, the microprocessor 35 uses the epochs to calculate spectral correlation coefficient (r), as described in Steps B1 and B2 of V. Swarnkar, U. R. Abeyratne, and C. Hukins, “Inter-hemispheric asynchrony of the brain during events of apnoea and EEG arousals,” Physiol. Meas., vol. 28, pp. 869-880, 2007 (the disclosure of which is hereby incorporated in its entirety by reference). r ^j (q) represent the coherence coefficient computed between C4-A2 and C3-A1 EEG data, where j∈ {δ, θ, α, β}. The IHA time series is defined, using (1) as

In the presently described embodiment, the inter-hemispheric asynchrony (I HA) time-series, which is a measure of asynchrony, is formed for the following six frequency bands: · delta (0.1-4 Hz),

• theta (4.1-8 Hz),

• alpha (8.1-12 Hz),

• beta (12.1-16 Hz),

• high beta (16.1-25 Hz), and · gamma (25.1-35 Hz).

The term IHA can also refer to “intra-hemispheric asynchrony” where the electrodes are located at different places on the same hemisphere. Although in the present example the IHA time series is formed for all of the frequency bands in other embodiments it may be formed over a lesser number of bands or over different bands or over a single range of frequencies.

The microprocessor 35 is programmed by software 40 to form the I HA time series by computing the spectral correlation between the first and second EEG signals, e.g. two EEG channels C4-A2 (right hemisphere) and C3-A1 (left hemisphere), or other EEG electrodes when needed. As an alternative to using electrodes placed on left and right hemispheres of the subjects brain first and second EEG signals may be processed for asynchrony (using the same IHATS process that has been described) from electrodes placed to monitor brain activity from different regions of the same hemisphere of a subject's brain, for example the first electrode may monitor the frontal lobe of the hemisphere whereas the second electrode monitors the occipital lobe of the same hemisphere.

At box 307, the microprocessor determines, for each IHA of the six IHA time series, values for the following five statistical features which are comuted from the measure of asynchrony in the form of the IHA, features in the present example are: mean, variance, standard deviation, skewness and kurtosis as follows:

B1) Compute the mean of the ψ ^j time series for all frequency bands and for NREM and REM sleep states as μ ^j _NREM and μ ^j _REM , respectively, using (2)

B2) Compute the standard deviation of the ψ ^j time series for all frequency bands and for NREM and REM sleep states as σ _NREM ^j and σ _REM ^j , respectively, using

(3)

B3) Compute the variance of the ψ ^j time series for all frequency bands and for NREM and REM sleep states as Var ^j _NREM and Var ^j _REM , respectively, using (4)

B4) Skewness (γ) is the measure of asymmetry associated with probability distribution of any random variable. Compute the skewness associated with the probability distribution of the ψ ^j (q) time series for all frequency bands and for NREM and REM sleep states as γ _NREM ^j and γ _REM ^j , respectively, using (5) B5) Kurtosis ( k ) is a measure of the peaked ness associated with a probability distribution of any random variable. Compute the kurtosis associated with the probability distribution of the ψ ^j (q) time series for all frequency bands and for NREM and REM sleep states as κ ^j _NREM and κ ^j _REM , respectively, using (6)

The five values calculated according to equations (2)-(6) for each of the six frequency bands results in 30 values. At box 309 microprocessor 35 forms the 30 values into a test feature vector based on the measure of asynchrony and/or one or more of the values for the features (1)-(6) that have been computed from the measure of asynchrony. The test vector in the present is of dimension 30, for example the test vector may comprise a data structure such as an array of dimension 30 which contains the values that have been calculated at box 307.

At box 311 the microprocessor 35 applies the test vector to a pre-trained classifier 42 which is stored in secondary memory 47 as part of the Impairment Assessment Software 40. The classifier produces an output signal that represents the likelihood of the subject falling into a neuropsychologically impaired or non (or lesser) neuropsychologically impaired group depending on how the classifier was previously trained.

At box 313 the microprocessor operates display 47 to produce a display corresponding to the output from the pre-trained classifier 42 which indicates whether or not the classifier deems the test vector to be associated with a subject that suffers from a neuropsychological impairment or not. A clinician observing the display can then take steps to assist the subject if the display indicates that the subject is neuropsychologically impaired for example by applying a suitable therapy. For example the clinician may apply therapy to alleviate sleep apoena.

The classifier, which has been previously trained, may comprise a pre-trained logistic regression model (LRM) that has been trained using training vectors derived according to each of boxes 301 to 307 from a population of subjects that fall into either of two groups being a first group G1 of people that suffer from a neuropsychological impairment and a second group G2 of people that do not suffer, or suffer less from a neuropsychological impairment than the first group. As will be discussed, whether or not the subject suffers from the neuropsychological impairment is determined by using a manual test for such an impairment, for example a psychomotor vigilance task (PVT) test.

A PVT test was performed on each subject on the evening of the subject's scheduled overnight PSG study. A portable tablet computer loaded with an open-source software system (PEBL) [16] was used to conduct the test. The PEBL allows design and run simple-response- time (SRT) task similar to the one described in the original PVT test [17]. Test was conducted in a quiet examination room. Subjects were instructed to respond to the appearance of a visual stimulus (an ‘X’ in the middle of the screen) by pushing a response button as quickly as possible. The test ran for 10-15 minutes with visual stimuli appearing at random time intervals spanning from 250-2500ms. Subject reaction times (RTs) were recorded from each PEBL-SRT trial.

According to existing literature, sleepiness and sleep loss are associated with increased variability in the reaction time [17, 18] which could be evaluated using the following PVT measures: (i) mean 1/RT measuring response speed, (ii) median RT, (iii) fastest 10% of RT, (iv) the slowest 10% of RT, and (v) number of lapses defined as number of RT≥500ms.

Even though SRT-PVT test has proven capability to capture aspects of neuropsychological impairment in OSA, it suffers from the non-availability of published normative data. Considering that the use of SRT-PVT data is limited to answering the question that whether EEG asynchrony carries information on neuropsychological impairment, the Inventors followed a process of PVT clustering to overcome the absence of normative data.

The Inventors clustered subjects into two groups, G1 and G2, based on five PVT parameters:

(i) mean of 1/RT, (ii) median of RT,

(iii) mean of fastest 10% of RT,

(iv) mean of slowest 10% of RT, and

(v) lapse count.

The “good” PVT group (G2), has lower RT values and a lower lapse count and the inferior PVT group (G1) has them relatively higher.

Standard k-means algorithm [19] was used on these parameters to group the subjects into two groups. It should be noted that this clustering algorithm is an unsupervised technique and hence ‘discovers’ the two clusters based on the parameters themselves. Once the two groups G1 and G2 are formed, the next task was to find out if EEG asynchrony-based features could be used to build classification models that can map subjects to the associated group (G1 or G2) with reasonable accuracy.

A classifier in the form of a logistic regression model (LRM) is preferably used to map each patient to one of the groups, G1 or G2. It will be realised that in other embodiments of the invention other suitable classifiers may be used, such as an artificial neural network or a Bayesian decision machine and thus the invention is not limited to the use of an LRM only.

LRM is a generalized linear model which uses several independent features to estimate the probability of a categorical event (dependent variable). In this work, the dependent variable Y is assumed to be equal to “one” (Y=1) for ‘poor PVT performance subjects in GY and “zero” (Y=0) for ‘good PVT performance subjects in G2’. A model is derived using a regression function to estimate the probability Y=1, (that is, subject belongs to group G1) given the independent variable (IHATS statistical features) as follows: where f ₁ , f ₂ , ... , f _F are independent variables (IHATS features as calculated at box 307), β ₀ is the intercept, and β ₁ , β ₂ , ... , β _F are the regression coefficients of independent variables.

Leave-one-out cross-validation (LOOCV) technique was used for LRM design. That is, one subject was left out for testing and the remaining subjects used for training the model, one at a time. The performance of the model is evaluated using sensitivity and specificity, assuming the PVT based /(-means grouping of G1/G2 as the reference technique. Confidence intervals of the estimates are provided. When clustering PVT data to two groups (‘good’: G1 and ‘bad’: G2), each PVT measurement m is normalized to have a unity variance across the training set, for instance, m _n = (m)/σ _m , where μ _m and σ _m are mean and standard deviation of a PVT measurement m. Subjects from the training and validation dataset (n = 250) were subjected to k- means clustering algorithm to form G1 and G2.

PVT parameters used for group were : number of lapses, slowest 10% of RT, fastest 10% of RT, mean of 1/(slowest 10% RT), median RT

Number of subjects in the cluster G1 = 204 Number of subjects in the cluster G2 = 46

The ROC curves in Figure 9 indicate the success of this clustering approach. Figure 9 shows the performance of each PVT measurement in separating the data into G1 and G2 (during learning). The grouping G1-G2 depends on all available PVT measurements, but the process is dominated by several measurements such as lapses and mean 1/(RT). The target is to pick G1 from a mix of G1 and G2.

For the embodiment that has previously been discussed, i.e. where 30 EEG Features were generated (five from each of six frequency bands) from sleep epochs without discriminating REM and NREM the results were as set out in Table 1 below.

Table 1

Further embodiments of the invention will now be discussed in relation to the following Tables 2 to 24 which document the results of using additional features to derive training and test vectors for training the LRM and for applying to the LRM in respect of an unclassified subject who may or may not be suffering from a neuropsychological impairment.

In another embodiment microprocessor 35 was programmed to process 60 EEG Features (30 features from NREM sleep & 30 features from REM sleep). The results were as set out in table 2 below.

Table 2

60 EEG Features + Age, BMI, Gender, ESS [R1] ESS: Epworth Sleepiness Score (a questionnaire-based method to assess sleepiness. This is the method commonly used in clinical practice). BMI: Body Mass Index = weight in kg/(height in meters) ²

Table 3

60 EEG Features + RDI, Arl [R2]. Note: Arl = Arousal Index, which is available after a routine PSG study. This is a valid measurements to use, because a target of an embodiment is to use night-time PSG measurements and predict daytime neurocognitive impairments as seen in PVT studies.

Table 4 Table 5 indicates results obtained with a range of individual measurements available after routine PSG and/or demographics data. It is clear such parameters do not carry much capacity to separate the group G1 from G2. This is the problem sleep clinicians face in their daily practice — not having a reliable measure to assess neurocog nitive deficiencies in apnea patients, even though such deficiencies are a driving force behind prescribing apnea treatment devices (e.g. CPAP, MAS).

Table 5

Note: RDI - Respiratory Disturbance Index, WSFI - weighted sleep fragmentation index.

In Table 6, the performance that is possible by augmenting EEG based features with SpO2 (blood oxygen saturation) channel, which is measurement available with PSG systems is set out. The SpO2 channel plays a vital role in defining hypopnea events, which is usually the main component in the RDI. SpO2 Features (24 Features) are as follows: SpO2 Event Types - Mark the SpO2 events satisfying following conditions:

• MinPH - Minimum Drop in the SpO2 value from 100% reference level is d1%

• MinPP - Minimum Drop in the SpO2 value from the nearby background reference level is d2%

• MaxPH - Maximum Drop in the SpO2 value from 100% reference level is d3%. This is to avoid unreliable values of SpO2 due to sensor coming off.

• MinPD - Minimum distance from the previous peak is t seconds. From each event compute following 3 parameters:

• Drop in SpO2 value from 100% reference level during the events

• Half width of the event

• Drop in SpO2 value from the nearby background. Compute the following statistical features from each of the parameter array. Mean, Variance, Skewness, Kurtosis, NGS & 25 ^th , 50 ^th & 75 ^h percentile.

EEG + SpO2 Features:

Table 6 EEG + SpO2 + Demographic Features. Demographic features include for example the age and gender of the subject his/her BMI and ESS.

Table 7 Best outcomes are underlined in Table 8:

Table 8 3. Results obtained with all EEG channels C3, C4, F3, F4, A1, A2, O1 and O2 of the International 10/20 system of electrodes.

Table 9 - Channel C3A1 ; C4A2. (30 features from NREM sleep & 30 from REM sleep) [ResT1]. Note C3A1 means the EEG measured between the channels C3 and A1 of the International 10/20 standard and so on.

Table 9 Table 10 - Channel F3A1; F4A2. (30 features from NREM sleep & 30 from REM sleep) [ResT2]

Table 10 Table 11 - Channel O1A1; O2A2. (30 features from NREM sleep & 30 from REM sleep) [ResT3]

Table 11 Table 12 - All Channels C3C4, F3F4, O102. (Total 180 features from NREM and REM sleep) [ResT4]

Table 12

Table 13 - All Channels C3C4, F3F4, O102 + Age, BMI, Gender, ESS, RDI,

Arl (Total 186 features )[ResT7]

Table 13 Use of Heart Rate Variability (HRV) in predicting PVT outcomes in some embodiments:

The Inventors have conceived that the effects of apnea may be described along three different axes: the respiratory axis, neurophysiological axis, and the cardiovascular-metabolic axis. In this work, the respiratory axis is largely represented by the Respiratory Disturbance Index (RDI); the neurophysiological axis is represented by the EEG, SpO2 and HRV are added to represent the cardiovascular-metabolic axis. The Inventors expect that cognitive deficiency as revealed by the PVT tests, to have the highest correlation with EEG which measure brain function. However, the Inventors also believe SpO2 and HRV parameters to carry non-random information supplementing EEG.

Heart Rate Variability

1. Identify the QRS waveform in the ECG signal by implementing QRS detection algorithm, eq. 1, by Baida et al. [Chapter 4, Biomedical Signal Analysis by Rangaraj M Rangayyan, the disclosure of which is incorporated herein in its entirety by reference].

In (3), x(n) is ECG signal and N is the moving average filter length.

2. Using a peak detection algorithm identify the locations of the QRS peak. Parameters controlled in this step are, Minimum peak height, Minimum peak distance from the previous peak.

3. Once the QRS locations is known it is then possible to compute the Heart Rate time series (HRTS) by computing the R-R interval between QRS peaks.

4. Resulting HRTS signal would be a non-uniformly sampled signal. Resample this signal to convert it into a uniformly sampled signal. 5. Compute 31 mel-frequency components from the HRTS

A better capture of the probability density function. NGS (.) like operators in some embodiments: The Inventors recognise that the inter-hemispheric asynchrony time series (IHTS) which is equivalently an “intra-hemispheric asynchrony time series” depending on the electrodes used to source the EEG signals, is a stochastic signal, and the probability density function carries the information in the signal. The Inventors believe that features such as the NGS of IHATS, and Chi- Squared of IHATS, Lilliefors test statistic and p-values of such tests, to gain information from the IHATS distribution. In the following one such result is illustrated in tables 19 and 20.

NGS of IHATS

1. EEG data (C3 - A1 ; C4 - A2; F3 - A1 ; F4-A2) was filtered using band pass filter with lower cut-off at 0.1 Hz and high cut-off at 35 hz.

2. Segment the EEG data into blocks of size t = 2s.

3. Compute the FFT of EEG data in each blocks. Let the resulting spectral magnitude be denoted by Y _i ^q (ω), where q represent the spectral data from the q ^th block and i represents the EEG channel (C3A1 , C4A2, F3A1 , F4A2).

4. Separate Y _i ^q (ω) (spectral EEG data, from step 3) into 6 spectral frequencies. Let the resulting spectral magnitude be represented by Y _ij ^q (ω). The subscript j represents EEG frequency bands, i.e. , j ∈ {δ, θ, α, β ₁ , β ₂ , γ} delta (δ, 0.1-4 Hz), theta (θ, 4.1-8 Hz), alpha (α, 8.1-12 Hz), beta (β ₁ , 12.1-16 Hz), high beta (β ₂ , 16.1-25 Hz), and gamma (γ, 25.1-35 Hz). 5. Compute the spectral correlation coefficient between two blocks of spectral EEG data, for all the blocks to form the Inter-hemispheric Asynchrony Time Series (IHATS). Twelve IHATS were formed, one each for 6 EEG frequency bands using 2 pair of EEG channels (pair 1 C3A1 & C4A2; pair 2 F3A1 & F4A2). Let ψ _j ^p [n] denotes the IHATS data, where j represents the EEG bands and p indicates the channel pair p ∈ { pair 1 C3A1 & C4A2; pair 2 F3A1 & F4A2}

6. Taking M blocks of IHATS data at a time compute the Non-Gaussianity Score. Let ξ _j ^p [m] represents the NGS time series for j EEG frequency bands and p pair of EEG channels.

7. Next NGS value in ξ _j ^p [m] were compared against a threshold η to define L _j ^p [m] using (1).

8. Compute IHATS-NGS Index, for all EEG frequency bands and both the EEG channel pairs using (2).

9. F _j ^p in (2) represents a feature vector.

Table 14 - NGS of IHATS features (Total 18 features) [ResT12]

Table 14

Table 15 - NGS of IHATS features + Age, BMI, Gender, ESS, RDI, Arl (Total 24 features) [ResT13]

Table 15

Combining all available features - the best outcome so far Table 16 - Combined optimal IHATS, NGS of IHATS, SpO2, HRV features + Age, BMI, Gender, ESS, RDI, Arl (Total 37 features) [ResT15]

Table 16

Table 17

Another feature set obtained by choosing a different feature optimization pathway:

Table 18 - [ResT5],

Table 18 Table 19

The Inventors have also devised, according to some embodiments, a method to establish a relationship between neural behaviour of brain and the performance on vigilance tasks of OSA patients by using features derived from EEG data and PVT test results but also taking into account the impact of other morbidities such as anxiety and morbid obesity, which may be prevalent in an OSA population, on the performance of vigilance tasks. An overview of an method for investigating the factors that impact on performance of cognitive tasks is provided in the flowchart of Figure 14 which illustrates the following stages:

(1) data acquisitions,

(2) feature computation,

(3) feature combination-based feature classes, PVT feature clustering, (4) PCA-based feature selection (optimisation) and forming PVT outcome-based classes,

(5) Training / Validating Logistic Regression models, and

(6) Testing for sensitivity, specificity and accuracy

In a retrospective study, the Inventors used patient data gathered Type 1 PSG (Grael Compumedics®, Sydney, Australia) for patients with suspected OSA. Informed consent was obtained from all participants and ethics approval was obtained from the relevant authorities. Scoring was according to the Subjects with complete data were included in the analysis.

Subjects were grouped into two groups based on following five Response Time (RT) based PVT outcomes using /(-means clustering:

• Number of lapses (number of RT ³ 500ms)

• Slowest 10% of RT

• Fastest 10% of RT

• Mean of

• Median RT Prior to clustering the above PVT data was normalized for mean 0 and variance 1. The Good PVT group (G1) has lower RT and lower lapse count whereas the inferior or poor PVT group (G2) has relatively higher RT and higher lapse count. The Inventors formed a number of comorbidity groups based on the absence or presence of comorbidities (depression, anxiety and morbid obesity) individually or in combination. The groups formed are shown in Table 20 with their distribution in PVT outcome-based groups: G1 and G2.

Table 20: Comorbidity groups with depression, anxiety and morbid obesity

Different types of features were used in the study: PSG-based features, demographic features, and clinical features as set out in Table 21.

Table 21 : Different types of features used.

A number of different features classes were formed using different combinations of features to examine the impact of different sources of information on determining the vigilance performance of OSA patients as set out in Table 22. Table 22 - Exemplary Feature Classes

Higher Order Statistics features that are referred to in the above Table 22 can include calculating a higher order spectrum of the EEG signals. For example: EEG filtered using a band-pass filter of lower cut-off = 0.1 Hz and higher cut-off = 45Hz. Higher Order Spectrum of the EEG signal was computed using following parameters:

Epoch Length = 30s EEG sampling rate = 1024 Hz (512 Hz)

Bispectrum Segment Length = 1 s Lag length = 64 points FFT length = Sampling rate.

Correlation is computed between the diagonal slices from the two EEG electrodes for the overnight data to form the correlation time series. From the correlation time series following statistical features were computed for each subject separately for NREM sleep data, REM sleep data and WAKE stage data: Mean, Variance, Standard Deviation, Skewness and Kurtosis. A Bispectrum Slice Correlation time-series may then be computed for following pairs of the EEG channels.

To study Inter-hemispheric difference electrode pairs used are preferably:

F3-A1 & F4-A2 C3-A1 & C4-A2 O1-A1 & O2-A2

To study Intra-hemispheric differences electrode pairs used are preferably: F3-A1 & O1-A1 F3-A1 & C3-A1 C3-A1 & O1-A1

F4-A2&O2-A2 F4-A2 & C4-A2 C4-A2 & O2-A2

The dataset was split into M equal parts by randomly picking epoch numbers. Each part was considered as a new dataset and all the features including, Inter- and Intra- hemispheric Asynchrony Time Series, Non Gaussianity Score Index of Inter- and Intra- hemispheric Asynchrony Time Serie, Oxygen Desaturation (SpO2), and Heart Rate Variability were separately computed from each part.

In order to identify the best set of features with higher discriminatory power, Principal Component Analysis (PCA) was used, which can transform the higher dimensional input space to an orthogonal feature space which is significantly lower dimensional.

The Inventors performed PCA on each feature class defined in Table 21 to optimize the performance of the classifiers trained for classifying subjects into PVT groups: G1 and G2 defined above. A set of Logistic Regression Models (LRM) were trained, validated and tested for all the feature classes defined in Table 21 after applying PCA feature optimization.

The Inventors trained, validated and tested a set of Logistic Regression Models (LRM) for all the feature classes defined in Table 2 after applying PCA feature optimization.

The LRMs were trained, validated and tested using an outer-loop-inner-loop strategy described below. In the outer-loop overall dataset is split into K = 10 equal, non-overlapping folds. Each fold of data was stratified to have the same percentage of two classes of subjects as in the original overall dataset. The data from one outer fold ("Test Fold") were treated as the independent test set. The data from the remaining nine folds were used for training an Optimal Model and validating it in a process termed the Inner-Loop. Note that the Inner and Outer loops did not share any data. This process was repeated 10 times until all the folds in the outer loop is used just once as the Test Fold.

In the Inner-Loop associated with each outer Test Fold, data from the nine remaining folds is used to build, optimise and validate the models, through another cross-validation process. In this process data from the nine folds were again divided into K1 = 50 inner folds. Then data from 49 folds were used to train the model and data from the remaining fold was used to test the model. This process is repeated K1 = 50 times to so that each inner fold is used for testing exactly once. Once all 50 folds are completed, features that were nonsignificant (which have mean p-value less than predetermined threshold) are removed. With the remaining features, once again models were trained following the 50-fold cross validation approach. This process is repeated until the optimal features are obtained. Once the optimal features are obtained, one model from the 50 models was selected to be tested in the outer-loop (Test Fold). The outer-loop-inner-loop strategy allows us to test the models on independent partitions of data, which were not used in any manner to build the models undergoing testing. The Inventors used data from 576 patients for the study. As previously described, the patients were grouped into two PVT outcome-based groups G1 and G2 using k-means clustering. The data set was further grouped into comorbidity groups defined in Table 20.

As shown in Table 23, OSA patients were grouped into G1 and G2 based on their PVT outcomes and comorbidity groups with depression, anxiety and morbid obesity.

Table 23 - OSA patients grouped into G1 and G2 based on their PVT outcomes and comorbidity groups with depression, anxiety and morbid obesity.

In feature extraction, as described in section 2.6, the dataset with 576 patients was split into M= 3 parts to produce 576 x 3 = 1728 datasets. These datasets were then used to train LR models following PCA for feature reduction as described in section sections 2.6 and 2.7.

The performance of the classifiers is tabulated in Table 24 with their sensitivity, specificity and accuracy with and without each comorbidity group and their accuracy gain after excluding the respective comorbidity group. The classifier performance is arranged according to the descending order with respect to the accuracy gain without each comorbidity group.

Table 24 shows the classification performance of the LRM models trained / tested for classifying OSA patients with and without comorbidities into PVT outcome-based groups (G1 and G2) in presented in terms of sensitivity (sen), specificity(spe) , accuracy(acc) their respective performance gain/loss sorted with respect to the descending order of accuracy gain.

Table 24 Part 1/3

Table 24 Part 2/3

Table 24 Part 3/3

According to the results shown in Table 24, it is clear that classification performance of the LRM trained with data from OSA patients without anxiety and/or morbid obesity (comorbidity group 5) is significantly higher in both sensitivity and specificity. This difference is 9%-14% in sensitivity and 14% - 17% in specificity when compared with performance of LRM trained using OSA patients with anxiety and/or morbid obesity.

The classifiers trained with comorbidity groups 2 and 7 show 11% and 10% accuracy gains respectively after excluding the respective comorbidity groups. However, there is significant drop in specificity which is 6%-12% in both classifiers with respect to that of the comorbidity group 5 that shows the highest performance gain.

Table 25 sets out classification performance of the LRM models trained / tested for classifying OSA patients with and without co-morbidities into PVT outcome- based groups (G1 and G2) in presented in terms of sensitivity (sen), specificity(spe) and accuracy(acc).

Table 25

When OSA patients with anxiety and/or morbid obesity are used for training, it is evident that specificity degrades significantly. However, according to Table 25, classifiers trained using feature classes 1 and 4 produces relatively higher sensitivity when compared to the classifiers trained using other feature classes, for example 2,3 and 5.

The results show that OSA patients can be groups into PVT outcome-based classes with relatively higher sensitivity (72%) and specificity (72%). This performance was achieved by excluding OSA patients with comorbidities: (1) anxiety and/or morbid obesity, (2) depression and/or anxiety, and (3) morbid obesity from the training data. The reasons why these comorbidities deteriorate cognitive performance should be further investigated. As previously discussed, in the past the physician's decision to begin OSA treatment, for example, in respect of a subject was not straightforward. The PSG-provided Respiratory Disturbance Index (RDI) alone, which is insufficient for the purpose. At present clinicians rely on self-reported questionnaires such as ESS.

The physician has had to make a judgment call on how severe the effects of OSA are on a given patient. Embodiments discussed herein are believed to be helpful by providing an objective method to quantify neuropsychological manifestations of OSA.

The disclosures of each of the following references are incorporated herein in their entireties by reference.

References

[1] C. Iber, S. Ancoli-lsrael, A. L. Chesson, and S. F. Quan, The AASM manual for the scoring of sleep and associated events: Rules, terminology and technical specifications. Westchester, Illinois: American Academy of Sleep Medicine, 2007.

[2] S. H. Sheldon, R. Ferber, and M. H. Kryger, Principles and practice of pediatric sleep medicine. W.B. Saunders, 2005.

[3] R. S. Bucks, M. Olaithe, and P. Eastwood, "Neurocognitive function in obstructive sleep apnoea: A meta-review," Respirology, vol. 18, no. 1 , pp. 61- 70, 2013.

[4] D. Leger, "The cost of sleep-related accidents: A report for the national commission on sleep disorders research," Sleep, vol. 17, no. 1, pp. 84-93, 1994.

[5] M. A. Carskadon, "Guidelines for the multiple sleep latency test (MSLT): A standard measure of sleepiness," Sleep, vol. 9, no. 4, pp. 519-524, 1986.

[6] M. W. Johns, "A new method for measuring daytime sleepiness: The Epworth sleepiness scale," Sleep, vol. 14, no. 6, pp. 540-545, 1991.

[7] M. S. Aloia, J. T. Arnedt, J. D. Davis, R. L. Riggs, and D. Byrd, "Neuropsychological sequelae of obstructive sleep apnea-hypopnea syndrome: A critical review," Journal of the International Neuropsychological Society, vol. 10, no. 5, pp. 772-785, 2004.

[8] C. F. Salorio, D. A. White, J. Piccirillo, S. P. Duntley, and M. L. Uhles, "Learning, memory, and executive control in individuals with obstructive sleep apnea syndrome," Journal of Clinical and Experimental Neuropsychology, vol. 24, no. 1, pp. 93-100, 2002.

[9] P. M. Macey, L. A. Henderson, K. E. Macey, J. R. Alger, R. C. Frysinger, M. A. Woo, et al., "Brain morphology associated with obstructive sleep apnea," American Journal of Respiratory and Critical Care Medicine, vol. 166, no. 10, pp. 1382-1387, 2002.

[10] M. J. Morrell, D. W. McRobbie, R. A. Quest, A. R. C. Cummin, R. Ghiassi, and D. R. Corfield, "Changes in brain morphology associated with obstructive sleep apnea," Sleep Medicine, vol. 4, no. 5, pp. 451-454, 2003.

[11] N. Canessa, V. Castronovo, S. F. Cappa, M. S. Aloia, S. Marelli, and L. Ferini-Strambi, " Obstructive sleep apnea: Brain structural changes and neurocognitive function before and after treatment," American Journal of Respiratory and Critical Care Medicine, vol. 183, no. 10, pp. 1419-1426, 2011.

[12] P. M. Macey, M. K. Sarma, R. Nagarajan, R. Aysola, J. M. Siegel, R. M. Harper, et al., "Obstructive sleep apnea is associated with low GABA and high glutamate in the insular cortex," Journal of Sleep Research, vol. 25, no. 4, pp. 390-394, 2016.

[13] R. H. Fatouleh, L. C. Lundblad, P. M. Macey, D. K. McKenzie, L. A. Henderson, and V. G. Macefield, "Reversal of functional changes in the brain associated with obstructive sleep apnoea following 6 months of CPAP," Neuroimage: Clinical, vol. 7, pp. 799-806, 2015.

[14] U. R. Abeyratne, V. Swarnkar, C. Hukins, and B. Duce, "Interhemispheric asynchrony correlates with severity of respiratory disturbance index in patients with sleep apnea," IEEE Transactions on Biomedical Engineering, vol. 57, no. 12, pp. 2947-2955, 2010.

[15] V. Swarnkar, U. R. Abeyratne, and C. Hukins, "Inter-hemispheric asynchrony of the brain during events of apnoea and EEG arousals," Physiological Measurement, vol. 28, no. 8, pp. 869-880, 2007. [16] S. T. Mueller and B. J. Piper, "The psychology experiment building language (PEBL) and PEBL test battery," Journal of Neuroscience Methods, vol. 222, pp. 250-259, 2014.

[17] J. Dorrian, N. L. Rogers, and D. F. Dinges, "Psychomotor vigilance performance: Neurocognitive assay sensitive to sleep loss," in Sleep deprivation: Clinical issues, pharmacology, and sleep loss effects, C. A. Kushida, Ed. New York: Marcel Dekker, 2005, pp. 39-70.

[18] M. Basner and D. F. Dinges, "Maximizing sensitivity of the psychomotor vigilance test (PVT) to sleep loss," Sleep, vol. 34, no. 5, pp. 581-591, 2011.

[19] J. A. Hartigan, Clustering algorithms. New York: Wiley, 1975.

[20] P. J. Rousseeuw, "Silhouettes: A graphical aid to the interpretation and validation of cluster analysis," Journal of Computational and Applied Mathematics, vol. 20, pp. 53-65, 1987.

[21] S. Banoo, D. Bell, P. Bossuyt, A. Herring, D. Mabey, F. Poole, et ai, "Evaluation of diagnostic tests for infectious diseases: general principles," Nature Reviews Microbiology, vol. 4, pp. S21-S31, 2006.

[22] P. M. Macey, R. Kumar, M. A. Woo, E. M. Valladares, F. L. Yan-Go, and R. M. Harper, "Brain structural changes in obstructive sleep apnea," Sleep, vol. 31, no. 7, pp. 967-977, 2008.

[23] C.-C. Hsu, Z.-G. Cai, H. Mei, H.-C. Chiu, and C.-M. Lin, "Detecting the duration of incomplete obstructive sleep apnea events using interhemispheric features of electroencephalography," International Journal of Innovative Computing, Information and Control, vol. 9, no. 2, pp. 705-725, 2013.

[24] R. D. Chervin, J. W. Burns, and D. L. Ruzicka, "Electroencephalographic changes during respiratory cycles predict sleepiness in sleep apnea," American Journal of Respiratory and Critical Care Medicine, vol. 171 , no. 6, pp. 652-658, 2005.

[25] U. R. Abeyratne, V. Swarnkar, S. I. Rathnayake, and C. Hukins, "Sleep- stage and event dependency of brain asynchrony as manifested through surface EEG," in 29th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, Lyon, France, 2007, pp. 709-712.

Conditional language used herein, such as, among others, "can," "could," "might," "may," "e.g.," and the like, unless specifically stated otherwise, or otherwise understood within the context as used, is generally intended to convey that certain embodiments include, while other embodiments do not include, certain features, elements and/or states. Thus, such conditional language is not generally intended to imply that features, elements and/or states are in any way required for one or more embodiments or that one or more embodiments necessarily include logic for deciding, with or without author input or prompting, whether these features, elements and/or states are included or are to be performed in any particular embodiment.

Depending on the embodiment, certain acts, events, or functions of any of the methods described herein can be performed in a different sequence, can be added, merged, or left out altogether (e.g., not all described acts or events are necessary for the practice of the method). Moreover, in certain embodiments, acts or events can be performed concurrently, e.g., through multi-threaded processing, interrupt processing, or multiple processors or processor cores, rather than sequentially.

The various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the embodiments disclosed herein can be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. The described functionality can be implemented in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the disclosure.

The various illustrative logical blocks, modules, and circuits described in connection with the embodiments disclosed herein can be implemented or performed with a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general purpose processor can be a microprocessor, but in the alternative, the processor can be any conventional processor, controller, microcontroller, or state machine. A processor can also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration.

The blocks of the methods and algorithms described in connection with the embodiments disclosed herein can be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module can reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, a hard disk, a removable disk, a CD-ROM, or any other form of computer-readable storage medium known in the art. An exemplary storage medium is coupled to a processor such that the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium can be integral to the processor. The processor and the storage medium can reside in an ASIC. The ASIC can reside in a user terminal. In the alternative, the processor and the storage medium can reside as discrete components in a user terminal.

Any modules described herein of certain embodiments may be implemented as software modules, hardware modules, or a combination thereof. In general, the word "module," as used herein, can refer to logic embodied in hardware or firmware or to a collection of software instructions executable on a processor. Additionally, the modules or components thereof may be implemented in analog circuitry in some embodiments.

While the above detailed description has shown, described, and pointed out novel features as applied to various embodiments, it will be understood that various omissions, substitutions, and changes in the form and details of the devices or algorithms illustrated can be made without departing from the spirit of the disclosure. As will be recognized, certain embodiments of the inventions described herein can be embodied within a form that does not provide all of the features and benefits set forth herein, as some features can be used or practiced separately from others. The scope of certain inventions disclosed herein is indicated by the appended claims rather than by the foregoing description. All changes which come within the meaning and range of equivalency of the claims are to be embraced within their scope.

In compliance with the statute, the invention has been described in language more or less specific to structural or methodical features. The term “comprises” and its variations, such as “comprising” and “comprised of” is used throughout in an inclusive sense and not to the exclusion of any additional features.

It is to be understood that the invention is not limited to specific features shown or described since the means herein described comprises preferred forms of putting the invention into effect. The invention is, therefore, claimed in any of its forms or modifications within the proper scope of the appended claims appropriately interpreted by those skilled in the art.

Throughout the specification and claims (if present), unless the context requires otherwise, the term "substantially" or "about" will be understood to not be limited to the value for the range qualified by the terms.