In-ear brain activity sensor for detecting cognitive state

Abstract

In one example, an apparatus includes a main body formed from a compliant material, an electronics housing positioned within the main body and housing electronics for measuring biometric signals, a conical tip coupled to a first end of the compliant body, and a plurality of dry electrodes in communication with the electronics. The plurality of dry electrodes includes a first subset of the plurality of dry electrodes formed on a surface of the conical tip, and a second subset of the plurality of dry electrodes formed on a surface of a second end of the main body.

Description

IN-EAR BRAIN ACTIVITY SENSOR FOR DETECTING COGNITIVE STATECROSS REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of United States Provisional Patent Serial No. 63 / 508,493, filed June 15, 2023, which is herein incorporated by reference in its entirety.FIELD OF THE INVENTION

[0002] The present disclosure relates generally to monitoring brain activity and relates more particularly to an in-ear electro-mechanical device for monitoring brain activity.BACKGROUND OF THE DISCLOSURE

[0003] Many short-term and long-term cognitive states can promote mental and physical performance while compromised states can cause declines that can be dangerous, and even fatal, in many situations. For instance, compromised cognitive states such as fatigue, lack of sleep, sleep disorders, stress, and age- related cognitive decline can reduce a person’s reaction time and attentiveness to their surroundings, as well as impair their memory, ability to estimate risk, and ability to make decisions. As a specific example, undiagnosed sleep disorders are often cited as a contributing factor in hypertension, diabetes, obesity, depression, heart attack, and strokes. It is estimated that sleep disorders are responsible for almost $100 billion in annual incremental health costs in the United States. Fatigue from impaired sleep is also often cited as a contributing factor in accidents (such as workplace accidents, driving accidents, and other types of accidents). It is estimated that over 16.5% of fatal vehicle accidents in the United States include a drowsy driver, resulting in over 8,000 deaths and $109 billion in damages per year. In addition to private and commercial accidents (e.g., trucking), the National Safety Council has also cited fatigue as the most critical hazard in construction and mining.SUMMARY

[0004] In one example, an apparatus includes a main body formed from acompliant material, an electronics housing positioned within the main body and housing electronics for measuring biometric signals, a conical tip coupled to a first end of the compliant body, and a plurality of dry electrodes in communication with the electronics. The plurality of dry electrodes includes a first subset of the plurality of dry electrodes formed on a surface of the conical tip, and a second subset of the plurality of dry electrodes formed on a surface of a second end of the main body.

[0005] In another example, a method includes printing a layer of a rigid plastic on a curved substrate using a three-dimensional printing process to form an electrode base that is sized and shaped to fit within or around an ear of a user, forming a layer of a base metal on the electrode base, forming an intermediate metal layer on the layer of the base metal, and forming a final metal layer on the intermediate metal layer.

[0006] In another example, a non-transitory computer readable medium storing instructions which, when executed by a processing system including at least one processor, cause the processing system to perform operations. The operations include receiving, from a brain activity sensor positioned to observe brain activity of a user from within an ear of the user, a plurality of measurements of the brain activity of the user, wherein the plurality of measurements of the brain activity of the user were observed while the user was performing a known control task, receiving, from a computing device separate from the in-ear brain activity sensor, a plurality of objective measurements of a cognitive state of the user, based on responses of the user to the known control task, receiving, from the computing device that is separate from the brain activity sensor, a plurality of subjective measurements of the cognitive state of the user, correlating the plurality of measurements of the brain activity of the user, the plurality of objective measurements of the cognitive state of the user, and the plurality of subjective measurements of the cognitive state of the user, detecting, based on the correlating, an event in which both: a measurement of the plurality of objective measurements of the cognitive state of the user exceeded a first threshold and a measurement of the plurality of subjective measurements of the cognitive state of the user exceeded a second threshold, identifying, based on the correlating, a measurement of the plurality of measurements of the brain activity of the userthat is correlated in time with the event, labeling the measurement of the plurality of measurements of the brain activity of the user as a triggering event, adding the triggering event to a set of training data for use in training a machine learning model, and using the set of training data to train the machine learning model to predict a cognitive state of a generic user based on in-ear measurements of brain activity of the generic user.BRIEF DESCRIPTION OF THE DRAWINGS

[0007] The teaching of the present disclosure can be readily understood by considering the following detailed description in conjunction with the accompanying drawings, in which:

[0008] FIG. 1 illustrates a high level block diagram of an example system for monitoring a user’s brain activity via an in-ear sensor of the present disclosure;

[0009] FIG. 2 illustrates side views of one example of the in-ear brain activity monitoring device 102 of FIG. 1 ;

[0010] FIG. 3 is a flow diagram illustrating one example of a method for fabricating electrodes for an in-ear brain activity sensor, according to the present disclosure;

[0011] FIG. 4 illustrates one example of a method for training a machine learning model to predict a user’s cognitive state, based at least in part on brain activity measurements observed by an in-ear brain activity sensor of the present disclosure;

[0012] FIG. 5 illustrates one example of a method for predicting when a cognitive state of a user is compromised, based at least in part on brain activity measurements observed by an in-ear brain activity sensor of the present disclosure; and

[0013] FIG. 6 depicts a high-level block diagram of a computing device or processing system specifically programmed to perform the functions described herein.

[0014] To facilitate understanding, identical reference numerals have been used, where possible, to designate identical elements that are common to the figures.DETAILED DESCRIPTION

[0015] The present disclosure provides examples of an in-ear brain activity sensor for detecting cognitive states, as well as methods for monitoring a user’s brain activity via an in-ear brain activity sensor and for manufacturing a robust in- ear brain activity sensor. As discussed above, many short-term and long-term cognitive states can promote mental and physical performance while compromised states can cause declines that can be dangerous, and even fatal, in many situations. For instance, conditions such as fatigue, lack of sleep, sleep disorders, stress, and age-related cognitive decline can reduce a person’s reaction time and attentiveness to their surroundings, as well as impair their memory, ability to estimate risk, and ability to make decisions. As a specific example, undiagnosed sleep disorders are often cited as a contributing factor in hypertension, diabetes, obesity, depression, heart attack, and strokes. It is estimated that sleep disorders are responsible for almost $100 billion in annual incremental health costs in the United States. Fatigue from impaired sleep is also often cited as a contributing factor in accidents (such as workplace accidents, driving accidents, and other types of accidents). It is estimated that over 16.5% of fatal vehicle accidents in the United States include a drowsy driver, resulting in over 8,000 deaths and $109 billion in damages per year. In addition to private and commercial accidents (e.g., trucking), the National Safety Council has also cited fatigue as the most critical hazard in construction and mining. While these deaths and damages may be prevented with common risk assessments, individuals experiencing these cognitive states may often miss or ignore the signs until it is too late to prevent an accident.

[0016] The onset of many of the above-described cognitive states can be detected through monitoring of brain activity. Existing solutions for monitoring brain activity have used camera-based eye-tracking, steering trajectory sensors, or electrophysiological recording devices. While eye tracking and steering trajectory sensor systems can be a good fit in automotive scenarios, the accuracy of eye tracking systems can be limited by sunglasses, hats, and / or other objects that may obscure a driver’s eyes, and steering sensor systems can be susceptible to false alarms on roads whose surfaces are uneven. User-centered recording modalities such as electrocardiography (ECG / EKG),electrooculography (EOG), and electroencephalography (EEG) are becoming increasingly popular because they can be taken anywhere and fit any work environment. Of these methods, EEG has been shown to achieve the highest detection accuracy when monitoring for brain state.

[0017] EEG is a safe, non-invasive method of monitoring the brain’s electrical activity from the scalp. Disorders related to sleep and epilepsy produce characteristic neurological signals that can be monitored via EEG in a clinical setting with large wet-electrode arrays. Moreover, EEG systems can be deployed in wearable devices. However, to date, few if any of the wearable devices that have been proposed for fatigue monitoring via EEG are feasible for long-term use. For instance, EEG systems that rely wet electrode arrays are limited by the abrasive skin cleaning that is required to ensure the adhesion of the electrodes to the wearer’s skin (which can also cause skin irritation and lesions over time), the large amount of training data that is needed to ensure a desired prediction accuracy, and the bulky electronics that are neither discreet nor comfortable to wear long-term. Although some wearable wet electrode EEG-based devices have been designed to fit discreetly inside a wearer’s ear canal, these devices still require the use of hydrogels which may limit the wearer’s comfort and may require custom fitting and configuration.

[0018] Examples of the present disclosure provide a discreet, in-ear brain activity sensor that is capable of estimating a wearer’s brain state based on a combination of brain activity and other biometrics. Within the context of the present disclosure, “brain state” may be understood to refer to brain activity that may be symptomatic of a cognitive state such as fatigue, lack of sleep, a sleep disorder, stress, age-related cognitive decline, and / or other types of cognitive states. In one example, the brain activity sensor includes a housing that is compact, compressible, and assembled from a minimal number of individual pieces, making the brain activity sensor unobtrusive and relatively comfortable to wear. In further examples, the housing houses a dry electrode-based EEG sensor for recording measures of a wearer’s brain activity and a small form factor wireless interface that allows the recorded measures to be transmitted to an external computing device forfurther processing. The dry electrodes are capable of establishing electrical contact with the wearer’s skin without the use ofhydrogel, thereby simplifying the electrode-to-skin application process and improving user comfort.

[0019] Further examples of the present disclosure provide a machine learningbased process for analyzing the measures of the wearer’s brain activity in combination with other biometrics of the wearer in order to estimate the wearer’s brain state and to generate an alert to notify the wearer when the estimated brain state meets a rule-based criterion for alert (e.g., when a quantifiable measure of brain state is above or below a threshold). In one example, the machine-learning based process relies on a low-power algorithm that can evaluate spontaneous EEG on-chip, such as logistic regression, a support vector machine (SVM), or a random forest algorithm. These algorithms are able to compensate for the limited spatial coverage and amplitude signals, of the in-ear brain activity sensor (as compared to scalp EEG systems) as well as the increased electrode-skin impedance (ESI), system noise, and interference susceptibility of dry electrodes. The use of a low power algorithm allows for the provision of timely notifications to a wearer of the in-ear brain activity sensor even when used in ambulatory environments with inconsistent wireless connectivity or limited access to large processing power.

[0020] Experimental results have shown that the disclosed process, when implemented in combination with the disclosed in-ear brain activity sensor, can achieve a detection accuracy for threshold levels of fatigue of approximately 93.2 - 93.3% when evaluating a new wearer (e.g., a wearer who has not been previously monitored using the in-ear brain activity sensor and for whom there is no existing baseline for brain activity). Not only is this detection accuracy on par with existing state-of-the-art wet electrode in-ear and scalp EEG systems, but it also motivates the possibility of using deploy-and-forget machine-learning models in wearable ear EEG applications.

[0021] Further examples of the present disclosure provide a process for manufacturing the in-ear brain activity sensor by plating metal electrodes on curved substrates (which may be formed from plastic or other materials). As discussed above, one example of the disclosed in-ear brain activity sensor utilizes a dry electrode-based EEG sensor. However, conventional materials for planar and three-dimensional (3D) dry electrodes, such as silver / silver chloride(Ag / AgCI) and gold (Au), suffer from drawbacks including tarnishing, depletion of conductive ions over time, and / or expensive processing produce in 3D, anatomically optimized form factors. The use of exotic materials (such as such as poly(3,4-ethylenedioxythiophene) polystyrene sulfonate and Iridium(IV) oxide), conductive composites (such as silvered-glass silicone and carbon- infused silicone), and / or high-surface area 3D electrodes can minimize some of these drawbacks, but these techniques are typically not feasible for large-scale deployment from yield and lifespan perspectives. Examples of the present disclosure provide a dry electrode fabrication technology suitable for repeated, comfortable, low impedance in-ear sensors. These and other aspects of the present disclosure are described in greater detail below in connection with FIGs. 1-6.

[0022] FIG. 1 illustrates a high level block diagram of an example system 100 for monitoring a user’s brain activity via an in-ear sensor of the present disclosure. The system 100 may include an in-ear brain activity monitoring device 102 and a computing device 104.

[0023] The in-ear brain activity monitoring device 102 may be worn in the ear of a user 106. The in-ear brain activity monitoring device 102 may be worn throughout the day to comfortably record neural signals from inside the ear canal. As discussed in further details below, the in-ear brain activity monitoring device 102 may measure neural signals / brain activity of the user 106 and transmit the measured neural signals to the computing device 104.

[0024] The computing device 104 may comprise a mobile computing device such as a laptop computer, a desktop computer, a tablet device, a mobile phone, a smart phone, a wearable smart device (e.g., a smart watch, a fitness tracker, or the like) and the like. For instance, the computing device 104 may be configured in a manner similar to the computing device 600 illustrated in FIG. 6 and discussed in greater detail below.

[0025] The in-ear brain activity monitoring device 102 may communicate wirelessly with the computing device 104. For example, the in-ear brain activity monitoring device 102 may transmit measured brain activity signals to the computing device 104 via a short range low power wireless signal, such as Bluetooth. The computing device 104 may then compare the measured brainactivity of the user 106 to trained models to detect brain states such as fatigue or drowsiness in the user 106 and provide feedback to the user 106.

[0026] The system 100 is designed to be user-generic, user-friendly, and perform as a testbed for long term in-ear EEG recording applications. The system 100 relies on a generic earpiece design and fabrication technique, a lightweight, low-power neural recording frontend, and a transparent machine learning pipeline for post-processing and classification.

[0027] A user-generic electrode scheme and physical earpiece design capable of recording across multiple individuals is critical for a ubiquitous, wearable detection platform. The in-ear brain activity monitoring device 102 is designed to ensure comfort and consistent low electrode-skin impedance (ESI). The sizes and locations of the electrodes on the in-ear brain activity monitoring device 102 are selected to minimize channel-to-channel correlation while maximizing electrode area.

[0028] FIG. 2 illustrates side views of one example of the in-ear brain activity monitoring device 102 of FIG. 1. Generally, the in-ear brain activity monitoring device 102 comprises a main body 200, a tip 202, an electronics housing 204, a sensor window 206, and a plurality of electrodes 208i-208n(hereinafter individually referred to as an “electrode 208” or collectively referred to as “electrodes 208”).

[0029] In one example, the main body 200 is formed from a soft, compliant material such as a foam or plastic and has a generally oval or fabiform shape, without cantilevers or protrusions. In one example, the main body may be fabricated using injection molding, printing, vacuum formation, or other manufacturing processes. The compliant material allows the main body 200 to be compressed for insertion into a user’s ear canal, while the fabiform shape allows the main body to confirm to and sit comfortably within the concha cavum of the user’s ear for an extended period of use. The comfort and size versatility of the main body 200 allows the in-ear brain activity monitoring device 102 to be worn across a wide range of ear canal shapes and minimizes ESI across users.

[0030] The tip 202 is attached to one end (e.g., a first end) the main body 200 and may similarly be formed of a foam or plastic. The tip 202 may have a conical or umbrella shape that allows the tip 202 to conform to the shape of the user’sear canal while at the same time exerting a slight pressure against the interior or the user’s ear canal to produce a gentle friction fit. Thus, when the in-ear brain activity monitoring device 102 is in use, the in-ear brain activity monitoring device 102 will be inserted tip-first into the user’s ear canal.

[0031] The electronics housing 204 comprises a rigid compartment that resides within or is surrounded by the main body 200. The electronics housing 204 is sized to house the electronics (such as some or all of the electronics illustrated in FIG. 6) necessary to record biometrics, electrophysiological signals, and / or brain activity measurements observed by the electrodes 208 and to transmit those brain activity measurements to an external (i.e., external to the in- ear brain activity monitoring device 102) computing device, such as the computing device 104 of FIG. 1. For instance, the electronics housing 204 may house one or more of: an EEG sensor, an electromyogram (EMG) sensor, an electrooculogram (EOG) sensor, an ECG / EKG sensor, a motion sensor, a photoplethysmography (PPG) sensor, a signal processor, a signal amplifier, a digitizier, a transponder, a Bluetooth radio or other short-range wireless radio, and / or other electronics.

[0032] The sensor window 206 may comprise a transparent opening in the main body 200 which is sized to house one or more optical sensors, such as PPG sensors (not shown) and sensors to measure other biometrics (e.g., blood oxygen levels, ambulation / motion, temperature (e.g., body temperature), respiration, and the like). In one example, the sensor window 206 is located to face toward the concha bowl of the user’s ear during use of the in-ear brain activity monitoring device 102.

[0033] In one example, the plurality of electrodes 208 comprise a plurality of dry electrodes. The dry electrodes may comprise a plurality of layers of material plated with an outer plating material such as gold, methacrylate photopolymer, nickel, or copper. In this example, the outer plating material comprises the portion of an electrode 208 that contacts a user’s skin. The electrodes 208 may be fabricated using injection molding, three-dimensional (3D) printing, milling, or other manufacturing processes. One example of a method for fabricating the electrodes 208 is discussed in greater detail with respect to FIG. 3.

[0034] The electrodes 208 may be attached to various locations on the main body 200 and / or tip 202 of the in-ear brain activity monitoring device 102. For instance, at least some of the electrodes 208 may be attached to the tip (e.g., similar to the electrodes 208i and 2082 of FIG. 2) to function as sensing electrodes. For instance, a sensing electrode may be attached to the tip 200 in a location to contact the contact the aperture of the user’s ear canal (e.g., to enter the first ten mm of the ear canal). At least one other electrode 208 may be attached to the main body 200 (e.g., similar to the electrodes 2083 and 208n) to function as a multi-contact reference or ground electrode. For instance, a reference electrode may be attached to the main body 200 in a location to contact the user’s concha cymba and concha cavity during use of the in-ear brain activity monitoring device 102. The unique shapes of the reference electrodes may also act as fiducial markers to help ensure that the in-ear brain activity monitoring device 102 is effectively locked in the same position with every use. Although FIG. 2 illustrates an example in which the in-ear brain activity monitoring device 102 includes four electrodes 208 (e.g., two sensing electrodes and two reference electrodes), it will be appreciated that the in-ear brain activity monitoring device 102 may include any number of electrodes (e.g., any number of sensing electrodes and / or any number of reference electrodes).

[0035] The electrodes 208 may have different sizes and shapes depending upon the locations on the in-ear brain activity monitoring device 102 to which the electrodes 208 attach. For instance, sensing electrodes attached to the tip 202 may have a shape that follows a curvature of the profile of the tip 200, while reference electrodes attached to the main body 200 may have semi-spherical shapes.

[0036] It will be appreciated that although FIG. 2 illustrates a single in-ear brain activity monitoring device 102, in use, a user may utilize two in-ear brain activity monitoring devices configured as illustrated in FIG. 2, i.e., one in-ear brain activity monitoring device 102 in each ear. This arrangement can provide up to eleven EEG channels with two possible reference electrodes.

[0037] FIG. 3 is a flow diagram illustrating one example of a method 300 for fabricating electrodes for an in-ear brain activity sensor, according to the present disclosure. For instance, the method 300 may be utilized to fabricate theelectrodes 208 of the in-ear brain activity monitoring device 102 illustrated in FIG. 2.

[0038] The method 300 begins in step 302. In step 304, a layer of a rigid plastic is printed on a curved substrate using a three-dimensional printing process to form an electrode base. The layer of rigid plastic can also be formed through vacuum forming, injection molding, or stamping.

[0039] In one example, the substrate may comprise a main body or a tip of an in-ear brain activity monitoring device, such as the main body 200 or tip 202 of the in-ear brain activity monitoring device 102 of FIG. 2. Thus, in one example, the substrate may comprise a soft, compliant foam, plastic, or polymer material (which may be fabricated through another 3D printing process).

[0040] The 3D printing process used to print the layer of rigid plastic in step 302 may be an additive manufacturing process or a subtractive manufacturing process. 3D printing may allow any arbitrary shape (including curves, non-linear surfaces, and the like) to be formed with the rigid plastic. For instance, in one example, the layer of rigid plastic may be printed using a stereolithography (SLA) 3D printer. SLA printers create 3D structures by precisely laser-curing photosensitive polymer resins layer-by-layer. This 3D printing method results in finer surface finishes and resolutions than fused deposition modeling printing, which comprises melted plastic filaments. Furthermore, resultant structures from SLA printers can be significantly more mechanically robust and survive high temperatures as well as greater impacts.

[0041] In one example, the layer of rigid plastic may comprise a layer of clear methacrylate photopolymer, although in other examples the layer of rigid plastic may comprise any other rigid printable plastic (e.g., thermoformed plastic or injection molded plastic), formable polymer, thermoplastic, or metal (e.g., stamped metal).

[0042] In step 306, the electrode base may be cured.

[0043] In one example, the curing process performed in step 306 may include processing the electrode base in an isopropyl alcohol (I PA) bath to rinse uncured resin. In one example, the IPA bath may have a duration of approximately twenty minutes. Following the IPA bath, an ultraviolet curing (UV) process may be usedto fully cure the surface of the electrode base. In one example, the UV curing process may have a duration of approximately one hour.

[0044] In step 308, the electrode base is sandblasted.

[0045] In one example, the sandblasting process may utilize one hundred grit white fused aluminum oxide blasting media. Sandblasting increases the effective surface area of the electrode base, which, in turn, promotes better film adhesion and lower skin-electrode impedance. In one example, the sandblasted electrode base may be sonicated in a bath of deionized water and cleaning solution for approximately ten minutes before rinsing with deionized water to remove excess media. In a further example, the surface energy of the electrode base may be further modified in a bath of one percent benzalkonium chloride surfactant solution for ten minutes. These steps ensure that the surface of the electrode base is clean for subsequent metal plating steps and promote adhesion of catalyst.

[0046] In step 310, a layer of a base metal is formed on the electrode base.

[0047] In one example, the base metal comprises copper. However, in other examples, the base metal may comprise another metal material, such as tin, nickel, or silver. For instance, in one example, a layer of copper may be formed via an electroless copper plating process. The electroless copper plating process may comprise first submerging the electrode base in a catalyst such as palladium-tin. Submersion in the catalyst may be performed for approximately ten minutes and may result in a thin layer of catalyst being formed on the electrode base. A copper sulfate plating solution may then be applied to form a thick layer of copper (i.e., the layer of the base metal) on the thin layer of catalyst. In one example, the copper sulfate plating solution may be applied for approximately four hours.

[0048] In another example, a layer of tin or nickel may be formed by blasting tin or nickel at the electrode base. In yet another example, a layer of silver (or other conductive ink) may be formed by spraying onto the electrode base.

[0049] In step 312, an intermediate metal layer is formed on the layer of the base metal.

[0050] In one example, the intermediate metal layer comprises nickel. However, in other examples, the intermediate layer may be formed of anothermetal material, such as tin or palladium. For instance, in one example, a layer of nickel may be formed on the layer of the base metal by submerging the electrode base (including the layer of the base metal) in an electroless nickel plating solution for approximately one minute. In one example, the nickel plating solution may be suspended in a water bath heated to approximately 95° C to ensure proper plating.

[0051] In step 314, a final metal layer is formed on the intermediate metal layer.

[0052] In one example, the final metal layer comprises gold. However, in other examples, the final metal layer may be formed of another metal material, such as platinum. For instance, in one example, a layer of gold may be formed on the intermediate metal layer by submerging the electrode base (including the layer of the base metal and the intermediate metal layer) in a gold plating solution for approximately fifteen minutes. In another example, a layer of platinum may be formed by electroplating.

[0053] Collectively, the layer of the base metal, the intermediate metal layer, and the final metal layer may form a functional electrode on the plastic electrode base, where the final metal layer comprises the surface of the functional electrode that contacts a user’s skin when the in-ear brain activity monitoring device 102 is in use. The example fabrication process illustrated in FIG. 3 ensures a 0.25 pm thick gold (final metal) layer and at least 0.5 pm of nickel and copper (layer of base metal and intermediate metal layer). The thicknesses of these layers ensures that the plated surface of the final metal layer effectively acts as a short circuit and that any electrode impedance would be dominated by the interface and skin itself. Furthermore, the exterior, inert gold surface ensures a long- electrode lifetime and biocompatibility for surface electrophysiological monitoring.

[0054] In step 316, conductive wires may be soldered to the final metal layer.

[0055] In one example, the conductive wires may be formed of tinned copper. The conductive wires may facilitate integration of the electrode with recording systems.

[0056] The method 300 may end in step 318.

[0057] Thus, by starting with a 3D printed mold with a plastic, portions of the plastic base can be plated with metal to form curved electrodes on the curved surfaces of the 3D printed mold. The plastic portion of the electrode (electrodebase) may be soft to provide comfort in the ear of a user, while the metal plated portions (functional electrode) may be rigid to form the electrodes of the in-ear brain activity monitoring device 102.

[0058] FIG. 4 illustrates one example of a method 400 for training a machine learning model to predict a user’s cognitive state, based at least in part on brain activity measurements observed by an in-ear brain activity sensor of the present disclosure. For instance, the method 400 may train the machine learning model to predict when the user is fatigued, feeling stressed, or falling asleep. The method 400 may be executed by a computing device that executes the machine learning model, such as the computing device 104 of FIG. 1 , or by a processing system, such as the processing system 602 of FIG. 6. For the sake of example, the method 400 is described below as being executed by a processing system including at least one processor.

[0059] The method 400 may begin in step 402. In step 404, the processing system may receive, from a brain activity sensor positioned to observe a user’s brain activity from within the user’s ear, a plurality of measurements of brain activity of the user, wherein the plurality of measurements of the user’s brain activity were observed while the user was performing a known control task.

[0060] In one example, the brain activity sensor may comprise a pair of contralaterally worn in-ear brain activity monitoring devices, similar to the in-ear brain activity monitoring device 102 of FIG. 1 . The pair of contralaterally worn in- ear brain activity monitoring devices may be communicatively coupled to each other via a wired or wireless connection. In one example, the plurality of measurements of the brain activity of the user may be transmitted from the brain activity sensor to the processing system via Bluetooth low energy (BLE). In one example, each measurement of the plurality of measurements of the brain activity of the user may comprise an EEG reading. In a further example, EEG readings may be received in a MATLAB, comma-separated values (CSV), or high-density array format. The plurality of measurements of the brain activity of the user may represent objective measures of the user’s level of cognitive state over time.

[0061] In one example, the known control task may comprise a reaction time task that is designed to help detect when the cognitive state of the user is associated with fatigue. In this example, the reaction time task may comprise agame that periodically (e.g., every x seconds) prompts the user to press a random number between zero and nine on an input device (e.g., keyboard) that is communicatively coupled to the processing system. The amount of time elapsed between the presentation of the prompt and the pressing of the corresponding number by the user may be recorded as a measurement of the reaction time of the user for that particular prompt.

[0062] In another example, the known control task may comprise sleeping. In this example, the in-ear brain activity sensor may be equipped with additional sensors such as motion sensors, heart monitors, or the like. Alternatively, these additional sensors may be integrated into monitoring devices that are separate from the in-ear brain activity sensor and send measurements to the processing system separately from the in-ear brain activity sensor. In a further example, the additional sensors may further include brain activity sensors having electrodes that are adhered to the user’s scalp.

[0063] In step 406, the processing system may receive, from a computing device separate from the in-ear brain activity sensor, a plurality of objective measurements of a cognitive state of the user, based on user responses to the known control task.

[0064] In one example, where the known control task comprises a reaction time task, the objective measurements of the cognitive state of the user may comprise observed reaction times of the user (e.g., how long it took the user to input a number and / or whether the number input by the user matched the prompted number).

[0065] In another example, where the known control task comprises sleeping, the objective measurements of the cognitive state of the user may comprise observed sleep stages experienced by the user, observed heart activity of the user, or measurements of other activity from which sleep quality can be inferred. For instance, EEG, motion, and / or heart sensors may provide measurements from which sleep stages can be inferred. In this example, sleep stages may include rapid eye movement (REM) sleep and non-REM sleep (where the latter can be further classified as N1 , N2, or N3 sleep which respectively correspond to light, moderate, and deep sleep). In one example, measurements from the EEG,motion, and / or heart sensors may be labeled to indicate which stage of sleep the measurements are associated with (e.g., awake, REM, N1 , N2, or N3).

[0066] Instep 408the processing system may receive, from the computing device that is separate from the brain activity sensor, a plurality of subjective measurements of the cognitive state of the user.

[0067] In one example where the subjective measurements are subjective measurements of fatigue, each subjective measurement may comprise a response of the user to a periodic prompt that prompts the user to enter a score on the Karolinska Sleepiness Scale (KSS) that indicates how sleepy the user currently feels. The KSS scale ranges from 0 = “extremely alert,” to 10 = “extremely sleepy, fighting to stay awake.” In one example, the interval between prompts to enter the KSS scale score may be selected to minimize interruptions while maximizing drowsiness label granularity. For instance, in one example, the user may be prompted to enter a KSS scale score every five minutes. In a further example, user responses to the prompts for KSS scale scores may be received in a CSV format. In step 410, the processing system may correlate the plurality of measurements of the brain activity of the user, the plurality of objective measurements of the cognitive state of the user, and the plurality of subjective measurements of the cognitive state of the user.

[0068] For instance, in one example, the plurality of measurements of the brain activity of the user, the plurality of objective measurements of the cognitive state of the user, and the plurality of subjective measurements of the cognitive state of the user may be plotted over time.

[0069] In step 412, the processing system may detect, based on the correlating, an event in which both: a measurement of the plurality of objective measurements of the cognitive state of the user exceeded a first threshold and a measurement of the plurality of subjective measurements of the cognitive state of the user exceeded a second threshold.

[0070] For example, where the plurality of measurements of the brain activity of the user, the plurality of objective measurements of the cognitive state of the user, and the plurality of subjective measurements of the cognitive state of the user have been plotted overtime, the processing system may detect a single time at which a measurement of the plurality of objective measurements of thecognitive state of the user exceeded the first threshold and a measurement of the plurality of subjective measurements of the cognitive state of the user exceeded the second threshold. In one example that trains the machine learning model to predict fatigue, the first threshold may comprise a reaction time that is at least double an average reaction time computed over the first five minutes’ worth of measurements of reaction times of the user to a reaction time task, and the second threshold may comprise a KSS scale score that exceeds five. However, both the first threshold and the second threshold may be tunable and may vary based on the user.

[0071] In one example, the event may be one of a plurality of events that is detected by the processing system, where each event of the plurality of events comprises an instance in which a measurement of the plurality of objective measurements of the cognitive state of the user exceeded a first threshold and a measurement of the plurality of subjective measurements of the cognitive state of the user exceeded a second threshold.

[0072] In step 414, the processing system may identify, based on the correlating, a measurement of the plurality of measurements of the brain activity of the user that is correlated in time with the event.

[0073] That is, based on the correlating that was performed in step 410, the processing system may determine a measurement of the plurality of measurements of the brain activity of the user that was observed at the same point in time that the event was observed.

[0074] In one example, a measurement of the plurality of measurements of the brain activity of the user may be identified for each event that was detected in step 412.

[0075] In step 416, the processing system may label the measurement of the plurality of measurements of the brain activity of the user as a triggering event.

[0076] In one example, a triggering event is an event that will cause an alert to be generated. For instance, any measurement of the plurality of measurements of the brain activity of the user that is associated with an event that indicates a potentially dangerous or compromised cognitive state (e.g., fatigued, sleepy, stressed, etc.) may be labeled as a triggering event in step 416. In anotherexample, specific stages of sleep (e.g., awake, REM, N1 , N2, N3, or the like) may be labeled as triggering events.

[0077] Using both objective measurements of cognitive state (e.g., reaction time, when monitoring for fatigue) and subjective measurements of cognitive state (e.g., user-assessed KSS scale score when monitoring for fatigue) enables robust label creation that is agnostic to temporary user error. In one example, the label may be put, along with other similarly generated labels, through a three sample rolling average filter and thresholded to achieve a binary label.

[0078] In one example, a correlated measurement of the plurality of measurements of the brain activity of the user may be labeled as a triggering event for each event that was detected in step 412. In a further example, any measurements of the brain activity of the user that are not correlated to any events detected in step 412 may be labeled as non-triggering events (i.e., events that do not trigger an alert).

[0079] In step 418, the processing system may add the triggering event to a set of training data for use in training a machine learning model.

[0080] That is, a set of training data may be built to assist in the training of a machine learning model. Steps 404-416 may generate additional data points that may be added to the set of training data to improve the accuracy of any predictions made by the machine learning model.

[0081] In step 420, the processing system may use the set of training data to train the machine learning model to predict a cognitive state of a generic user is , based on in-ear measurements of brain activity of the generic user.

[0082] In one example, the machine learning model may comprise a supervised machine learning model. In a further example, the machine learning model may be based on a low-complexity classifier model, such as one or more of the following machine learning algorithms: a logistic regression algorithm, a support vector machine (SVM), or a random forest algorithm. Once trained, the machine learning model should be able to take as an input an unlabeled measurement of the brain activity of a user and generate an output a prediction or classification of the current cognitive state of the user from whom the measurement of the brain activity was taken (e.g., fatigued or not fatigued, stressed or not stressed / calm, which stage of sleep, etc.). In one example, theprediction may be associated with a confidence score indicating the machine learning model’s confidence in the accuracy of the prediction.

[0083] For instance, in one example, binary (alert / fatigue, stressed / not stressed) classification may be performed using logistic regression with a stochastic average gradient descent solver and L1 regularization, which adds a penalty equal to the absolute value of the magnitude of feature coefficients. In another example, binary classification may be performed using a support vector machine (SVM) with a radial basis function (RBF) kernel that utilizes a maximum of 400 support vectors and a regularization parameter, C=1 . In another example, binary classification may be performed using a random forest classifier with 100 trees and a maximum depth of five. In one example, class probabilities returned from these models may be filtered with a 3-tap Hamming window FIR filter and thresholded to achieve final binary outputs.

[0084] In another example, a multi-state (e.g., awake, REM, N1 , N2, N3, etc.) classification may be performed using logistic regression. In another example, a multistate classification may be performed using a SVM with an RBF kernel. In another example, a multistate classification may be performed using a random forest classifier.

[0085] In further examples, the machine learning model may be trained to initiate an action in response to generating a prediction that indicates fatigued triggering event. The action may comprise, for instance, sending an alert to a computing device of the user (e.g., a smart phone, a smart watch, a tablet computer, or the like) to inform the user that they are fatigued or stressed and should take a break from any activity they are currently engaged in.

[0086] The method 400 may end in step 422.

[0087] In one example, post-processing, feature extraction, and classification of the EEG readings may be conducted to detect alert and fatigued states, stressed and not stressed / calm states, sleep stages, and the like. To take advantage of contralateral readings, each in-ear brain activity sensor may be rereferenced to the left concha cymba electrode to maximize spatial covering. Both originally referenced and re-referenced EEG channels may then be band pass filtered from 0.05-50Hz in order to target the EEG frequency bands of interest forfatigue detection, stress detection, or sleep staging, namely, delta (5, 0.05-4Hz), theta (9, 4-8Hz), alpha (a, 8-13Hz), beta (0, 13-30Hz), and gamma (y, 30-50Hz).

[0088] In one example, measurements of the user’s brain activity may be divided into a plurality of epochs of equal duration (e.g., 50 second epochs. Features may be calculated for each epoch, since maximizing the feature extraction window size will maximize classifier accuracy.

[0089] Both temporal and spectral features that appear in EEG based classification literature may be calculated. Time-domain features may include voltage standard deviation and maximum peak-to-peak voltage amplitude. For frequency characteristics, the power spectral density (PSD) may be calculated using Welch’s method. Spectral features including maximum PSD, frequency of maximum PSD, and PSD variance may be calculated for each EEG band. Absolute and relative band powers may also calculated for the following bands and ratios: 5, 9, a, 0, y, a / 0, 9 / 0, (a + 9) / 0, and (a + 9) / (a + 0). Furthermore, different filters can be used to highlight the appearance of sleep spindles, k- complexes, or potential sleep debt (PSD) slope changes associated with sleep stages. The previous epoch features may be used to capture changes between subsequent epochs related to onset of fatigue, transition between sleep stages, or other cognitive state transitions. To account for outliers and signal artifacts, features may be scaled by subtracting the median and scaling according to the interquartile range. During training of the machine learning model, analysis of variance feature selection may be used to find a predefined number of (e.g., 20) features that maximize class variation and minimize redundancy.

[0090] FIG. 5 illustrates one example of a method 500 for predicting a cognitive state of a user, based at least in part on brain activity measurements observed by an in-ear brain activity sensor of the present disclosure. The method 500 may be executed by a computing device, such as the computing device 104 of FIG. 1 , or by a processing system, such as the processing system 602 of FIG. 6. For the sake of example, the method 500 is described below as being executed by a processing system including at least one processor.

[0091] The method 500 may begin in step 502. In step 504, the processing system may obtain, from a brain activity sensor positioned within an ear of a user, a measurement of current brain activity of the user.

[0092] The brain activity sensor may be configured in a manner similar to the brain activity sensor 104 of FIGs. 1 and 2. In one example, the brain activity sensor may comprise two (wired or wirelessly connected) brain activity sensors which are both configured in a manner similar to the brain activity sensor 104 of FIGs. 1 and 2, where one brain activity sensor is worn in each ear of the user, contralaterally.

[0093] In one example, the measurement of the current brain activity of the user may comprise an EEG reading made by an EEG sensor that is integrated in the brain activity sensor. The measurement of the current brain activity of the user may be received in a MATLAB array form.

[0094] In step 506, the processing system may execute a machine learning model that predicts, based at least in part on the measurement of the current brain activity of the user, a current cognitive state of the user.

[0095] The machine learning model may be trained in accordance with the method 400, described above. Thus, the machine learning model may be trained to take as an input a measurement of a user’s brain activity and to generate a prediction, based on the input, of the user’s current cognitive state (e.g., fatigued or alert, stressed or calm, in a specific sleep stage, etc.). In one example, the prediction may be a binary classification (i.e., fatigued or not fatigued / alert, stressed or not stressed / calm). In another example, the classification may comprise a label of the KSS scale, described above. In another example, the classification may comprise a label indicating a specific stage of sleep (e.g., awake, REM, N1 , N2, N3, etc.).

[0096] In one example, the machine learning model may comprise a logistic regression model, an SVM model, or a random forest model. For instance, the machine learning model may employ logistic regression with a stochastic average gradient descent solver and L1 regularization, which adds a penalty equal to the absolute value of the magnitude of feature coefficients. In another example, the machine learning model may employ a support vector machine (SVM) with a radial basis function (RBF) kernel that utilizes a maximum of 400 support vectors and a regularization parameter, C=1. In another example, the machine learning model may employ a random forest classifier with 100 trees and a maximum depth of five. In one example, class probabilities returned from these modelsmay be filtered with a 3-tap Hamming window FIR filter and thresholded to achieve final binary outputs.

[0097] In one example, the machine learning model may generate the prediction on the basis of further inputs in addition to the measurement of the current brain activity of the user. For instance, the brain activity sensor may include sensors that detect additional biometrics of the user, including blood oxygen levels, ambulation / motion, respiration, photoplethysmography data, and the like.

[0098] In step 508, the processing system may determine whether the cognitive state of the user satisfies a triggering condition.

[0099] In one example, the triggering condition may be a threshold (e.g., maximum) fatigue level beyond which it may be considered dangerous for a user to participate in certain activities (such as driving, operating heavy machinery, or the like). The threshold fatigue level may be predefined and may comprise a score on the KSS scale. In this case, the threshold fatigue level may vary by user. For instance, if the processing system has previous knowledge of the user, the processing system may have learned a threshold score on the KSS scale beyond which the user is expected to be unable to reliably participate in certain activities.

[0100] In one example, where the machine learning model performs a binary classification of the brain activity measurement (e.g., fatigued or not fatigued / alert), the threshold fatigue level may simply be “fatigued” (i.e., any brain activity measurement that results in a classification of “fatigued” is considered to exceed the threshold fatigue level).

[0101] In another example, the triggering condition may be an amount of time that the user spends in a specific sleep stage, such as awake, REM, N1 , N2, or N3. In this case, the amount of time may be a threshold (e.g., minimum) amount of time that is recommended for the user to spend in the specific sleep stage in order to obtain a sufficient amount of rest. The threshold amount of time may vary by user. For instance, the user’s doctor or caregiver may specify the threshold amount of time.

[0102] If the processing system concludes in step 508 that the current cognitive state of the user does not satisfy the triggering condition, then the method mayreturn to step 504, and the processing system may continue as described above to obtain further measurements of the current brain activity of the user from the brain activity sensor.

[0103] If, however, the processing system concludes in step 508 that the current cognitive state of the user does satisfy the triggering condition, then the processing system may proceed to step 510. In step 510, the processing system may deliver an alert to a device of the user.

[0104] For instance, the alert may comprise a visible, audible, and / or haptic alert sent to the user’s smart phone, tablet computer, laptop computer, desktop computer, smart watch, or other computing device in the user’s vicinity. In one example, the alert may simply inform the user that he or she is fatigued or stressed. In another example, the alert may specify an amount of time that the user spent in each of a plurality of sleep stages over a period of time (e.g., awake, REM, N1 , N2, N3) In a further example, the alert may recommend an action to the user, such as to take a nap, take a break from whatever activity the user is currently engaged in, to seek assistance from another individual, consult a doctor, limit consumption of certain foods or beverages before sleeping, or the like.

[0105] Once the alert has been delivered to the device of the user, the method may return to step 504, and the processing system may continue as described above to obtain further measurements of the current brain activity of the user from the brain activity sensor. The processing system may continue to iterate through steps 504-510 until a signal instructing the processing system to stop processing the measurements of the user’s brain activity is received, until a connection with the brain activity sensor is lost or the brain activity sensor stops sending measurements of the user’s brain activity, or until some other event occurs.

[0106] Thus, the present disclosure provides the design, fabrication, assembly, and evaluation of a wireless, wearable, in-ear EEG platform for cognitive state monitoring in ambulatory environments. All aspects of this platform can be adapted to different use-cases. The 3D printed and electroless plated electrodes can be rapidly augmented for any anatomically optimized wearable, and the presented transparent classifiers demonstrate the potential of dry-electrode based brain-state classification. In contrast to other state-of-the-art in-ear recording platforms, the techniques and system presented can be used for alarge-scale deployment of user-generic, wireless ear EEG brain computer interfaces that use multiple machine learning algorithms.

[0107] The results are promising for the development of the next generation of standalone wearables that can monitor brain and muscle activity in addition to heart rate. Future work includes integrating these classifiers on-chip for real-time brain-state classification and miniaturizing all the hardware into a pair of earbuds. With further integration, a pair of ear EEG buds would significantly enable long term, daily recording EEG without interrupting a user’s day or stigma. These measurements would enable an entirely new era of research for tracking longterm cognitive changes from disorders such as depression, Alzheimer’s, narcolepsy, or stress.

[0108] FIG. 6 depicts a high-level block diagram of a computing device or processing system specifically programmed to perform the functions described herein. As depicted in FIG. 6, the processing system 600 comprises one or more hardware processor elements 602 (e.g., a central processing unit (CPU), a microprocessor, or a multi-core processor), a memory 604 (e.g., random access memory (RAM) and / or read only memory (ROM)), a module 605 for predicting when a user is fatigued, based at least in part on brain activity measurements observed by an in-ear brain activity sensor, and various input / output devices 606 (e.g., storage devices, including but not limited to, a tape drive, a floppy drive, a hard disk drive or a compact disk drive, a receiver, a transmitter, a speaker, a display, a speech synthesizer, an output port, an input port and a user input device (such as a keyboard, a keypad, a mouse, a microphone and the like)). Although only one processor element is shown, it should be noted that the computing device may employ a plurality of processor elements. Furthermore, although only one computing device is shown in the figure, if the methods 400 or 500 as discussed above are implemented in a distributed or parallel manner for a particular illustrative example, i.e., the steps of the above methods 400 or 500 or the entire methods 400 or 500 are implemented across multiple or parallel computing devices, e.g., a processing system, then the computing device of this figure is intended to represent each of those multiple computing devices.

[0109] Furthermore, one or more hardware processors can be utilized in supporting a virtualized or shared computing environment. The virtualizedcomputing environment may support one or more virtual machines representing computers, servers, or other computing devices. In such virtualized virtual machines, hardware components such as hardware processors and computer- readable storage devices may be virtualized or logically represented. The hardware processor 602 can also be configured or programmed to cause other devices to perform one or more operations as discussed above. In other words, the hardware processor 602 may serve the function of a central controller directing other devices to perform the one or more operations as discussed above.

[0110] It should be noted that the present disclosure can be implemented in software and / or in a combination of software and hardware, e.g., using application specific integrated circuits (ASIC), a programmable gate array (PGA) including a Field PGA, or a state machine deployed on a hardware device, a computing device or any other hardware equivalents, e.g., computer readable instructions pertaining to the method discussed above can be used to configure a hardware processor to perform the steps, functions and / or operations of the above disclosed methods 400 or 500. In one example, instructions and data for the present module or process 605 for predicting when a user is fatigued, based at least in part on brain activity measurements observed by an in-ear brain activity sensor (e.g., a software program comprising computer-executable instructions) can be loaded into memory 604 and executed by hardware processor element 602 to implement the steps, functions, or operations as discussed above in connection with the illustrative methods 400 or 500. Furthermore, when a hardware processor executes instructions to perform “operations,” this could include the hardware processor performing the operations directly and / or facilitating, directing, or cooperating with another hardware device or component (e.g., a co-processor and the like) to perform the operations.[oom] The processor executing the computer readable or software instructions relating to the above described method can be perceived as a programmed processor or a specialized processor. As such, the present module 605 for predicting when a user is fatigued, based at least in part on brain activity measurements observed by an in-ear brain activity sensor (including associated data structures) of the present disclosure can be stored on a tangible or physical(broadly non-transitory) computer-readable storage device or medium, e.g., volatile memory, non-volatile memory, ROM memory, RAM memory, magnetic or optical drive, device or diskette, and the like. Furthermore, a “tangible” computer- readable storage device or medium comprises a physical device, a hardware device, or a device that is discernible by the touch. More specifically, the computer-readable storage device may comprise any physical devices that provide the ability to store information such as data and / or instructions to be accessed by a processor or a computing device such as a computer or an application server.

[0112] It will be appreciated that variants of the above-disclosed and other features and functions, or alternatives thereof, may be combined into many other different systems or applications. Various presently unforeseen or unanticipated alternatives, modifications, variations, or improvements therein may be subsequently made by those skilled in the art which are also intended to be encompassed by the following claims.

Claims

IN THE CLAIMSWhat is claimed is:

1. An apparatus, comprising: a main body formed from a compliant material; an electronics housing positioned within the main body and housing electronics for measuring biometric signals; a conical tip coupled to a first end of the main body; and a plurality of dry electrodes in communication with the electronics, comprising: a first subset of the plurality of dry electrodes formed on a surface of the conical tip; and a second subset of the plurality of dry electrodes formed on a surface of a second end of the main body.

2. The apparatus of claim 1 , wherein the compliant material is at least one of: a foam, a plastic, or a polymer.

3. The apparatus of claim 1 , wherein a shape of the main body is at least one of: oval orfabiform.

4. The apparatus of claim 1 , wherein the electronics include at least one of: an electroencephalogram sensor, an electromyogram sensor, an electrooculogram sensor, an electrocardiogram sensor, a motion sensor, or a photoplethysmography sensor.

5. The apparatus of claim 4, wherein the electronics further include at least one of: a signal processor, a signal amplifier, a digitizer, a transponder, a Bluetooth radio, or a short-range wireless radio.

6. The apparatus of claim 1 , further comprising: a transparent opening formed in a side of the main body.

7. The apparatus of claim 6, wherein the transparent opening is sized to house at least one of: an optical sensor, a photoplethysmography sensor, a blood oxygen sensor, an ambulation sensor, a temperature sensor, a motion sensor, or a respiration sensor.

8. The apparatus of claim 6, wherein the transparent opening is located to face toward a concha bowl of an ear of a user when the apparatus is placed in the ear of the user.

9. The apparatus of claim 1 , wherein each electrode of the plurality of dry electrodes is formed from at least one of: gold, plastic, methacrylate photopolymer, nickel, or copper.

10. The apparatus of claim 1 , wherein the first subset of the plurality of dry electrodes comprises sensing electrodes, and the second subset of the plurality of dry electrodes comprises reference or ground electrodes.11 . The apparatus of claim 1 , wherein the second subset of the plurality of dry electrodes is located so that the second subset of the plurality of dry electrodes contacts a concha cymba and a concha cavity of an ear of a user when the apparatus is placed in the ear of the user.

12. The apparatus of claim 1 , wherein the first subset of the plurality of dry electrodes has a different shape than the second subset of the plurality of dry electrodes.

13. A method, comprising: printing a layer of a rigid plastic on a curved substrate using a three- dimensional printing process to form an electrode base that is sized and shaped to fit within or around an ear of a user; forming a layer of a base metal on the electrode base; forming an intermediate metal layer on the layer of the base metal; and forming a final metal layer on the intermediate metal layer.

14. The method of claim 13, wherein the layer of rigid plastic comprises a layer of clear methacrylate photopolymer, thermoformed plastic, injection molded plastic, or stamped metal.

15. The method of claim 14, wherein the base metal comprises at least one of: copper, tin, nickel, or silver.

16. The method of claim 15, wherein the intermediate metal layer comprises at least one of: nickel, tin or palladium.

17. The method of claim 15, wherein the final metal layer comprises at least one of: gold or platinum.

18. A non-transitory computer readable medium storing instructions which, when executed by a processing system including at least one processor, cause the processing system to perform operations comprising: receiving, from a brain activity sensor positioned to observe brain activity of a user from within an ear of the user, a plurality of measurements of the brain activity of the user, wherein the plurality of measurements of the brain activity of the user was observed while the user was performing a known control task; receiving, from a computing device separate from the brain activity sensor, a plurality of objective measurements of a cognitive state of the user, based on responses of the user to the known control task; receiving, from the computing device that is separate from the brain activity sensor, a plurality of subjective measurements of the cognitive state of the user; and generating a plurality of labels for the plurality of measurements of the brain activity of the user, based on the plurality of objective measurements of the cognitive state of the user and the plurality of subjective measurements of the cognitive state of the user, wherein the plurality of labels is applied to the plurality of measurements of the brain activity of the user for use in training a machine learning model.

19. The non-transitory computer readable medium of claim 18, wherein the generating comprises: correlating the plurality of measurements of the brain activity of the user, the plurality of objective measurements of the cognitive state of the user, and the plurality of subjective measurements of the cognitive state of the user; detecting, based on the correlating, an event in which both: a measurement of the plurality of objective measurements of the cognitive state of the user exceeded a first threshold and a measurement of the plurality of subjective measurements of the cognitive state of the user exceeded a second threshold; identifying, based on the correlating, a measurement of the plurality of measurements of the brain activity of the user that is correlated in time with the event; labeling the measurement of the plurality of measurements of the brain activity of the user as a triggering event; and adding the triggering event to the set of training data for use in training the machine learning model.

20. The non-transitory computer readable medium 19, further comprising: using the set of training data to train the machine learning model to predict a cognitive state of a generic user based on in-ear measurements of brain activity of the generic user.

Images