CHARACTERIZING STIMULI RESPONSE TO DETECT SLEEP DISORDERS

Abstract

A method of determining a sleep phenotype of a subject includes defining a stimulus profile, defining a response profile, administering the stimulus profile to the subject, collecting response data of the subject in accordance with the response profile; comparing the response data to a set of reference data, and determining sleep phenotype based on the comparison of the response data to the reference data. The method is characterized by determining sleep phenotype based on measuring responses to administered stimuli rather than observing typical physiological data observed during periods of sleep.

Description

BACKGROUND OF THE INVENTION

1. Field of the Invention

The disclosed concept pertains to systems and methods for detecting and characterizing sleep disorders, and, in particular, to systems and methods that are less obtrusive than methods traditionally used for detecting and characterizing sleep disorders.

2. Description of the Related Art

A person's physiological response to an external stimulus such as sound, light, vibration and others can be used for several varying purposes such as evaluation of consciousness during anesthesia or tailoring of pharmaceutical treatment (e.g. insulin dose in diabetic patients). In the context of sleep, excess sensitivity to physical pain, caused by application of an external stimulus, has been shown to be related to sleep disorders. For example, subjects with primary insomnia have been shown to exhibit reduced thresholds to pain. In the context of light therapy used for insomnia treatment, responses to light stimulus has been used to determine the phase and amplitude of the circadian phase in order to define the optimal light therapy schedule.

The current gold standard for assessing sleep disorders is polysomnography (PSG). PSG consists of monitoring multiple bio-signals such as electroencephalograms (EEG), electrocardiograms (ECG), oxygen saturations (SpO2), electromyography (EMG) and body motion for one or two nights in a sleep lab, and requires the use of expensive equipment and trained personnel. Home monitoring solutions for sleep screening also exist, but home solutions still require detection of several bio-signals overnight with obtrusive equipment. In addition, phenotyping populations of healthy versus afflicted sleepers by quantifying differences in existing physiological signals may fail to present strong characterizations in any but the most severe or fully developed populations of afflicted sleepers.

Accordingly, there is room for improvement in methods and systems for detecting and characterizing sleep disorders.

SUMMARY OF THE INVENTION

Accordingly, it is an object of the present invention to provide, in one embodiment, a method of characterizing stimulus response for detecting sleep disorders using a plurality of subjects including a first group exhibiting a first sleep phenotype and a second group exhibiting a second sleep phenotype, the method comprising: defining a stimulus profile including a number of parameters for administering a selected stimulus to the subjects; defining a response profile, the response profile including a number of parameters for collecting data about a response of the subjects to administration of the selected stimulus according to the stimulus profile; administering the selected stimulus to the plurality of subjects according to the stimulus profile; collecting, in a controller, response data of the plurality of subjects according to the response profile and in response to the selected stimulus; determining, with the controller, a response difference between the first group and the second group based on the response data; and characterizing in the controller that the stimulus profile is effective based on a determination that the response difference is significant.

In another embodiment, a method of training a predictive machine learning system for sleep phenotype determination comprises: receiving training data for a plurality of training subjects, wherein the training data comprises subject stimulus data and subject response data for each of the training subjects, wherein the subject stimulus data for each training subject comprises data indicative of administration of a selected stimulus to the training subject according to the stimulus profile of a method of characterizing stimulus response for detecting sleep disorders, and wherein the subject response data for each training subject comprises data of the training subject obtained according to the response profile of the method of characterizing stimulus response for detecting sleep disorders and in response to the selected stimulus; and using the training data to train a predictive machine learning system to be able to predict the sleep phenotype of an unclassified subject based on unclassified subject stimulus data and unclassified subject response data, wherein the unclassified subject stimulus data comprises data indicative of administration of the selected stimulus to the unclassified subject according to the stimulus profile, and wherein the unclassified subject response data comprises data of the unclassified subject according to the response profile and in response to the selected stimulus.

In another embodiment, a phenotyping system for determining a sleep phenotype of a subject comprises: a controller configured to implement a predictive machine learning system, the controller being configured to: receive unclassified subject stimulus data and unclassified subject response data for an unclassified subject, wherein the unclassified subject stimulus data comprises data indicative of administration of a selected stimulus to the unclassified subject according to the stimulus profile of a method of characterizing stimulus response for detecting sleep disorders, and wherein the unclassified subject response data comprises data of the unclassified subject according to the response profile of the method of characterizing stimulus response for detecting sleep disorders and in response to the selected stimulus; and use the predictive machine learning system and the unclassified subject stimulus data and the unclassified subject response data to predict the sleep phenotype of the unclassified subject, wherein the predictive machine learning system has been previously trained using training data for a plurality of training subjects, wherein the training data comprises subject stimulus data and subject response data for each of the training subjects, wherein the subject stimulus data for each training subject comprises data indicative of administration of the selected stimulus to the training subject according to the stimulus profile, and wherein the subject response data for each training subject comprises data of the training subject obtained according to the response profile and in response to the selected stimulus.

In yet another embodiment, a sleep phenotype determination method comprises: receiving unclassified subject stimulus data and unclassified subject response data for an unclassified subject, wherein the unclassified subject stimulus data comprises data indicative of administration of a selected stimulus to the unclassified subject according to the stimulus profile of a method of characterizing stimulus response for detecting sleep disorders, and wherein the unclassified subject response data comprises data of the unclassified subject according to the response profile of the method of characterizing stimulus response for detecting sleep disorders and in response to the selected stimulus; and providing the unclassified subject stimulus data and the unclassified subject response data to a predictive machine learning system and using the predictive machine learning system and the unclassified subject stimulus data and the unclassified subject response data to predict the sleep phenotype of the unclassified subject, wherein the predictive machine learning system has been previously trained using training data for a plurality of training subjects, wherein the training data comprises subject stimulus data and subject response data for each of the training subjects, wherein the subject stimulus data for each training subject comprises data indicative of administration of the selected stimulus to the training subject according to the stimulus profile of the method of characterizing stimulus response for detecting sleep disorders, and wherein the subject response data for each training subject comprises data of the training subject obtained according to the response profile of the method of characterizing stimulus response for detecting sleep disorders and in response to the selected stimulus.

These and other objects, features, and characteristics of the present invention, as well as the methods of operation and functions of the related elements of structure and the combination of parts and economies of manufacture, will become more apparent upon consideration of the following description and the appended claims with reference to the accompanying drawings, all of which form a part of this specification, wherein like reference numerals designate corresponding parts in the various figures. It is to be expressly understood, however, that the drawings are for the purpose of illustration and description only and are not intended as a definition of the limits of the invention.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a flow chart containing the steps of a method representing the learning phase of a process for determining sleep phenotypes of subjects in accordance with an exemplary embodiment of the disclosed concept;

FIG. 2 is a table of example stimulus profiles that can be used in the method represented in FIG. 1;

FIG. 3 is a flow chart containing the steps of a method for training a machine learning model to determine a sleep phenotype of a subject in accordance with an exemplary embodiment of the disclosed concept;

FIG. 4 is a schematic diagram of a phenotyping system which incorporates the machine learning model trained by the process depicted in the flow chart in FIG. 3 in accordance with an exemplary embodiment of the disclosed concept; and

FIG. 5 is a flow chart of a process by which the phenotyping system depicted in FIG. 4 determines a sleep phenotype of a subject in accordance with an exemplary embodiment of the disclosed concept.

DETAILED DESCRIPTION OF EXEMPLARY EMBODIMENTS

As used herein, the singular form of “a”, “an”, and “the” include plural references unless the context clearly dictates otherwise.

As used herein, the statement that two or more parts or components are “coupled” shall mean that the parts are joined or operate together either directly or indirectly, i.e., through one or more intermediate parts or components, so long as a link occurs.

As used herein, the term “number” shall mean one or an integer greater than one (i.e., a plurality).

As used herein, the term “controller” shall mean a number of programmable analog and/or digital devices (including an associated memory part or portion) that can store, retrieve, execute and process data (e.g., software routines and/or information used by such routines), including, without limitation, a field programmable gate array (FPGA), a complex programmable logic device (CPLD), a programmable system on a chip (PSOC), an application specific integrated circuit (ASIC), a microprocessor, a microcontroller, a programmable logic controller, or any other suitable processing device or apparatus. The memory portion can be any one or more of a variety of types of internal and/or external storage media such as, without limitation, RAM, ROM, EPROM(s), EEPROM(s), FLASH, and the like that provide a storage register, i.e., a non-transitory machine readable medium, for data and program code storage such as in the fashion of an internal storage area of a computer, and can be volatile memory or nonvolatile memory.

As used herein, the term “afflicted sleeper” shall mean a person with unsatisfactory sleep characteristics, identified by either him- or herself or an objective third party, such as short sleep (less than six hours), inability to fall or stay asleep, disruptions in sleep architecture or continuity, atypical SpO2 oxygen levels, or any other characteristic being investigated by a diagnostic implementation of a proposed system described in the present disclosure. Afflicted sleepers may alternatively be referred to “afflicted subjects”.

As used herein, the term “healthy sleeper” shall mean the converse of afflicted sleeper, a person lacking abnormal sleep characteristics being investigated by the diagnostic implementation of a proposed system described in the present disclosure. Healthy sleepers may alternatively be referred to “healthy subjects”.

As used herein, the term “machine learning model” or “machine learning system” or “predictive machine learning system” shall mean a software system that develops and builds a mathematical model based on sample data, known as “training data”, in order to make predictions or decisions without being explicitly programmed to do so, including, without limitation, a computer software system that has been trained to recognize patterns from a set of training data, and subsequently develops algorithms to recognize patterns from the training data set in other data sets.

As used herein, the terms “sleep arousal” or “arousal” shall mean an abrupt shift of EEG frequency including alpha, theta, and/or frequencies greater than 16 Hz, excluding spindles, that lasts at least 3 seconds, with at least 10 seconds of stable sleep preceding the change.

As used herein, the term “sleep phenotype” shall mean an observable pattern of sleep characteristics used to categorize one or more groups of people being investigated by an implementation of a proposed system described in the present disclosure, including but not limited to metrics describing sleep architecture, duration, disturbances, or other physiological parameters such as oxygen saturation (SpO2), resting heartrate, and movements while asleep.

Directional phrases used herein, such as, for example and without limitation, top, bottom, left, right, upper, lower, front, back, and derivatives thereof, relate to the orientation of the elements shown in the drawings and are not limiting upon the claims unless expressly recited therein.

The disclosed concept, as described in greater detail herein in connection with various particular exemplary embodiments, provides improved methods and systems for the detection and characterization of sleep phenotypes in subjects such as sleep study subjects. The present disclosure is directed toward determining a subject's sleep phenotype by providing a number of stimuli to the subject and measuring the subject's responses to the stimuli. The disclosed concept is in contrast with known methods for detecting sleep disorders that measure typical non-stimulated physiological signals. Referring to FIG. 1, a flow chart for a method 100 representing the learning phase of a process for determining sleep phenotypes of subjects is shown. The method 100 may be used, for example, in connection with an exemplary process 200 that is described in further detail herein with respect to FIG. 3. Referring to FIG. 1, at 101, a group of sleep study subjects is compiled from subjects who exhibit one of two predetermined sleep phenotypes (but not both), the subjects having previously been categorized as healthy sleepers (a first phonotype) or afflicted sleepers (a second phonotype) using traditional sleep study PSG methods, e.g., by monitoring multiple bio-signals such as EEGs, ECGs, SpO2, EMGs, and body motion in a sleep lab. Other steps of method 100 may also employ PSG methods, as the primary utility of method 100 lies in how it is employed in process 200, which subsequently eliminates the need for PSG methods to determine sleep phenotype. For economy of disclosure, the steps of method 100 are initially described herein with respect to the group chosen at 101 comprising healthy sleepers and afflicted sleepers, but it will be appreciated that the group chosen at 101 can be comprised entirely of afflicted sleepers with different affliction types, i.e., different sleep phenotypes, as described subsequently herein without departing from the scope of the disclosed concept. The number of subjects included in the group selected at 101 can vary, but it will be appreciated that in general, collecting data from a greater number of subjects will yield more meaningful results during the learning phase than collecting data from a lesser number of subjects.

At 102, a stimulus profile is defined. The stimulus profile defines the type of stimulus to be administered to a subject and an evaluation function for the stimulus. Non-limiting examples of the types of stimuli that can be administered to a subject include: auditory, visual, and ingestible stimuli. The evaluation function defines the parameters of how the stimulus is to be administered to the subject. With respect to an auditory stimulus, the evaluation function can, for example and without limitation, define the length of the tones in the auditory sequence to be played (such as shorter pulsed tones spaced an interval of time apart or sustained tones), the frequency/pitch of the tones, the volume of the tones, and/or whether the sequence is played before, during, and/or after sleep onset. With respect to a visual stimulus, the evaluation function can, for example and without limitation, define the types or wavelengths of light (such as blue light) to which the subject would be exposed, the duration of exposure, the variability of exposure to the wavelength (such as constant brightness or variable brightening and dimming over the duration of the exposure), and/or whether the light exposure occurs before, during, and/or after sleep onset. With respect to an ingestible stimulus the evaluation function can, for example and without limitation, define a quantity of a type of ingestible (such as caffeine) to be consumed and define a time of day or a number of hours before typical sleep onset at which the ingestible should be consumed. At a minimum, in an exemplary embodiment, a meaningful evaluation function may include temporal attributes of the stimulus. An optimal stimulus is designed to maximize the differentiation of healthy versus afflicted sleepers, and the attributes of an optimal stimulus profile can be found using classical optimization methods or using a learning system such as a closed-loop reinforcement learning model using rewards which maximize cluster differences between the target populations, i.e., healthy sleepers and afflicted sleepers. It will be appreciated that a stimulus profile can be adapted for highly sensitive afflicted sleepers such as subjects with severe insomnia by, for example and without limitation, reducing the magnitude or intensity of the stimulus and increasing the duration over which the stimulus is administered (such as three nights rather than one) to sensitive subjects as compared to non-sensitive sleepers.

At 103, a response profile is defined. The response profile is a set of parameters defining how the physiological responses induced in a subject by the stimulus profile defined at 102 should be measured. In one non-limiting example, it can be determined that, if the stimulus profile includes an audible stimulus, one of the characteristics that should be included in the response profile is an alpha-band EEG measured at anatomical landmark Fpz (just above the bridge of the nose) for 50 ms after the audible stimulus is administered. In another non-limiting example, it can be determined that, if the stimulus profile includes an ingestible stimulus of 200 mg of caffeine to be ingested by the subject at 6:00 pm, one of the characteristics that should be included in the response profile is duration of average sleep cycle measured during the evening following the caffeine consumption. The response profile can also define parameters to adjust for the influence of confounding factors on the physiological responses of the subjects to the stimulus profiles. Non-limiting examples of confounding factors includes background noise, age of the subjects, and physical characteristics of the subjects. For example and without limitation, the response profile could prescribe that background noise be filtered out from the subject's environment during the administration of the stimulus profile and observation of the subject's response to the stimulus profile, and/or that the number of detected arousals be normalized by the age of the subjects. The response profile can be optimized using classical optimization methods or a learning system such as closed-loop reinforcement learning. For example and without limitation, maximum likelihood estimation may show for a particular stimulus profile that observing 75 ms of an EEG differentiates healthy sleepers from afflicted sleepers more effectively than observing 150 ms or 50 ms of an EEG does.

The stimulus profile and response profile can be strategically defined at steps 102 and 103 of method 100 to detect and characterize specific afflictions, as opposed to simply detecting the presence of a general affliction. In a first non-limiting example, at step 102 a stimulus profile designed to detect sleep onset insomnia and/or maintenance insomnia prescribes that subjects be administered at least one of the following types of stimuli: sub-threshold audible tones, external vibratory therapy, light therapy, or inner-ear electrical stimulation. In the same first example, at step 103 the response profile defined for a stimulus profile designed to detect sleep onset insomnia and/or maintenance insomnia prescribes characterization of sleep EEG signals in each subject, for example and without limitation: spindles, arousals, depth, etc. In a second non-limiting example, at step 102 a stimulus profile designed to differentiate chronic sleep restriction from insomnia prescribes that subjects be administered at least one of: light therapy followed by dim-light onset, or relaxation programs/methods (e.g., sound relaxation sequences, paced breathing exercises, and mindfulness exercises). In the same second example, at step 103 the response profile defined for a stimulus profile designed to differentiate chronic sleep restriction from insomnia prescribes characterization of wake after sleep onset (WASO), sleep onset latency (SOL), total sleep time (TST), and/or sleep survival in each subject.

At 104, the stimulus profile from 102 is administered to both the healthy subjects and the afflicted subjects. At 105, data about the responses of the subjects to the stimulus profile is collected according to the response profile defined at 103. At 106, a response difference for the entire group of subjects is determined by finding the average number of sleep arousals among the healthy subjects and the average number of sleep arousals among the afflicted subjects, finding the difference between the average of the healthy subjects and the average of the afflicted sleepers, and finding the square of the difference, as shown in expression (1) below:

Δ_pr=(a_h−a_u)_pr²  (1)

where a_h is the average number of sleep arousals among healthy sleepers, a_u is the number of sleep arousals among afflicted sleepers, pr denotes a particular stimulus profile, and Δ_pr is the response difference. The closer to zero the value of Δ_pr is, the lower the difference in the average number of arousals between the healthy subjects and the afflicted subjects is and the less useful the particular profile pr is for differentiating between healthy and afflicted sleepers. In contrast, the higher the value of Δ_pr is, the higher the difference in the average number of arousals between the healthy subjects and the afflicted subjects is and the more useful the particular profile pr is for differentiating between healthy and afflicted sleepers. An optimal stimulus profile will cause zero arousals and ideally zero changes to sleep activity in healthy sleepers and one or more arousals in all afflicted sleepers. It will be appreciated that the value of Δ_pr that provides a meaningful delineation between healthy sleepers and afflicted sleepers can vary widely depending on what stimulus profile is defined at 102 and that choosing an appropriate value for Δ_pr is highly specific to each stimulus profile.

Step 106 is particularly significant when the efficacies of similar stimulus profiles in bifurcating a group of subjects into healthy and afflicted sleepers are being compared. Referring to FIG. 2, in one non-limiting exemplary embodiment, sound can be used as the primary stimulus, and several different example sound profiles that can be used as the stimulus profile in step 102 are shown in Table 1 of FIG. 2. The values used in Table 1 are chosen for illustrative purposes and it will be appreciated that other values can be selected instead of the values shown in Table 1 without departing from the scope of the disclosed concept. Each sound profile in Table 1 consists of a volume (measured in dB) and a pitch (measured in Hz), with no two profiles having the same volume and pitch. Finding the response difference at step 106 of method 100 for each sound profile can determine which sound profile produces the maximum response difference and is accordingly the most effective in characterizing subjects as either healthy sleepers or afflicted sleepers. Gradient descent optimization is one non-limiting example of a method that can be used to find the stimulus profile that produces the maximum response difference among a group of stimulus profiles, but any optimization method can be used to find the maximum response difference without departing from the scope of the disclosed concept.

At 107, the response difference found at 106 is determined to be either significant or insignificant depending on what value of Δ_pr is chosen to be determinative for a given stimulus profile and whether the response difference found at 106 is greater than or lower than the chosen determinative value of Δ_pr, a response difference greater than the chosen determinative value of Δ_pr, being significant and a response difference lesser than the chosen determinative value of Δ_pr, being insignificant. If the response difference for the stimulus profile is determined to be significant, then the stimulus profile is characterized as effective at 108. If the response difference for the stimulus profile is determined to be insignificant, the stimulus profile is characterized as ineffective and the method can return to 102 to either incrementally alter the existing stimulus profile or define an entirely new stimulus profile.

Method 100 can be tailored to produced highly nuanced results. In one non-limiting example, rather than selecting subjects for the group at step 101 based merely on whether they are healthy sleepers or afflicted sleepers, afflicted sleepers can be chosen based on affliction phenotype, including but not limited to: sleep onset insomnia, maintenance insomnia, and chronic sleep restriction. Furthermore, as previously stated with respect to step 101, in an additional exemplary embodiment, instead of including healthy subjects and afflicted subjects in the group at step 101, two different types of afflicted subjects can be included in the group (i.e., with subjects exhibiting either one of two selected sleep phenotypes, but not both) such that the determination of the response difference at 106 and the determination of the significance of the response difference at 107 can be used to determine the efficacy of a stimulus profile in differentiating subjects exhibiting a particular type of sleep affliction from subjects exhibiting another type of sleep affliction, as opposed to differentiating healthy sleepers from afflicted sleepers. In another non-limiting exemplary embodiment, instead of using sleep arousals to differentiate between the two types of sleepers in the group and determine the response difference at 106, other parameters such as sleep continuity, frequency of rousing, and length of deep sleep periods can be evaluated without departing from the scope of the disclosed concept.

In one non-limiting exemplary embodiment of method 100, sound is selected at 102 as the main stimulus to discriminate between healthy sleepers and afflicted sleepers, and the evaluation function of the stimulus profile prescribes that several different sound profiles (such as the sound profiles defined in Table 1 of FIG. 2) all be transmitted to each subject at different times (e.g., each sound profile is administered to a given subject on a different day or night than every other sound profile). At 103, the response profile is defined to use EEG signals to measure the physiological responses induced in the subjects, and changes of the EEG waves in amplitude and power are used to quantify the physiological response in each subject. Non-limiting examples of specific characteristics that can be extracted from the EEG signals and used to quantify the physiological response of each subject to each sound profile include number of sleep arousals (where arousal indicates an abrupt change in the pattern of brain wave activity) and absolute value or variation in amplitude or power of EEG waves in different frequency bands. The response profile can also optionally define parameters to adjust for the influence of confounding factors (e.g., background noise, age of the subjects, and physical characteristics of the subjects) on the physiological responses of the subjects to the sound profiles. For example and without limitation, as previously described with respect to step 103, the response profile could prescribe that background noise be filtered out from the subject's environment during administration of the sound profiles and observation of the subject's response to the sound profiles, and/or that the number of detected arousals be normalized by the age of the subject.

Method 100 represents an improvement to existing sleep phenotyping methods for its ability to be incorporated into machine learning models for automation of sleep phenotyping. Referring to FIG. 3, a flow chart of a process 200 for training a machine learning model to determine a sleep phenotype of a subject according to an exemplary embodiment is shown. At 201, a learning phase is executed, the learning phase being method 100 described herein with respect to FIG. 1. At 202, classifier training is performed, wherein, for each stimulus profile characterized as effective at step 108 of method 100, the machine learning model is provided with the data obtained from the learning phase, referred to as training data or reference data, such that the machine learning model can analyze the stimulus profiles defined at step 102 along with the response data collected at step 105 in order to detect patterns indicative of specific sleep phenotypes. The subjects to whom the stimulus profiles are administered and whose responses to the stimulus profiles are collected according to the response profiles during the learning phase at step 201 may be referred to as training subjects. In addition, non-phenotype data about each training subject such as, for example and without limitation, demographic, physical characteristics, and comorbidities, can also be provided to the machine learning model at step 202 so that the machine learning model can analyze said non-phenotype data in order to determine whether a relationship exists between the provided non-phenotype data and sleep phenotype of each training subject. At 203, validation is performed, wherein unclassified subject stimulus data and unclassified subject response data collected in accordance with step 105 of method 100 is provided to the machine learning model for the machine learning model to analyze in order to make a determination of sleep phenotype. The unclassified subject stimulus data and unclassified subject response data are not part of the training data set. For a given unclassified subject, the unclassified subject stimulus data comprises data about the administration of a stimulus profile to the subject, and the unclassified subject response data comprises data obtained according to a corresponding response profile about the response of the subject to the administered stimulus profile. After the machine learning model makes a determination of sleep phenotype based on the unclassified subject stimulus data and the unclassified subject response data, the machine learning model's performance is evaluated manually by sleep engineers, physicians, or other qualified personnel. In the present disclosure, a machine learning model that has undergone training, such as through a process such as process 200, is referred to as a trained predictive machine learning system, as the trained predictive machine learning model is considered capable of determining sleep phenotypes and other associated characteristics with a satisfactory degree of accuracy when provided with unclassified subject stimulus data and unclassified subject response data.

Step 203 of method 200 is closely related to how the response profile is defined in step 103 of method 100, i.e. the learning phase of method 200 executed at step 201. In another non-limiting exemplary embodiment of method 100, at step 103 the response profile prescribes measuring multiple types (as opposed to a single type) of physiological signals of a subject's response to a stimulus response, for example and without limitation, heart rate variability, SpO2, and body motion, as well as EEG. Using multiple types of physiological signals to evaluate the response of a subject to a stimulus profile will likely lead to better detection and characterization of sleep phenotype in a subject during the learning phase than using a single type. When validation of the machine learning model is performed at step 203 of method 200, data about the response of a subject to a stimulus profile is provided to the machine learning model to analyze in order to make a determination of sleep phenotype. It is expected that, if data from multiple physiological signals are compiled as training data to use during classifier training at step 202 of method 200, the machine learning model will be able to accurately determine and characterize sleep phenotype even if the response data collected for validation at step 203 is collected using data from only one of the types of physiological signals or a subset of the types of physiological signals used to compile the training data used at step 202.

FIG. 4 is a schematic representation of a phenotyping device 10 in accordance with an exemplary embodiment of the disclosed concept which incorporates a trained predictive machine learning system as described herein, such as a machine learning model 20 trained by process 200 as described with respect to FIG. 3. FIG. 5 is a flow chart of a process 300 by which phenotyping device 10 in FIG. 4 determines a sleep phenotype of a subject in accordance with an exemplary embodiment of the disclosed concept. Phenotyping device 10 includes an input apparatus 12 (e.g., a plurality of buttons), a display 14, a stimulus apparatus 15 (e.g., a number of speakers for providing auditory stimulus tones and/or a light source for producing visual stimuli), a response collection apparatus 16 (e.g., a plurality of sensors configured to be affixed to a subject, the sensors being electrically connected via a wired or wireless connection to controller 18), and a controller 18 into which machine learning model 20 is incorporated as a phenotyping analysis software application that is executed by controller 18. A user is able to provide input to controller 18 using input apparatus 12, and controller 18 provides output signals to display 14 to enable display 14 to display information to the user (e.g., a clinician) as described in detail herein. A memory portion of controller 18 has stored therein a number of routines that are executable by a processor portion of controller 18. One or more of the routines implement (by way of computer/processor executable instructions) the phenotyping analysis software, which is configured (by way of one or more algorithms) to, among other things, instruct stimulus apparatus 15 to output a stimulus profile (selected from a number of stimulus profiles stored in the memory portion of controller 18) that is administered to a subject.

At step 301 of process 300, the user of phenotyping device 10 selects a stimulus profile (such as the stimulus profiles described with respect to step 102 of method 100) to be administered to the subject and inputs the selection to phenotyping device 10 via input apparatus 12. At step 302 of process 300, stimulus apparatus 15 outputs the stimulus profile to the subject. As the selected stimulus profile is administered to the subject, response collection apparatus 16 senses the response of the subject (e.g., bio-signal data) relevant to the selected stimulus profile and transmits the response data to the phenotyping analysis software application at a step 303 of process 300. To equip phenotyping device 10 with this capability, specialized processing chips may need to be embedded as part of controller 18, to which the response data sensed by response collection apparatus 16 will be fed. At a step 304, the phenotyping analysis software application determines a sleep phenotype of the subject using the reference data provided to machine learning model 20 during the classifier training step 202 of process 200.

The applications of a machine learning model trained in accordance with method 200 are wide-ranging. In one non-limiting example, a machine learning model trained by method 200 can be incorporated into a controller of a computer system configured with equipment to administer various stimuli to sleep study subjects and record the results of the stimuli administration so that the controller can automatically commence determining a subject's sleep phenotype as soon as data collection is complete. In another non-limiting example, to the extent that systems for at-home measurement of a subject's sleep data (such as ambulatory EEG systems) are available or become available, the sleep data can be entered into or collected by a device application such as a personal computer software program or mobile phone application that incorporates the machine learning model trained by method 200 and used to determine the subject's sleep phenotype at home instead of a laboratory.

The impact of availability of at-home systems to determine sleep phenotype can be especially significant in contexts such as at-home continuous positive airway pressure (CPAP) therapy compliance, since the flow of pressurized air produced by CPAP machines is noticeably audible and can alter the quality of sleep for subjects whose sleep phenotype makes them prone to auditory sleep disturbances. For example and without limitation, if a subject uses an ambulatory EEG system while undergoing his or her overnight CPAP therapy and the EEG data collected and analyzed by an at-home device application (incorporating a machine learning model trained by method 200) indicates that the subject exhibits a sleep phenotype that is incompatible with the noise level of the CPAP machine settings, a care provider may be able to tailor the recommended CPAP settings to lessen the auditory disturbance to the subject or to suggest a different patient interface that alters the sound profile of the recommended CPAP settings in order to decrease disturbances to the subject's sleep quality and ensure compliance with the recommended CPAP regimen.

In addition to the non-limiting exemplary embodiments for determining and characterizing sleep phenotypes already described, an additional non-limiting embodiment of the disclosed concept couples the exemplary embodiments for determining sleep phenotype with a recommendation system that suggests specific treatment solutions to address the effects experienced by subjects with afflicted phenotypes, such treatment solutions including but not limited to devices and/or programs for lifestyle changes for improving sleep quality. In this non-limiting exemplary embodiment, method 100 constitutes the learning phase of a process for determining treatment solutions for mitigating the effects of affected sleep phenotypes in addition to detecting the sleep phenotypes of subjects. Steps 101 through 108 as previous described in the context of executing method 100 solely to detect sleep phenotypes of subjects are all performed in substantially the same manner in the additional context of determining solutions for mitigating the effects of affected sleep phenotypes, but steps 101 and 102 are still described in more detail below for illustrative purposes.

At 101, a group of subjects is compiled from a first sub-group of subjects who exhibit a predetermined afflicted sleep phenotype A (with phenotype A being classified as a disorder) and benefit most from a first type of treatment X, and a second sub-group of subjects who exhibit sleep phenotype A and benefit most from a second type of treatment Y, the subjects having previously been categorized as phenotype A and benefiting most from treatment X or Y using traditional sleep study PSG methods. The group of subjects could also be compiled from one of the sub-groups of subjects who exhibit phenotype A and a second sub-group compiled from healthy sleepers without departing from the scope of the disclosed concept. At 102, the stimulus profile is defined. An optimal stimulus in the context of determining solutions to mitigate the effects of a sleep disorder will maximize the differentiation between both the type of sleep disorder/phenotype and the best treatment for the disorder. Accordingly, the phenotyping based on stimulus response can be extended to the detection of best therapeutic and/or pharmacological treatment for specific detected sleep disorders. One non-limiting example of treatment differentiation that can be made is receptor antagonist or trazodone (SSRI) versus a GABA receptor agonist. In an embodiment where method 100 is executed to determine solutions for mitigating the effects of afflicted sleep phenotypes in addition to determining sleep phenotype, process 200 proceeds as previously described in training a machine learning model to determine sleep phenotypes and additionally trains the machine learning model to determine (best) solutions for mitigating the effects of afflicted phenotypes and/or whether said solutions are appropriate for use by a particular subject. In addition, process 300 by which the phenotyping device 10 of FIG. 4 determines a sleep phenotype of a subject also proceeds as previously described, and at step 304, the phenotyping analysis software application determines (best) solutions for mitigating the effects of afflicted phenotypes in addition to determining sleep phenotype, as well as whether said solutions are appropriate for use by a particular subject.

In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word “comprising” or “including” does not exclude the presence of elements or steps other than those listed in a claim. In a device claim enumerating several means, several of these means may be embodied by one and the same item of hardware. The word “a” or “an” preceding an element does not exclude the presence of a plurality of such elements. In any device claim enumerating several means, several of these means may be embodied by one and the same item of hardware. The mere fact that certain elements are recited in mutually different dependent claims does not indicate that these elements cannot be used in combination.

Although the invention has been described in detail for the purpose of illustration based on what is currently considered to be the most practical and preferred embodiments, it is to be understood that such detail is solely for that purpose and that the invention is not limited to the disclosed embodiments, but, on the contrary, is intended to cover modifications and equivalent arrangements that are within the spirit and scope of the appended claims. For example, it is to be understood that the present invention contemplates that, to the extent possible, one or more features of any embodiment can be combined with one or more features of any other embodiment.

Claims

1. A method of characterizing stimulus response for detecting sleep disorders using a plurality of subjects including a first group exhibiting a first sleep phenotype and a second group exhibiting a second sleep phenotype, the method comprising: defining a stimulus profile including a number of parameters for administering a selected stimulus to the subjects;defining a response profile, the response profile including a number of parameters for collecting data about a response of the subjects to administration of the selected stimulus according to the stimulus profile;administering the selected stimulus to the plurality of subjects according to the stimulus profile;collecting, in a controller, response data of the plurality of subjects according to the response profile and in response to the selected stimulus;determining, with the controller, a response difference between the first group and the second group based on the response data; andcharacterizing in the controller that the stimulus profile is effective based on a determination that the response difference is significant.

2. The method of claim 1, further comprising: choosing a determinative value for the response difference,wherein a determination is made that the response difference is significant if the response difference is greater than the determinative value, andwherein a determination is made that the response difference is insignificant if the response difference is not greater than the determinative value.

3. The method of claim 1, wherein the first phenotype is a healthy sleep phenotype, andwherein the second phenotype is an afflicted sleep phenotype.

4. The method of claim 1, wherein the first phenotype is a first afflicted sleep phenotype, andwherein the second phenotype is a second afflicted sleep phenotype.

5. The method of claim 1, further comprising: receiving unclassified subject stimulus data and unclassified subject response data for an unclassified subject, wherein the unclassified subject stimulus data comprises data indicative of administration of the selected stimulus to the unclassified subject according to the stimulus profile, and wherein the unclassified subject response data comprises data of the unclassified subject collected according to the response profile and in response to the selected stimulus; andusing the unclassified subject stimulus data and the unclassified subject response data to predict the sleep phenotype of the unclassified subject.

6. A method of training a predictive machine learning system for sleep phenotype determination, the method comprising: receiving training data for a plurality of training subjects, wherein the training data comprises subject stimulus data and subject response data for each of the training subjects, wherein the subject stimulus data for each training subject comprises data indicative of administration of the selected stimulus of claim 1 to the training subject according to the stimulus profile of claim 1, and wherein the subject response data for each training subject comprises data of the training subject obtained according to the response profile of claim 1 and in response to the selected stimulus; andusing the training data to train a predictive machine learning system to be able to predict the sleep phenotype of an unclassified subject based on unclassified subject stimulus data and unclassified subject response data, wherein the unclassified subject stimulus data comprises data indicative of administration of the selected stimulus of claim 1 to the unclassified subject according to the stimulus profile of claim 1, and wherein the unclassified subject response data comprises data of the unclassified subject according to the response profile of claim 1 and in response to the selected stimulus.

7. The method of claim 6, wherein the selected stimulus comprises a number of auditory stimuli, the number of auditory stimuli comprising a number of tones.

8. The method of claim 7, wherein the parameters of the stimulus profile provide for a variance in at least one of a length of the number of tones, a pitch of the number of tones, a volume of the number of tones, and commencement of playing of the number of tones with respect to an onset of a sleep state of the subject.

9. The method of claim 6, wherein the selected stimulus comprises a number of visual stimuli, the visual stimuli comprising a number of wavelengths of light.

10. The method of claim 9, wherein the parameters of the stimulus profile provide for a variance in at least one of a length of the number of wavelengths of light, a duration of exposure of the subject to the number of wavelengths of light, an illuminance of the number of wavelengths of light, and commencement of the exposure of the subject to the number of wavelengths of light with respect to an onset of a sleep state of the subject.

11. The method of claim 6, wherein the number of parameters of the response profile comprises parameters for measurement of electroencephalogram (EEG) data.

12. A phenotyping system for determining a sleep phenotype of a subject, the system comprising: a controller configured to implement a predictive machine learning system, the controller being configured to:receive unclassified subject stimulus data and unclassified subject response data for an unclassified subject, wherein the unclassified subject stimulus data comprises data indicative of administration of the selected stimulus of claim 1 to the unclassified subject according to the stimulus profile of claim 1, and wherein the unclassified subject response data comprises data of the unclassified subject according to the response profile of claim 1 and in response to the selected stimulus; anduse the predictive machine learning system and the unclassified subject stimulus data and the unclassified subject response data to predict the sleep phenotype of the unclassified subject, wherein the predictive machine learning system has been previously trained using training data for a plurality of training subjects, wherein the training data comprises subject stimulus data and subject response data for each of the training subjects, wherein the subject stimulus data for each training subject comprises data indicative of administration of the selected stimulus of claim 1 to the training subject according to the stimulus profile of claim 1, and wherein the subject response data for each training subject comprises data of the training subject obtained according to the response profile of claim 1 and in response to the selected stimulus.

13. The phenotyping system of claim 12, further comprising: an input apparatus configured to accept user input;a stimulus apparatus configured to administer the selected stimulus to the unclassified subject according to the stimulus profile; anda response collection apparatus configured to collect unclassified response data of the unclassified subject according to the response profile and to transmit the unclassified response data to the controller.

14. The phenotyping system of claim 12, wherein the stimulus apparatus is configured to administer a number of auditory stimuli to the unclassified subject.

15. The phenotyping system of claim 12, wherein the stimulus apparatus is configured to administer a number of visual stimuli to the unclassified subject.

16. The phenotyping system of claim 13, wherein the response collection apparatus comprises EEG sensors.

17. The phenotyping system of claim 13, wherein the response collection system is configured to collect unclassified response data of the unclassified subject according only to a subset of the response profile,wherein the predictive machine learning system is configured to predict the sleep phenotype of the unclassified subject using the unclassified subject stimulus data and unclassified subject response data collected by the response collection system according only to a subset of the response profile.

18. The phenotyping system of claim 12, further comprising: wherein the training data used to train the predictive machine learning system comprises subject stimulus data and subject response data for each of the training subjects indicative of the selected stimulus comprising a treatment solution,wherein the predictive machine learning system is configured to predict an appropriateness of the treatment solution for use by the unclassified subject, andwherein predicting the sleep phenotype of the unclassified subject using the predictive machine learning system includes predicting the appropriateness of the treatment solution for use by the unclassified subject.

19. A sleep phenotype determination method, comprising: receiving unclassified subject stimulus data and unclassified subject response data for an unclassified subject, wherein the unclassified subject stimulus data comprises data indicative of administration of the selected stimulus of claim 1 to the unclassified subject according to the stimulus profile of claim 1, and wherein the unclassified subject response data comprises data of the unclassified subject according to the response profile of claim 1 and in response to the selected stimulus; andproviding the unclassified subject stimulus data and the unclassified subject response data to a predictive machine learning system and using the predictive machine learning system and the unclassified subject stimulus data and the unclassified subject response data to predict the sleep phenotype of the unclassified subject, wherein the predictive machine learning system has been previously trained using training data for a plurality of training subjects, wherein the training data comprises subject stimulus data and subject response data for each of the training subjects, wherein the subject stimulus data for each training subject comprises data indicative of administration of the selected stimulus of claim 1 to the training subject according to the stimulus profile of claim 1, and wherein the subject response data for each training subject comprises data of the training subject obtained according to the response profile of claim 1 and in response to the selected stimulus.

20. The sleep phenotype determination method of claim 19, wherein the training data used to train the predictive machine learning system comprises subject stimulus data and subject response data for each of the training subjects indicative of the selected stimulus comprising a treatment solution,wherein the predictive machine learning system is configured to predict an appropriateness of the treatment solution for use by the unclassified subject, andwherein predicting the sleep phenotype of the unclassified subject by the predictive machine learning system includes predicting the appropriateness of the treatment solution for use by the unclassified subject.