Methods and apparatus for sleep monitoring

Abstract

Apparatus and methods detect sleep staging events. The apparatus (100) may be configured to obtain a facial biopotential signal measured between two electrodes connected to a user's face, which may be configured to form a transverse-ocular measurement vector. The biopotential signal may be measured by a biopotential measurement device comprising the two electrodes. The apparatus (100, 200) may be configured to derive, from the facial biopotential signal, a plurality of biosignals suggestive of sleep staging events. The apparatus may be configured to classify individual segments of the plurality of biosignals as belonging to one of a plurality of sleep stating events. The classification may involve using a trained machine learning model, such as a recurrent neural network, to predict sleep staging events for different segments or epochs of the plurality of biosignals.

Description

METHODS AND APPARATUS FOR SLEEP MONITORING1.1. CROSS REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of United States Provisional Patent Application No. 63 / 662,455, filed 21 June 2024, the entire content of which is incorporated herein by reference.2. BACKGROUND OF THE TECHNOLOGY2.1. FIELD OF THE TECHNOLOGY

[0002] The present technology generally relates to one or more of the screening, detection, diagnosis, monitoring, treatment, prevention and amelioration of disorders, such as sleep-related disorders. The present technology also relates to medical devices or apparatus, and their use. More particularly, some embodiments of the present technology relate to detecting biosignals from the biopotential of the head or face. In some implementations, the present technology involves the screening, detection, diagnosis, monitoring, treatment, prevention, and / or amelioration of sleep disorder events.2.2. DESCRIPTION OF RELATED ART

[0003] Screening and diagnosis generally describe the identification of a disorder from its signs and symptoms. Screening typically gives a true / false result such as indicating whether or not a patient’s disorder is severe enough to warrant further investigation, while diagnosis may result in clinically actionable information. Screening and diagnosis tend to be one-off processes, whereas monitoring the progress of a disorder can continue indefinitely. Some screening / diagnosis systems are suitable only for screening / diagnosis, whereas some may also be used for monitoring.

[0004] Polysomnography (PSG) is typically used for the evaluation of sleep architecture and common sleep disorders. See L.A. Smolley, Adult poly somnogi iphy, in ENCYCLOPEDIA OF SLEEP AND CIRCADIAN RHYTHMS, Oxford: Academic Press, pp. 474-477 (2nd Ed., 2023). One advantage of PSG is its inclusion of a large number of sensing modalities, which allows for a comprehensive assessment of sleep conditions. These sensing modalities are typically used to measure and record various biosignals, such as brain activity by electroencephalography (EEG), eye movements by electrooculography (EOG), muscle activity / activation by electromyography (EMG), and heart rhythm by electrocardiography (ECG), which can accurately determine a patient’s sleep stages, thereby classifying sleep in one of five stages: Wakefulness, Rapid-Eye-Movement (REM) sleep, and three stages of non-REM sleep (Nl, N2, N3). Stages N1 and N2 are referred to as light sleep whereas stage N3 is denominated as deep sleep. See Berry et al., The AASM Manual for the Scoring of Sleep and Associated Events: Rules, Terminology and Technical Specifications, A ERICAN ACADEMY OF SLEEP MEDICINE, V. 2.4 (2017). The assessment of these sleep characteristics and other derivations, such as sleep onset and sleep latencies, is a crucial step in sleep disorderdiagnostics. For example, the diagnosis of Obstructive Sleep Apnea (OSA) relies on the accurate estimation of the Total Sleep Time (TST) to calculate the Apnea-Hypopnea-Index (AHI). Furthermore, interest in comprehensive OSA phenotyping for the development of personalized OSA management is growing. See Zinchuk et al., Phenotypes in obstructive sleep apnea: A definition, examples and evolution of approaches, Sleep Med Rev., vol. 35, pp. 113-123 (2017). For example, rapid eye movement (REM)-predominant OSA is one such phenotype that can inform healthcare providers on the choice of an optimal treatment strategy. See Massie et al., Phenotyping REM OSA by means of peripheral arterial tone-based home sleep apnea testing and polysomnography: A critical assessment of the sensitivity and specificity of both methods, J. Sleep Res., vol. 31, no. 2, p. el3481 (3 Apr. 2022).

[0005] Common PSG setups are deployed in hospitals, clinics, sleep labs, and / or other healthcare facilities, involve the placement of 15 to 20 contact sensors on a person in order to measure the various biosignals, and usually involve expert clinical staff to properly operate and apply the PSG system. However, analysis from PSG can be a costly and time-consuming procedure, which requires the assistance of highly trained personnel for sensor application, patient monitoring and data scoring. The high density of attached sensors and electrodes combined with an unfamiliar sleep environment may cause patient discomfort and may alter the patient’s natural sleep patterns. Factors such as night-to-night variability and the first-night effect can further decrease the reliability of the PSG analysis. See Newell et al., Is a one-night stay in the lab really enough to conclude? First-night effect and night-to-night variability in polysomnographic recordings among different clinical population samples, Psychiatry Research, vol. 200, no. 2-3, pp. 795-801 (30 Dec. 2012). For these reasons, ambulatory versions of the PSG that support in-bedroom, multi-night assessment of sleep have gained popularity.

[0006] Notably, reduced channel home sleep apnea testing (HSAT) devices, such as those based on peripheral arterial tone, are well-positioned to address this need. See Van Pee et al., A multicentric validation study of a novel home sleep apnea test based on peripheral arterial tonometry, Sleep, vol. 45, no. 5 (1 May 2022). Peripheral arterial tone-based HSAT devices typically include a finger-based photoplethysmography (PPG) and accelerometery sensing module from which signal modalities such as SpCh, Pulse Rate (PR), activity and peripheral arterial tone can be derived and analyzed to estimate the AHI. A significant limitation of these devices is their lack of EEG, EOG, and EMG sensing capabilities, which prohibits the conventional and accurate assessment of sleep stages.3. BRIEF SUMMARY OF THE TECHNOLOGY

[0007] The present technology is directed towards providing medical devices used in the screening, monitoring, diagnosis, amelioration, treatment, or prevention of sleep disorders havingone or more of improved comfort, cost, efficacy, ease of use and manufacturability.

[0008] Some implementations of the present technology may include a single-channel biopotential measurement device, which can be used for detecting sleep staging events and / or sleep disorder events and which can be used for home sleep apnea testing setup (HSAT).

[0009] Some versions of the present technology may include a processor-implemented method for detecting sleep staging events. The method may include obtaining a facial biopotential signal measured between two electrodes connected to a user’s face. The method may include deriving, from the facial biopotential signal, a plurality of biosignals suggestive of sleep staging events. The method may include classifying individual segments of the plurality of biosignals as belonging to one of a plurality of sleep staging events.

[0010] In some implementations, the two electrodes may form a transverse-ocular measurement vector when the facial biopotential signal is measured between the two electrodes. The plurality of biosignals may be derived by filtering. The filtering may produce electroencephalography (EEG)data, electrooculography (EOG) data and / or electromyography (EMG) data. The filtering may include applying frequency-based filtering to the facial biopotential signal. The applying frequency -based filtering to the facial biopotential signal may include any one or more of: filtering a frequency range between 0.5 Hertz (Hz) and thirty-five Hz from the facial biopotential signal to produce the EEG data; filtering a frequency range between 0.1 Hz and thirty -five Hz from the facial biopotential signal to produce the EOG data; and filtering a frequency range between seventy Hz and one hundred ten Hz from the facial biopotential signal to produce the EMG data. The facial biopotential signal may be measured based on a voltage difference between the two electrodes. The facial biopotential signal may be measured based on a current between the two electrodes.

[0011] In some implementations, the obtaining may include obtaining the facial biopotential signal from a biopotential measurement device may include the two electrodes. The biopotential measurement device may be a single-channel biopotential measurement device that may include two electrodes. The biopotential measurement device may be a single-channel biopotential measurement device that may include only two electrodes. The two electrodes may be skin contact electrodes including respective adhesive patches. The two electrodes may include a sub-ocular facial electrode and a supra-ocular facial electrode. The sub-ocular facial electrode may be a sagittal-right facial electrode or a sagittal-left facial electrode. The supra-ocular facial electrode may be a sagittal-neutral facial electrode. The sub-ocular facial electrode may be a measurement electrode and the supra-ocular facial electrode may be a reference electrode. The reference electrode may be configured to be attached to a center of the user’s forehead, and the measurement electrode may be configured to be attached sub-ocular to an eye of the user. The supra-ocularfacial electrode may be a measurement electrode and the sub-ocular facial electrode may be a reference electrode. The measurement electrode may be configured to be attached to a center of the user’s forehead, and the reference electrode may be configured to be attached below the user’s eye. The reference electrode may be a ground electrode. The measurement electrode may be connected to the reference electrode via an electric wire.

[0012] In some implementations, the biopotential measurement device may further include a base, and the base may include a controller with at least one processor configured to measure the biopotential signal. The base may be directly connected to the reference electrode. The base may further include a communication interface connected to the at least one processor. The method may be performed by the controller in the base. The method may further include: causing transmission, using the communication interface, of any one or more of: (a) the classified individual segments, (b) the measured facial biopotential signal, and (c) the derived plurality of biosignals, to an external computing system over a wired or wireless connection between the communication interface and the external computing system. The method may be performed by a computing system external to the biopotential measurement device, and the obtaining may include receiving the biopotential signal from the biopotential measurement device over a wired or wireless connection with the biopotential measurement device, wherein the biopotential signal may be transmitted by the biopotential measurement device using the communication interface. Any one or more of the steps of the method may be performed by a computing system external to the biopotential measurement device and / or by the biopotential measurement device. Thus, some of the steps may be performed by the computing system external to the biopotential measurement device and some of the steps may be performed by the biopotential measurement device.

[0013] In some implementations, the method may further include segmenting each of the plurality of biosignals into the individual segments, each individual segment may include one of a plurality of epochs. The classifying may include feeding the plurality of epochs to a sleep staging model (SSM) to predict a classification label for each epoch of the plurality of epochs. Each classification label may correspond to one or more sleep staging events. The SSM may be a trained machine learning (ML) classifier model. The ML classifier model may be a recurrent neural network (RNN). The SSM may include a filterbank configured to extract frequency sub-bands. The SSM may include an epoch-level bidirectional attention-based SSM configured to perform sequential feature extraction based on the extracted frequency sub-bands. The SSM may include a sequencelevel bidirectional SSM configured to perform modelling of sequential features extracted by the epoch-level bidirectional attention-based SSM. The SSM may include a softmax layer configured to predict probabilities for each classification label.

[0014] In some implementations, the method may further include transforming each epoch of the plurality of epochs into a time-frequency image via short-time Fourier transform. The method may further include outputting the predicted classification labels for each epoch.

[0015] Some versions of the present technology may include a controller. The controller may include at least one processor and at least one memory including processor control instructions. The at least one memory and processor control instructions may configured to, with the at least one processor, cause the controller to perform any one or more of the aspects and / or steps described herein of the method(s) for detecting sleep staging events.

[0016] Some versions of the present technology may include a processor-readable storage medium. The medium may include processor-executable instructions, wherein execution of the processor-executable instructions by one or more processors of a computing system may cause the computing system to perform any one or more of the aspects described herein of the method(s) for detecting sleep staging events.

[0017] Some versions of the present technology may include a biopotential measurement device. The biopotential measurement device may include a reference electrode. The biopotential measurement device may include a measurement electrode. The biopotential measurement device may include a positioning structure connecting the reference electrode and the measurement electrode. The positioning structure may be configured to, when in use on a patient's face, enable positioning of the reference electrode with respect to the measurement electrode so that the electrodes form a transverse-ocular measurement vector. The biopotential measurement device may include a base. The base may include at least one processor. The at least one processor may be configured to measure a biopotential between the measurement electrode and the reference electrode. The at least one processor may be configured to generate a biopotential signal based on the measured biopotential.

[0018] In some implementations, the positioning structure may include an electric wire that electrically couples one of the reference electrode and the measurement electrode to the base. The positioning structure may be configured to connect the reference electrode and the measurement electrode so that, when in use on a patient's face, one of the reference electrode and the measurement electrode may be positioned in a central forehead location of the patient's head, and the other one of the reference electrode and the measurement electrode may be positioned on a cheek location of the patient's face. The facial biopotential signal may be measured based on a voltage difference between the reference electrode and the measurement electrode. The facial biopotential signal may be measured based on a current between the reference electrode and the measurement electrode.

[0019] In some implementations, the reference electrode and the measurement electrode may be skin contact electrodes including respective adhesive patches. The two electrodes include a subocular facial electrode and a supra-ocular facial electrode. The sub-ocular facial electrode may be a sagittal -right facial electrode or a sagittal-left facial electrode. The supra-ocular facial electrode may be a sagittal-neutral facial electrode. The sub-ocular facial electrode may be a measurement electrode and the supra-ocular facial electrode may be a reference electrode. The reference electrode may be configured to be attached to a center of the user’s forehead, and the measurement electrode may be configured to be attached sub-ocular to an eye of the user. The supra-ocular facial electrode may be a measurement electrode and the sub-ocular facial electrode may be a reference electrode. The measurement electrode may be configured to be attached to a center of the user’s forehead, and the reference electrode may be configured to be attached below the user’s eye. The reference electrode may be a ground electrode.

[0020] In some implementations, the base may include the reference electrode. The base may further include a communication interface connected to the at least one processor. The at least one processor may be configured to cause the communication interface to transmit the biopotential signal to an external computing system over a wired or wireless connection between the communication interface and the external computing system. The at least one processor may be configured to derive, from the biopotential signal, a plurality of biosignals indicative of sleep staging events. The at least one processor may be configured to classify individual segments of the plurality of biosignals as belonging to one of a plurality of sleep staging events. The plurality of biosignals may include electroencephalography (EEG) signals, electrooculography (EOG) signals, and electromyography (EMG) signals. The at least one processor may be configured to: cause the communication interface to transmit any one or more of: (a) classified individual segments of a plurality of biosignals derived from the measured facial biopotential signal, (b) the measured facial biopotential signal, and (c) the derived plurality of biosignals, to an external computing system over a wired or wireless connection between the communication interface and the external computing system. The at least one processor may be configured to segment each of the plurality of biosignals into a plurality of epochs. The at least one processor may be configured to operate a sleep staging model (SSM) to predict a classification label for each epoch of the plurality of epochs, wherein each classification label corresponds to one or more sleep staging events. The SSM may be a trained recurrent neural network (RNN) classifier model. The SSM classifier model may include a filterbank configured to extract frequency sub-bands. The SSM classifier model may include an epoch-level bidirectional attention-based SSM configured to perform sequential feature extraction based on the extracted frequency sub-bands. The SSM classifier model may include a sequence-level bidirectional SSM configured to perform modellingof sequential features extracted by the epoch-level bidirectional attention-based SSM. The SSM classifier model comprisesmay include a softmax layer configured to predict probabilities for each classification label.

[0021] In some implementations, the filterbank, such as one implemented with software and / or by hardware, filterbank, may be configured to apply frequency-based filtering to the facial biopotential signal. The frequency sub-bands of the fiterbank may include any one or more of (a) a frequency range between 0.5 Hertz (Hz) and thirty-five Hz from the facial biopotential signal to produce the EEG data; (b) a frequency range between 0.1 Hz and thirty -five Hz from the facial biopotential signal to produce the EOG data; and (c) a frequency range between seventy Hz and one hundred ten Hz from the facial biopotential signal to produce the EMG data.

[0022] In some implementations, the device, such as one or more processors of the device, and / or a computing system external to the device, may include a filterbank that may be configured to apply frequency-based filtering to the facial biopotential signal to derive one or more biosignals. The one or more frequency bands of the filterbank may include any one or more of: a frequency range between 0.5 Hertz (Hz) and thirty-five Hz from the facial biopotential signal to produce the EEG data; a frequency range between 0.1 Hz and thirty -five Hz from the facial biopotential signal to produce the EOG data; and a frequency range between seventy Hz and one hundred ten Hz from the facial biopotential signal to produce the EMG data. One or more processors of the device may be configured to transform each epoch of the plurality of epochs into a time-frequency image via short-time Fourier transform.

[0023] In some implementations, the device may further include a microphone. At least one processor of the device may be configured to detect snoring sounds. The device may further include an accelerometer. At least one processorof the device may be configured to detect body position and / or user motion.

[0024] Some versions of the present technology may include a computing system. The computing system may include a communication interface configured to receive, over a wired or wireless connection, a facial biopotential signal from a biopotential measurement device may include two electrodes, wherein the facial biopotential signal may be measured between the two electrodes connected to a user’s face and configured in a transverse-ocular measurement vector. The computing system may include one or more memories. The computing system may include one or more processors connected to the one or more memories. The one or more processors may be configured to obtain the facial biopotential signal. The one or more processors may be configured to derive, from the facial biopotential signal, a plurality of biosignals suggestive of sleep staging events. The one or more processors may be configured to classify individual segments of the plurality of biosignals as belonging to one of a plurality of sleep stating events.

[0025] The one or more processors may be part of a computing system (e.g., one or more servers) external to the biopotential measurement device and / or may be part of the biopotential measurement device, such as any of the biopotential measurement devices described herein. Thus, some or all of the steps of the methods described herein may be performed by the computing system external to the biopotential measurement device and some or all of the steps may be performed by the biopotential measurement device.

[0026] In some implementations, the facial biopotential signal may be measured based on a voltage difference between the two electrodes or based on a current between the two electrodes. The two electrodes may include a reference electrode and a measurement electrode. The measurement electrode may be a sub-ocular facial electrode and one of a sagittal-right facial electrode or a sagittal-left facial electrode. The reference electrode may be a supra-ocular facial electrode and a sagittal -neutral facial electrode. The reference electrode may be configured to be attached to a center of the user’s forehead, and the measurement electrode may be configured to be attached to a portion of the user’s face underneath the user’s eye. The reference electrode may be a sub-ocular facial electrode and one of a sagittal-right facial electrode or a sagittal-left facial electrode. The measurement electrode may be a supra-ocular facial electrode and a sagittal -neutral facial electrode. The measurement electrode may be configured to be attached to a center of the user’s forehead, and the reference electrode may be configured to be attached to a portion of the user’s face underneath the user’s eye. The reference electrode may be a ground electrode. The measurement electrode may be connected to the reference electrode via an electric wire.

[0027] In some implementations, the plurality of biosignals may include electroencephalography (EEG) signals, electrooculography (EOG) signals, and electromyography (EMG) signals, or at least two thereof. The one or more processors may be configured to operate software to segment each of the plurality of biosignals into a plurality of epochs. The classifying may include operating a sleep staging model (SSM) to predict a classification label for each epoch of the plurality of epochs, wherein each classification label corresponds to one or more sleep staging events. The SSM may be a trained recurrent neural network (RNN) classifier model. The SSM or RNN may include a filterbank configured to extract frequency sub-bands. The SSM or RNN may include an epoch-level bidirectional attention-based RNN configured to perform sequential feature extraction based on the extracted frequency sub-bands. The SSM or RNN may include a sequence-level bidirectional RNN configured to perform modelling of sequential features extracted by the epochlevel bidirectional attention-based RNN. The SSM or RNN may include a softmax layer configured to predict probabilities for each classification label. The one or more processors may be configured to operate with software to transform each epoch of the plurality of epochs into a time-frequency image via short-time Fourier transform.

[0028] In some implementations, the one or more processors may be configured to operate with software to output the predicted classification labels for each epoch. The computing system may be, or include, any one or more of a server(s) and a mobile device(s). The one or more processors may include a filterbank configured to apply frequency-based filtering to the facial biopotential signal to derive one or more biosignals. The one or more frequency bands of the filterbank may include any one or more of a frequency range between 0.5 Hertz (Hz) and thirty-five Hz from the facial biopotential signal to produce the EEG data; a frequency range between 0.1 Hz and thirty- five Hz from the facial biopotential signal to produce the EOG data; and a frequency range between seventy Hz and one hundred ten Hz from the facial biopotential signal to produce the EMG data.

[0029] Some implementations of the present technology may include a processor-readable storage medium that may include processor-executable instructions for performing a method including any one or more aspects of the methodologies as describe herein when executed by one or more processors.

[0030] Some implementations of the present technology may include a processor-readable medium, having stored thereon processor-executable instructions which, when executed by one or more processors, cause the one or more processors to perform some or all aspects of the previously described method(s). Some implementations of the present technology may include a server with access to any of the processor-readable medium described herein. The server may be configured to receive requests for downloading the processor-executable instructions of the processor- readable medium to a processing device over a network.

[0031] Some implementations of the present technology may include a processing device. The processing device may include: one or more processors; and (a) any processor-readable medium described herein, or (b) wherein the processing device is configured to access the processorexecutable instructions with any server described herein. The processing device may be a respiratory therapy apparatus. The processing device may be configured to generate a pressure therapy or a flow therapy.

[0032] Some implementations of the present technology may include a method of a server having access to any processor-readable medium described herein. The method of the server may include receiving, at the server, a request for downloading the processor-executable instructions of the processor-readable medium to an electronic processing device over a network. The method of the server may include transmitting the processor-executable instructions to the electronic processing device in response to the request.

[0033] Some implementations of the present technology may include a method of one or more processors for detecting sleep staging events. The method may include accessing, with the one or more processors, any processor-readable medium described herein. The method may includeexecuting, in the one or more processors, the processor-executable instructions of the processor- readable medium.

[0034] Portions of the aspects may form sub-aspects of the present technology. Additionally, various ones of the sub-aspects and / or aspects may be combined in various manners and also constitute additional aspects or sub-aspects of the present technology. Other features of the technology will be apparent from consideration of the information contained in the following detailed description, abstract, drawings and claims.4. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The present technology is illustrated by way of example, and not by way of limitation, in the figures of the accompanying drawings, in which like reference numerals refer to similar elements including:

[0036] Fig. 1 shows an example environment illustrating an implementation of a data gathering system according to, or associated with, the present technology.

[0037] Fig. 2 shows an example computing system or processing device suitable for performing various methodologies according to example implementations of the present technology.

[0038] Fig. 3 shows example head and / or facial regions for sensor-based biopotential measurement of the present technology.

[0039] Figs. 4 and 5 show example biopotential measurement device in accordance with the present technology when worn by a patient.

[0040] Fig. 6 shows top and bottom views of the example biopotential measurement device of Figs. 4 and 5.

[0041] Fig. 7 shows side views of the example biopotential measurement device of Figs. 4 and 5.

[0042] Fig. 8 shows front and back views of a base of the example biopotential measurement device of Figs. 4 and 5.

[0043] Fig. 9 shows front, top, and bottom views of an adhesive electrode patch used in the example biopotential measurement device of Figs. 4 and 5.

[0044] Fig. 10 shows an example artificial neural network (NN) that can be used to practice aspects discussed in the present technology.

[0045] Fig. 11 shows an example process of a methodology for detecting sleep staging events of the present technology.5. DETAILED DESCRIPTION OF EXAMPLES OF THE TECHNOLOGY

[0046] Before the present technology is described in further detail, it is to be understood that the technology is not limited to the particular examples described herein, which may vary. It is also to be understood that the terminology used in this disclosure is for the purpose of describing only the particular examples discussed herein, and is not intended to be limiting.

[0047] The following description is provided in relation to various examples which may share one or more common characteristics and / or features. It is to be understood that one or more features of any one example may be combinable with one or more features of another example or other examples. In addition, any single feature or combination of features in any of the examples may constitute a further example.5.1. Overview

[0048] One aspect of the present technology relates to a methodology and apparatus that uses physiological data measured, collected, recorded by a data gathering system, to monitor, evaluate, and / or diagnose sleep staging performance of individual users. Aspects of such technology may be considered in relation to Figs. 1 and 2.5.1.1. Data Gathering System

[0049] Fig. 1 illustrates an example network environment 100 in which aspects of the present disclosure may be practiced. The environment 100 includes a data gathering system 101 that is configured to receive, detect, transform, and / or transfer data 108 to a target system 110. The data gathering system may be implemented in an HSAT. Such data 108 may include time varying numeric signals, such as physiological data and / or other data.

[0050] The data gathering system 101 may be any system that is configured to receive, detect, analyze, and / or transfer data 108, such as physiological data, associated with the user 102. Such physiological data may include, but not limited to, heart rate, step count, blood glucose, blood pressure, respiration rate, body temperature, blood volume, sound pressure, lactic acid, photoplethysmography (PPG), electroencephalogram (EEG), electrocardiogram (ECGZEKG), electromyogram (EMG), electrooculogram (EOG), electroretinogram (ERG), electrogastrogram (EGG), blood oxygen saturation (SpO?), skin conductance such as galvanic skin response (GSR) and / or electrodermal activity (EDA), PSG data, peripheral arterial tone, and biopotential signals, among other possibilities.

[0051] In some implementations, the data gathering system 101 is or includes one or more sensors or devices to monitor and / or detect physical stimuli, environmental changes, and / or other phenomena related to a user 102, such as physiological states of the user 102, and converts them into signals or data, such as data 108. For example, the data gathering system 101 may include one or more of the following sensors / devices: a smart watch 101a, a temperature sensor 101b, an electrocardiogram (ECG / EKG) device 101c, a respiratory pressure medical device 10 Id (e.g., also referred to as a respiratory therapy (RT) device, a respiratory pressure therapy (RPT) device, and / or a high flow therapy device (HFT)), a finger sensor device lOlf, a biopotential measurement device lOle, a health tracker, a blood monitor (e.g., a glucose meter, lactic acid meter / analyzer, and / or the like), flow sensors and / or flow rate sensors, pressure sensors, motion sensors, imagecapture devices (e.g., cameras), a smartphone, sonar sensors and / or microphones, among other possibilities. Examples of the finger sensor device lOlf, such as a pulse oximeter configured to derive a peripheral arterial tone signal, are disclosed in EP Pat. No. 3,593,707 and / or U.S. Pat. Pub. 2020 / 0015737, the entire disclosure of each of which is incorporated herein by reference. Examples of sonar-based sensors can include a microphone and speaker implemented processing device such as any of the processing devices described in U.S. Pat. Pub. 2021 / 0275056 and U.S. Pat. Pub. 2022 / 0007965, the entire disclosures of each of which are incorporated herein by reference. Additionally or alternatively, the data gathering devices 101 can include radiofrequency (RF) sensing devices described in U.S. Pat. Pub. 2014 / 0024917, Int’l App. No. PCT / EP2017 / 070773, and U.S. Pat. Pub. 2018 / 0239014, the entire disclosures of each of which are incorporated herein by reference. For purposes of the present disclosure, the data gathering system 101 may refer to an individual sensor / device, or a collection of multiple sensors / devices.

[0052] The data gathering system 101 may send data 108 to a target system 110 via a wired or wireless connection. In one example, the data gathering system 101 may include a built-in wireless transceiver, which may be regarded as a data transfer device, configured to transmit the data 108 wirelessly. Additionally or alternatively, the data gathering system 101 may include a built-in network interface controller (NIC), which may be regarded as a data transfer device, configured to transmit the data 108 over wired medium. The target system 110 may include one or more of the following a remote server 112 and / or a wireless device 114.5.1.2. Remote Server

[0053] The remote server 112 may be a remotely located computing system of one or more servers that receives data 108 provided by one or more data gathering systems 101. The remote server 112 may be implemented to monitor conditions or treatment progress of one or more users 102 based on data 108 provided by one or more data gathering systems 101. The remote server 112 may be a cloud-based server system that provides cloud computing services. Additionally, the remote server 112 may include various hardware components, such as any of those discussed infra with respect to Fig. 2.

[0054] The remote server 112 may be accessible to a clinician(s) and patients, such as user 102. Each user 102 may have a user account at the remote server 112. Each user account may store historical data 108 obtained from the data gathering system 101, so that the remote server 112 can track treatment progress of each individual user 102. Furthermore, as discussed in more detail infra, the remote server 112 may operate one or more machine learning (ML) models to predict various sleep stages of the user 102 and / or output other sleep monitoring related data associated with the user 102, which may be based on the collected data 108, such as biopotential data collected from the biopotential measurement device lOle.5.1.3. Wireless Device

[0055] The wireless device 114 may be a computing system accessible by the user 102, a user’s physician, and / or other healthcare or equipment provider. The wireless device 114 may collect, manage and / or monitor data 108 provided by the user’s data gathering system 101. Examples of the wireless device 114 may include mobile phone and / or smartphone, tablet, smart appliance, smart TV, netbook, laptop computer, desktop computer, and wearable computing device such as a smartwatch, among other possibilities. The wireless device 114 may include all of the components normally used in connection with a computing device such as a user interface subsystem and / or other hardware subsystems / components, such as any of those discussed infra with respect to Fig. 2.

[0056] In one example, when the wireless device 114 is connected with the data gathering system 101, the wireless device 114 may have two-way communication with the data gathering system 101. The wireless device 114 may transmit any user input, including any device setting of the data gathering system 101, to the data gathering system 101. The data gathering system 101 may receive the user’s input via the wireless device 114, and adjust any device setting according to the user input. The data gathering system 101 may send data 108 to the wireless device 114. The wireless device 114 may display, in its graphical user interface, data 108 received from the data gathering system 101 to the user 102. Additionally or alternatively, the wireless device 114 may process data 108 received from the data gathering system 101, and display any processed data to the user 102. For example, the wireless device 114 may operate one or more ML models to predict various sleep stages of the user 102 and / or output other sleep monitoring related data associated with the user 102 (including, for example, sleep disordered breathing events such as sleep apnea), which may be based on the collected data 108, such as biopotential data collected from the biopotential measurement device lOle.

[0057] Additionally or alternatively, the wireless device 114 may forward the data 108 received from the data gathering system 101 to the remote server 112. In another example, after the wireless device 114 processes the data 108 received from the data gathering system, the wireless device 114 may send the processed data to the remote server 112.5.2. Computing System

[0058] Fig. 2 shows an example computing system 200 comprising circuitry enabling the performance of any one or more of the processes, or any one or more of the step(s) therein, according to the described examples of the present technology. The computing system 200 may correspond to any of the computing systems / devices discussed previously with respect to Fig. 1, such as the data gathering system 101 and / or the target system 110. For example, the aforementioned implementations of the data gathering system 101 and / or the target system 110may include any one or more of the following components: one or more processors 202, memory 204, a storage element interface 206, one or more storage elements 208, a bus / interconnect 210, an input interface 214, an output interface 216, a communication interface 212, sensors, actuators, input devices, output devices, among many others.5.2.1. Processors

[0059] The processor(s) 202 may include any type of general-purpose and / or special-purpose processors or microprocessors that interprets and executes programming instructions and / or other types of executable code. The processor(s) 202 may be embodied as any suitable processors, such as one or more central processing units (CPUs), graphics processing units (GPUs), accelerated processing units (APUs), neural processing units (NPUs), tensor processing units (TPUs), microcontrollers, application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), hardware accelerators, and / or other hardware-based processors. References to a processor should be understood to include references to a single processor or a collection of processors that may or may not operate in parallel.5.2.2. Memory and Storage

[0060] The memory 204 may be of any type capable of storing information accessible by the processor(s) 202, including a computing device-readable medium. The memory may be a non- transitory medium, such as random access memory (RAM) and / or another type of dynamic storage device that stores information and instructions for execution by processor 202, read only memory (ROM) and / or another type of static storage device that stores static information and instructions for use by processor 202, a hard drive, memory card, optical disk, solid state drive, and / or other memory. The memory may include different combinations of the foregoing, whereby different portions of instructions and data are stored on different types of media. The instructions may be any set of instructions to be executed directly (such as machine code) or indirectly (such as scripts) by the processor(s). For example, the instructions may be stored as computing device code on the computing device-readable medium. In that regard, the terms “instructions”, “modules”, and “programs” may be used interchangeably herein. The instructions may be stored in object code format for direct processing by the processor, or in any other computing device language including scripts or collections of independent source code modules that are interpreted on demand or compiled in advance.

[0061] Storage element interface 206 may comprise a storage interface such as for example a Serial Advanced Technology Attachment (SATA) interface or a Small Computer System Interface (SCSI) for connecting bus 210 to one or more storage elements 208, such as one or more local disks, for example SATA disk drives, and control the reading and writing of data to and / or from these storage elements 208. Although the storage element(s) 208 is / are described as a local disk,in general any other suitable computer-readable media such as a removable magnetic disk, optical storage media such as a CD or DVD, -ROM disk, solid state drives, flash memory cards / drives, and / or the like.

[0062] Although Fig. 2 functionally illustrates the processor(s) 202, memory 204, and storage elements 208 as being within the same block, such devices may actually include multiple processor or memories that may or may not be stored within the same physical housing. Similarly, the memory 204 and / or storage elements 208 may include hard drive(s) or other storage media located in a housing different from that of the processor(s) 202, for example, in a cloud computing system. The processor(s) 202 may access the memory 204 and / or storage elements 208 over a network via the communication interface 212.5.2.3. Bus

[0063] Bus 210 may comprise one or more conductors such as metal or metal alloys (e.g., copper, aluminum, and / or the like), fiber, and / or some other interconnection means that permit communication among the various components of the computing system 200. The bus 210 may include any number of interconnect, fabric, and / or interface technologies such as, for example, SATA, peripheral component interconnect (PCI), PCI express (PCIe), USB, HyperTransport, and / or any number of other bus or interconnect technologies including proprietary buses or interconnects.5.2.4. User Interface

[0064] The user interface of the computing system 200 may include input devices and output devices connected to the input interface 214 and output interface 216, respectively.

[0065] The input interface 214 may comprise one or more mechanisms that permit an operator or user to input information to the computing device 200, such as input devices. In the example of Fig. 2, the input devices includes a keyboard 220 and a mouse 230, however, the input devices can additionally or alternatively include a pen, a stylus, physical buttons, switches, dials, knobs, keypads, touchpads, touchscreens, document scanners, headsets, voice recognition and / or biometric mechanisms (e.g., microphones, fingerprint readers, palm / hand geometry scanners, retina scanners, iris scanners, vein pattern scanners, heart rate monitors, blood pressure monitors, temperature sensors, pulse oximeters, respiratory rate monitors, blood glucose monitors, EEG sensors / electrodes / headsets, EOG sensors / electrodes, EMG sensors / electrodes, ECG sensors / electrodes, and / or the like), a camera, and / or the like. In some examples, one or more sensors, such as any of the various sensors mentioned herein, can additionally or alternatively be used as input devices. The input devices may be physical devices, or virtual devices (e.g., software components) accessible via a touchscreen or other physical input device (e.g., keyboard 220 and / or mouse 230). The input devices may, in one form, be physically connected or attached to a housingof the computing system 200, or may, in another form, be in wireless communication with the computing system 200.

[0066] Output interface 216 may comprise one or more mechanisms that output information to the operator or user, such as output devices. In the example of Fig. 2, the output device comprises a display 240, however, the output devices can additionally or alternatively include audio devices (e.g., speakers or other audio emitting devices), other visual output devices such as individual light emitting diodes (LEDs), projectors, printers, actuators and / or haptic feedback devices, and / or the like.

[0067] The input interface 214 and / or output interface 214 may include physical connectors, plugs, sockets, slots, fasteners, and / or the like for coupling the computing system 200 with input devices and output devices. In some examples, the interfaces may include USB ports, RJ45 connectors, audio jacks, power supply connectors, memory card slots / ports, and / or the like. Additionally or alternatively, the input interface 214 and / or output interface 214 may include virtual mechanisms, such as software connectors, APIs, drivers, middleware, and / or the like.5.2.5. Communication Interface

[0068] The communication interface 212 may be configured to establish wired and / or wireless communication with external computing devices 281, 282, 283. The external computing devices 281, 282, 283 may represent the data gathering system 101, the target system 110, or some other computing systems or devices. The communication interface 212 may connect the computing system 200 to one or more other computing system / devices 281, 282, 283 by means of a local area network (LAN) or a wide area network (WAN), such as, for example, the internet, an enterprise network, and / or the like. In one example, the communication interface 212 may be configured to detect and join a wired or wireless network so as to form a wired or wireless communication with one or more of the external computing devices 281, 282, 283. Additionally or alternatively, the communication interface 212 may form a low-bandwidth communication with one or more of the external computing devices 281, 282, 283.

[0069] The communication interface 212 may include one or more transceivers, which include various hardware elements to wirelessly transmit and receive data packets or to otherwise facilitate over-the-air communication. Such hardware elements may include, for example, baseband processors, digital signal processors (DSPs), switches, filters, amplifiers, antenna elements, and the like. Some transceivers may be configured to communicate over different networks and / or using different communication protocols than other transceivers. For example, the one or more transceivers may include a cellular transceiver to communicate over a cellular network, a Wi-Fi® transceiver to communicate over a wireless local area network (WLAN) / Wi-Fi network, a short- range communication transceiver to communicate over a personal area network and / or directlywith other computing systems / devices. Such short-range transceivers may include, for example, such as Bluetooth®, Adaptive Network Topology (ANT) / ANT+, Zigbee, consumer infrared protocol, and / or the like. The transceiver(s) may be regarded as a data transfer device. Additionally or alternatively, the one or more transceivers may include near-field communication (NFC) circuitry to enable the computing system 200 to transmit and receive data wirelessly with other proximate devices or over very short distances.

[0070] Additionally or alternatively, the communication interface 212 may include one or more network interface controllers (NICs) that enables computing system 200 to communicate with other computing devices 281, 282, 283 via wired connections. The one or more NICs may be configured to communicate with other devices 281, 282, 283 according to one or more protocols, such as Ethernet, ISO / IEEE 11073, fiber optics, and / or some other suitable protocol(s).5.2.6. Transducers and Sensors

[0071] In some implementations, one or more of the external computing devices 281, 282, 283 may represent one or more sensors such as, for example, inertia measurement units (IMU) comprising accelerometers, gyroscopes, and / or magnetometers; microelectromechanical systems (MEMS) or nanoelectromechanical systems (NEMS) comprising 3-axis accelerometers, 3-axis gyroscopes, and / or magnetometers; level sensors; flow sensors; temperature sensors (e.g., thermistors and / or the like); pressure sensors; barometric pressure sensors; gravimeters; altimeters; image capture devices (e.g., cameras); light detection and ranging (LiDAR) sensors; proximity sensors (e.g., infrared radiation detector and the like); depth sensors, ambient light sensors; optical light sensors; ultrasonic transceivers; microphones; skin-electrode mechanosensing structure (SEMS) devices; and / or any other suitable transducer or sensor, including any of those mentioned herein.5.2.7. Additional Components and Other Aspects

[0072] The computing system 200 may additionally or alternatively include other components not shown by Fig. 2. For example, one or more of the aforementioned sensors, input devices, and / or output devices may be embedded in or otherwise integrated with the computing system 200. Additionally or alternatively, the computing system 200 may include power circuitry to provide power for stationary and / or portable implementations, such as AC power inputs, DC power inputs, AC / DC or DC / AC converter(s), power regulators, transformers, charging / power management circuitry, batteries, wired power supply connectors / inputs, and / or wireless power charging circuitry. Additionally or alternatively, the computing system 200 may include clock circuitry (e.g., including crystal oscillators, clock generator, phased-locked loop (PLL) circuits, and / or the like) for synchronization, timing, power management, timestamping data, and / or the like.

[0073] In some implementations, some or all of the depicted components of the computing system200 may be contained in a housing, chassis, case, shell, or other type of enclosure. In some circumstances, the housing may be dimensioned for portability such that it can be shipped, carried by a human, worn by a human such as user 102, or otherwise moved from one location to another. Such housing may be formed from one or more materials that form one or more exterior surfaces that partially or fully protect the contents enclosed by such housing from potentially damaging or hazardous environmental conditions, for example, electromagnetic interference (EMI), radiofrequency interference (RFI), electromagnetic radiation, vibration, extreme temperatures, various fluids (liquids, gases, etc.), and the like, and / or to enable submergibility. Additionally or alternatively, the housing and / or surfaces thereof may include mounting implements to enable attachment of the housing to various objects or structures (e.g., furniture, buildings poles, and / or the like), racks (e.g., server racks, blade mounts, and / or the like), and / or individual users 102 (e.g., straps, belts, fasteners, and / or the like). Although Fig. 2 functionally illustrates various components as being within the same block, such devices may also be distributed across multiple circuit boards or devices, and / or may or may not be stored within the same physical housing.

[0074] As used in this application, the term "circuitry" may refer to one or more or all of the following: (a) hardware-only circuit implementations such as implementations in only analog and / or digital circuitry; (b) combinations of hardware circuits and software, such as (as applicable): (i) a combination of analog and / or digital hardware circuit(s) with software / firmware and (ii) any portions of hardware processor(s) with software (including digital signal processor(s)), software, and memory(ies) that work together to cause an apparatus, such as a mobile phone or server, to perform various functions); and (c) hardware circuit(s) and / or processor(s), such as microprocessor s) or a portion of a microprocessor(s), that requires software (e.g. firmware) for operation, but the software may not be present when it is not needed for operation. This definition of circuitry applies to all uses of this term in this application, including in any claims. As a further example, as used in this application, the term circuitry also covers an implementation of merely a hardware circuit or processor (or multiple processors) or portion of a hardware circuit or processor and its (or their) accompanying software and / or firmware. The term circuitry also covers, for example and if applicable to the particular claim element, a baseband integrated circuit or processor integrated circuit for a mobile device or a similar integrated circuit in a server, a cellular network device, or other computing or network device.

[0075] Thus, a processing device that may implement any of the processes discussed herein may include integrated chips / integrated circuits, a memory and / or other control instruction, data or information storage medium. For example, programmed instructions encompassing any of the assessment / signal processing methodologies described herein may be coded on integrated chips in the memory of the device or apparatus to form an ASIC, FPGA, and / or the like. Such instructions,with such processes, may also or alternatively be loaded as software or firmware using an appropriate processor-readable medium(s), data storage medium or memory. Optionally, such processing instructions may be downloaded such as from a server over a network (e.g., an internet, an intranet, enterprise network, and / or the like) to a processing device such that when the instructions are executed, the processing device serves as a screening, monitoring device, and / or treatment device.

[0076] In some examples, such a processing device may be the aforementioned target system 110, such as a wireless device 114 or a server 112, and may include a number of components such as a communication interface 212 to link / connect with a data gathering system 101 (or individual sensors / devices, such as any of those shown by Fig. 1) and / or receive data representing such signals from such data gathering system 101. The processing device may also include, among other components, processor(s) 202, display interface 240, a user control / input interface 214, one or more sensors, and a memory 204 / data storage 208, such as with the processing instructions of the processing methodologies / modules described herein. One or more sensors may be integral with or operably coupled with the processing device. For example, one or more sensors may be integrated or embedded in a same housing or enclosure as the processing device, and / or one or more sensors may be coupled with processing device such as through a wired (e.g., Ethernet, optical fibre, USB, and / or the like) and / or wireless link (e.g., Bluetooth™, Wi-Fi, cellular, and / or the like). The target system 110 may be configured with access to any of the processor-readable medium(s) or data storage medium(s) described herein that encompasses processor control instructions of any of the processes discussed herein. The processing device may be configured to receive requests for downloading the processor-executable instructions of the processor-readable medium to such processing devices over a network.

[0077] In other examples, such a processing device may be the aforementioned data gathering system 101, such as data gathering devices 101a, 101b, 101c, 101 d, lOle, lOlf, and may include a number of components such as a communication interface 212 to link / connect with a target system 110 and / or receive data from such target system 110. The processing device may also include, among other components, processor(s) 202, a user control / input interface 214, one or more sensors, and a memory 204 / data storage 208, such as with the processing instructions of the processing methodologies / modules described herein. One or more sensors may be integral with or operably coupled with the processing device. For example, one or more sensors may be integrated or embedded in a same housing or enclosure as the processing device, and / or one or more sensors may be coupled with processing device such as through a wired (e.g., Ethernet, optical fibre, USB, copper and / or other electrical wire, and / or the like) and / or wireless link (e.g., Bluetooth™, Wi-Fi, cellular, and / or the like). The data gathering system 101 may be configured with access to any ofthe processor-readable medium(s) or data storage medium(s) described herein that encompasses processor control instructions of any of the processes discussed herein. The processing device may be configured to receive requests for downloading the processor-executable instructions of the processor-readable medium to such processing devices over a network. In one example, such a processing device may be a biopotential measurement device lOle, which may be considered in reference to Figs. 3 through 9.5.3. HSAT Monitoring and Detection System

[0078] As previously mentioned, in one form, the present technology may include an HSAT monitoring and detection system for monitoring and / or evaluating sleep staging performance, such as for use in an HSAT, which may be involved in and / or configured for any of the processes discussed herein. Such a system may include a data gathering system 101, such as the biopotential measurement device lOle and optionally one or more data gathering devices 101a, 101b, 101c, lOld, lOlf, and a target system 101.

[0079] In some implementations, the data gathering system 101, such as the biopotential measurement device lOle, collects or captures biosignals and / or biopotential of a user 102, conveys the biosignals / biopotential (or data related to the biosignals / biopotential) to a target system 110, and the target system 110 evaluates the monitored biosignals / biopotential to detect and / or diagnose sleep (e.g., sleep states), sleep disordered breathing, and sleep patterns.

[0080] In other implementations, the data gathering system 101, such as the biopotential measurement device lOle, collects or captures biosignals / biopotential of a user 102 and evaluates the biosignals / biopotential to detect and / or diagnose sleep, sleep disordered breathing, and sleep patterns (hereinafter sleep related information). In these implementations, the data gathering system 101 may communicate results of the evaluation to the target system 110 for further processing and / or analysis. Optionally, the biopotential measurement device lOle may include multiple sensors, such as a microphone and / or accelerometer, to enable the device lOle to additionally measure sound such as with the microphone for detecting snoring sounds and / or to enable the device lOle to measure position and / or motion such as with the accelerometer for detecting, for example, body position and / or motions (e.g., breathing and / or wake related motions).

[0081] In any of the aforementioned implementations, data synchronization techniques may be used to synchronize data collected by different data gathering devices lOla-f. In one example, data collected by the biopotential measurement device lOle may be time synchronized with data collected by the finger sensor 10 If, such as, via short-range communication between the biopotential measurement device lOle and the finger sensor 10 If. In another example, data collected by the biopotential measurement device lOle may be time synchronized with data collected by the finger sensor 10 If via short-range communication between the biopotentialmeasurement device lOle, the finger sensor lOlf, and a target system 110. Otherwise, such measured data signals may be synchronized by time stamping.

[0082] In any of the aforementioned implementations, the evaluation or processing of the captured / collected biosignals and / or biopotential signals can include inputting or feeding the biosignals / biopotential, optionally with other data collected by other data gathering device(s) 101a, 101b, 101c, 101 d, and / or 10 If, to one or more untrained ML models (e.g., as a training dataset). The training dataset can include labeled or unlabeled data. The one or more untrained ML models use the input training data to learn one or more patterns about the sleep staging of monitored users 102. For example, model parameters (e.g., weights, biases, and / or the like) of the untrained model(s) may be initialized with random or default values, and during a training process, the untrained model(s) learn from the input training data by adjusting the model parameters based on the training dataset, which may be done using suitable optimization algorithms, loss functions, and / or the like. Additionally or alternatively, the evaluation or processing of the monitored biosignals and / or biopotential signals can include inputting or feeding the monitored biosignals / biopotential, optionally with other data collected by other data gathering device(s) 101a, 101b, 101c, 101 d, and / or 10 If (e.g., as an inference dataset), to one or more trained ML models. The one or more trained ML models uses the input data to generate one or more inferences about sleep related information, such as the sleep staging, of the monitored patient.

[0083] The HSAT monitoring and detection system of the present technology may provide automated sleep staging detection that is at least comparable to standard visual sleep scoring of typical PSG systems. The HSAT monitoring and detection system attains similar functionality of in-lab PSG systems while being usable in in-home scenarios, uses fewer components than typical in-lab PSG systems, and provides enhanced user experience in terms of improved ease-of-use and user comfort.5.3.1. Biopotential Measurement Device

[0084] As previously discussed, biopotential signals can be measured with a biopotential sensor, such as a head or face worn sensor. Such as sensor may be a single-channel biopotential sensor device, such as biopotential measurement device 10 le. Incorporating a single-channel biopotential measurement device lOle in an HSAT setup may widen the applicability of sleep apnea testing, improve accuracy of total sleep time (TST) estimation, and allow for extended phenotyping of sleep apnea. Additionally, the biopotential measurement device lOle may be capable of attaining similar functionality to in-lab / clinic PSG systems, while potentially optimizing ease-of-use, comfort, and cost. The biopotential measurement device lOle may be considered in reference to Figs. 3 through 9.

[0085] Referring to Fig. 3, example head / facial regions for sensor-based biopotential measurementof the present technology, are shown. As illustrated by Fig. 3, biopotential measurement device lOle includes facial electrodes 410, 420 and a positioning structure 419 connecting the electrodes and may comprise a wire (such as within insulation) electrically coupled to at least one of the electrodes. The positioning structure may be configured to, when in use on a patient's face, allow and restrict positioning of electrodes for measurement. That is, the positioning structure 419 may enable (or constrain) the positioning of a reference electrode, such as electrode 410, with respect to a measurement electrode, such as electrode 420, so that the electrodes can obtain a transverse- ocular measurement vector TOMV, which may preferably align at an angle (non-parallel) to the axis 310 attributable to the sagittal plane. Additionally or alternatively, the transverse-ocular measurement vector TOMV may be formed between the electrodes 410, 420 when in use, such as when the biopotential measurement device lOle is placed on the face for measuring biopotential signal(s). In this regard, the facial electrodes 410, 420 may be located in particular regions or quadrants of the face, which are, for example, delineated by imaginary axes shown by axis line 310 and axis line 320. Axis line 310 may be considered to be an imaginary axis along a sagittal plane, extending vertically through the middle region of the face. Thus, the axis line 310 may be referred to as a “sagittal axis 310”. Axis line 320 may be considered an imaginary axis perpendicular to the sagittal axis and extending horizontally approximately across a vertical middle of a rhinal region of the face. This vertical middle of the rhinal region of the face may be a region of the face referred to as the “rhinion”, which is the midline junction where the nasal bone meets the septal cartilage. Axis line 320 may divide the face into a supra-ocular region (e.g., including the eyes and the area above the eyes) and a sub-ocular region (e.g., including the area below the eyes). Thus, the axis line 320 may be referred to as an “ocular axis 310”.

[0086] Facial Quadrants 301, 302, 303, and 304 are considered in relation to axis lines 310 and 320. Thus, facial quadrant 301 is considered to be a quadrant that is the sagittal-left and supraocular, which generally corresponds to the upper right side of the face. Facial quadrant 302 is considered to be a quadrant that is sagittal-right and supra-ocular, which generally corresponds to the upper right side of the face. Facial quadrant 303 is considered to be a quadrant that is sagittal- left and sub-ocular, which generally corresponds to the lower right side of the face. Facial quadrant 304 is considered to be a quadrant that is sagittal -right and sub-ocular, which generally corresponds to the lower left side of the face.

[0087] Accordingly, in various implementations of the present technology the facial biopotential measurement device 400 may be positioned or configured for measurement of particular quadrants of the face. For example, the measurement electrode 420 may be configured as a sagittal -right facial electrode as shown by Figs. 3 and 4. In another example, the measurement electrode 420 may be configured as a sagittal -left facial electrode as shown by Fig. 5. In these examples, themeasurement electrode 420 may be configured as a sub-ocular facial electrode, being placed below the ocular axis 320. Additionally, the reference electrode 410 may be configured as a supra-ocular facial electrode, being placed above the ocular axis 320. As shown by Figs. 3-5, the reference electrode 410 is shown as being placed on the sagittal axis 310, and thus, may be configured as a sagittal -neutral facial electrode. In other examples, the reference electrode 410 may be configured as a sagittal-left facial electrode or as a sagittal-right facial electrode. In other implementations, the measurement electrode 420 may be configured as a supra-ocular facial electrode and the reference electrode 410 may be configured as a sub-ocular facial electrode.

[0088] Accordingly, the head or facial biopotential measurement device 400 may be employed as illustrated in Figs. 3, 4, or 5. The positioning structure 419 may be formed of a suitable conductive material, such as copper, aluminum, gold, silver, steel, tungsten, nickel, tin, among many others, and / or any combination(s) thereof. As an example, the positioning structure 419 may be formed of a conductive material, such as copper wire and / or the like, with a diameter of between about 1 millimeter (mm) and about 5mm. Such a positioning structure 419 may be suitably resilient to allow placement on the user’s face and arranged around an eye of the user and without the positioning structure 419 falling onto / against the eye. In some implementations, the positioning structure 419 may be formed of a suitable fiber optic cable / wire. The positioning structure 419 may also include an insulator, jacket, or non-conductive covering, such as polyvinyl chloride (PVC), polyethylene (PE), cross-linked PE (XLPE), teflon (e.g., polytetrafluoroethylene (PTFE), fluorinated ethylene propylene (FEP), perfluoroalkoxy alkanes (PFA), and / or the like), thermoplastic elastomers (TPE), thermoplastic polyurethane (TPU), natural and / or synthetic rubber, silicone, neoprene, polyurethane, acrylate polymer coating(s), among many others, and / or combination(s) thereof. In one form of the present technology, the positioning structure 419 may be embedded in, or otherwise coupled to, an adhesive strip that can be applied to the skin, directing the wire around the eye. The adhesive strip may be contiguous (such as along the length of the positioning structure 419), or comprise discrete portions, in which the positioning structure 419 may be embedded, or to which the positioning structure 419 is otherwise coupled. This may prevent the user’s hand / finger from inadvertently pulling on the positioning structure 419 while they sleep and potentially pulling the device from the face. In one example implementations, the positioning structure 419 is formed of 1.5mm copper wire with a TPU coating, the electrodes 410, 420 are embodied as Red Dot 2560 electrodes provided by 3M®, and the casing is formed from a polycarbonate material.

[0089] As previously noted, the positioning structure 419 connects the electrodes to position them, or promote such positioning, when in use on a patient's face, for forming a transverse-ocular measurement vector TOMV. In this regard, biopotential may be measured between two electrodesjuxtaposed across the eye, such as from the cheek to the forehead or forehead center. For example, as shown by Figs. 3, 4, or 5, a reference electrode 410 may be positioned by the structure to be attached to the center of the user’s 102 forehead while the measurement electrode 420 may be positioned to be attached the user’s 102 cheek. However, in other implementations the reference electrode 410 may be positioned by the structure to be attached to the cheek while the measurement electrode 420 may be positioned by the structure to be attached to the center of the forehead. In some implementations, the minimum distance or spacing between the electrodes 410, 420, as enabled by the positioning structure, may be any distance / spacing sufficient to span the user’s 102 eye. The positioning structure may be flexible and may be flexibly resilient. As illustrated, the positioning structure may be formed with resilient curve, or may be bendable, to curve about the eye so as not to obstruct vision. Thus, the positioning of the electrodes 410, 420 creates an imaginary axis, such as the axis shown at TOMV, between the electrodes 410, 420 that crosses the eye, and the biopotential signal is measured across that axis.

[0090] The biopotential measurement device 400 may be further applied to the user 102 by means of two adhesive electrode patches 415, one positioned centrally on the forehead, and one positioned below the eye. The reference electrode 410 and the measurement electrode 420 may be, for example, attached to respective patches 415 with a snap connector 901 (see Fig. 9). The reference electrode 410 is a stable electrode with a known and constant biopotential, and serves as a reference point for measuring the potential of the measurement electrode 420. The reference electrode 410 may be or act as a ground electrode. The measurement electrode 420 is an electrode that is used to measure the biopotential of the user 102. The potential difference between the measurement electrode 420 and the reference electrode 410 is measured, and this measured biopotential signal may be used to determine biopotential properties or parameters of the user 102, such as one or more of EEG signals related to brain activity, EOG signals related to ocular activity, EMG signals related to mandible (jaw) activity, and / or the like.

[0091] Specifically, the positioning of the biopotential measurement device 400, such as when forming the transverse-ocular measurement vector TOMV, can allow the biopotential measurement device 400 to capture a single biosignal having a superposition of signals, such as EEG (brain), EOG (eye), and EMG (chin / jaw muscle) biosignals. In other words, the biopotential measurement device 400 measures / captures a single biopotential signal with a sufficient resolution of a plurality of different biopotential signals, (e.g., each of the aforementioned biosignal types, or at least two of the types, (e.g., EEG, EOG, EMG)) so as to not require a different measurement electrode and lead for each Optimal electrode placement for measuring each EEG, EOG, and EMG is typically different than those shown by Figs. 3, 4, and 5. However, the electrode locations shown by Figs. 3, 4, and 5 may be considered a compromise that enables providing a single signalusing a single measurement electrode to obtain a sufficient resolution of each of these biosignals. That is, the electrode locations, and the structure of the biopotential measurement device 400, strikes a balance between minimization of sensing devices / electrodes while maximizing the number of detectable biosignals. The positioning of the electrodes in these ways allows for the measurement or derivation of multiple biosignals used for sleep related information detection using a minimum number of electrodes for such measurements. This advantageous electrode placement strikes a balance between biosignal measurement accuracy and reduced complexity, which enables ease of use and comfort for in-home PSG.

[0092] For example, a notable discrepancy when comparing the biopotential measurement device 400 with other sleep study devices is a relative underperformance of N3 determination compared to REM. This discrepancy can be explained by the specific choice of electrode placement of the biopotential measurement device 400. That is, the polarity that is provided with information across the eye helps to differentiate REM from deep sleep state classification. The absence of Al or A2 reference electrode reduces the amplitude of delta waves. Instead, structuring a device to position the electrodes 410, 420 in a diagonal angle across the lateral cross section of the eye ensures that the polarity of rapid eye movements can be adequately captured, which greatly enhances the capacity to discriminate between N3 and REM sleep when using only a single-channel setup, such as the biopotential measurement device 400.

[0093] The biopotential measurement device 400 demonstrates significant similarity in overall sleep staging performance as well as in the trade-off of specificity and sensitivity for each of the individual sleep stages when compared to other sleep study devices. As such, the biopotential measurement device 400 produces favorable results as it was designed into a small form factor with consideration to electrode placement so as to optimize for overall sleep stage classification (e.g., 5-stage classification) accuracy while minimizing patient discomfort by avoiding an excessive number of electrodes and avoiding electrode placement above the hairline.5.3.2. Biopotential Measurement Device Components

[0094] Referring now to Figs. 4 to 9, which show various views of the example biopotential measurement device 400. In particular, Figs. 4 and 5 show example positioning of biopotential measurement device 400 on a user 102; Fig. 6 shows top and bottom views of the example biopotential measurement device 400; Fig. 7 shows side views of the biopotential measurement device 400; Fig. 8 shows front and back views of a base 405 of the biopotential measurement device 400; and Fig. 9 shows front, top, and bottom views of an adhesive electrode patch 415 used in the biopotential measurement device 400.

[0095] In various implementations of the present technology, two electrodes 410, 420 may be employed for determining a biopotential signal, which may be measured as a difference in thereadings across the two electrodes 410, 420. Additionally or alternatively, the biopotential signal may be a measure of voltage, current, or field strength using suitable transducers and / or sensors.

[0096] The biopotential measurement device 400 includes a base 405 that contains a reference electrode 410 and a measurement electrode 420 that is connected to the base 405 or the reference electrode 410 via the positioning structure 419 that may comprise an electric wire 430. Specifically, the reference electrode 410 may be directly attached to the base 405 and the measurement electrode 420 may be attached to the base 405 via the positioning structure. The electric wire 430 may electrically couple the measurement electrode to the measurement circuits of the base 405, which may comprise a controller with one or more processors. As previously noted, in some implementations, the positioning structure may be formed to have a length suitable to naturally yield the positioning of the electrodes 410, 420 for measurement across the eye such that it may wrap or snake around the eye.5.3.2.1. Electrodes

[0097] The reference electrode 410 and the measurement electrode 420 are biopotential electrodes. A biopotential electrode is a transducer that senses ion distribution on the surface of tissue, and converts the ion current to electron current. One side of the electrode 410, 420 comes into contact with tissue and mechanically couples with the tissue for signal transduction. The other side of the electrode 410, 420 includes conductive metal attached to a lead, wire, or other interconnection element connected to a measuring instrument, such as a biopotential sensor / transducer / signal sampler in or on the base 405. Thus, each electrode 410, 420 may include an electrode pad / patch 415, a connector, such as a fastener or other type of connector, and a lead / wire that connects the electrode 410, 420 to the measurement circuitry. For example, the reference electrode 410 may include the patch 415a and connector, such as a snap connector, of the base 405 (not shown) and the measurement electrode 420 may include the patch 415b and a connector, such as a snap connector, of a connection cable / wire 430.

[0098] The electrodes 410, 420 may be implemented as skin contact electrodes. For example, the electrodes 410, 420 may be applied directly to the skin using a dry electrode patch 415. Alternatively, the electrodes 410, 420 may be applied indirectly to the skin using a wet electrode patch 415, wherein an electrolyte gel, paste, adhesive, or other conductive element is disposed between the tissue and the electrode patch 415. In the depicted examples, adhesive electrode patches 415 are used to attach respective electrodes 410, 420 to the face.

[0099] As shown by Figs. 3 and 4, the adhesive electrode patch 415 may be a crescent or bean shaped conductive pad. As shown by Fig. 5, the adhesive electrode patch 515 may be a square or rectangularly shaped conductive pad. It should be understood that the electrode patches / pads 415, 515 can be formed in any suitable shape, such as circular, needle-shaped, irregular shapes, amongmany other possibilities. Furthermore, aspects discussed in the present disclosure related to the electrode patch 415 may also apply to the electrode patch 515 unless explicitly stated otherwise.

[0100] As shown by Figs. 6 and 9, each electrode pad 415 includes a conductive section 601 and an adhesive section 602. The adhesive section 602 may mechanically couple the patch 415 with the tissue. The conductive section 601 may include a conductive material that interacts with the tissue / skin for signal transduction. In the example of Fig. 6, the conductive section 601 is shown as having a circular shape. In other implementations, the conductive section 601 may be formed in some other shape or pattern, such as a rectangular or square shape, a grid pattern, a coil or spiral pattern, and / or any other suitable shape / pattern. The conductive section 601 may be formed using any suitable materials, such as any combination of stainless steel, silver (Ag), gold (Au), chromium (Cr), copper (Cu), titanium (Ti), silicon (Si), platinum (Pt), conductive polymer(s) such as Nafion® (fluoropolymer-copolymer), carbon materials (e.g., carbon fiber, carbon nanotubes (CNT), graphene, carbon black, and the like), and / or other known materials, or composites thereof. The adhesive section 602 may be formed using any suitable materials, such as any combination of silicone, foam, polyester, nylon, various fabrics, polyethylene, polyethylene terephthalate (PET), polydimethylsiloxane (PDMS), polyurethane, rayon acetate, and / or other known materials, or composites thereof. The adhesive section 602 may also include any suitable adhesive material, or may be formed of a self-adhesive layer. In other implementations, the electrode pads 415 may be implemented as metal thin film electrodes, flexible printed circuit boards (FPCBs), graphene electronic tattoo electrodes, PWS films, and / or other sensor / electrode technologies.

[0101] In other implementations, the electrodes 410, 420 may be implemented as non-contact electrodes. For example, one or more electrodes 410, 420 may be embedded in a non-conductive component, such as a polymer (e.g., silicone, non-conductive fabric, and / or the like), and may be located near the skin with the non-conductive component separating the skin and the electrode. In these implementations, such non-contact sensors may be utilized to measure the biopotential signal by detecting field strength. Additionally or alternatively, the electrodes may be formed with any suitable metal or conductor. For example, electrodes may be formed with conductive inks, such as in the case of printing or dying the electrode on other components. For example, conductive inks may be employed to print or dye an electrode on a headgear strap, a mask cushion, a mask frame or some other headgear.

[0102] Similarly, the electrode may be formed as a fabric or cloth electrode. For example, an electrode may be formed or weaved with conductive threads or fine wire. For example, cloth may be impregnated with the wires or threads. Additionally or alternatively, such an electrode may be weaved into the fabric of a strap of headgear or a headgear support, or an eye mask or sleep mask. In one example implementation, the biopotential measurement device 400 can be embeddedin a sleep / eye mask wherein the reference electrode 410 is woven into a central section of the eye mask that is configured to be placed over the forehead, the measurement electrode 420 is woven into a lower section of the eye mask that is configured to be placed underneath one of the eyes, and the wire 430 is woven into the eye mask such that the wire 430 loops through a structure of the eye mask. Additionally or alternatively, electrodes may be constructed from or embedded in a conductive polymer.

[0103] As shown by Fig. 9, the conductive section 601 also includes a connector 901, such as a snap fastener or some other fastener, that allows the electrode patch 415 to be connected to a corresponding connector / fastener of its respective electrode 410, 420. The connector 901 may be configured to be connectable to conventional electrode snap clips or snap connectors, alligator clips, and / or other connectors / fasteners. Additionally or alternatively, the connector 901 may be a male connector configured to be inserted into a socket or female connector of an electrode 410, 420 (not shown). For example, the reference electrode 410 may be attached to an electrode patch 415a via a snap connector 901 of the electrode patch 415a, and the measurement electrode 420 may be attached to an electrode patch 415b via a snap connector 901 of the electrode patch 415b 5.3.2.2. Base

[0104] A biopotential measurement device 400 in accordance with the present technology may include a base 405. The base 405 comprises a housing that contains or otherwise includes a biopotential measurement sensor / transducer, which generates biopotential signals based on biopotential measured across the electrodes 410 and 420.

[0105] As shown by Figs. 6, 7, and 8, the base 405 includes an input device 605 in the form of a button, which allows the user 102 to interact with the device 400. Additional or alternative input devices 605, such as additional or alternative buttons, switches, dials, a touchscreen, microphones, and / or the like, may be additionally or alternatively included in / on the base 405 in other implementations. In the examples of Figs. 6, 7, and 8, the input device 605 is a physical button. However, in other implementations, the input device(s) 605 can include software components accessible via a touchscreen of the device 400 and / or software components accessible via a communication interface and a target system 110. The input device(s) 605 may be physically connected to the external housing, or may be in wireless communication with a communication interface that is in electrical connection to a controller / processor of the device 400.

[0106] The base 405 is also shown as including an output device 406 to display various operational states of the biopotential measurement device 400. In the examples of Figs. 6, 7, and 8, the output device 406 is in the form of at least one LED. Additional or alternative output devices 406, such as additional or alternative LEDs, a touchscreen display, audio speakers, haptic devices, and / or the like, may be additionally or alternatively included in / on the base 405 in other implementations.

[0107] Furthermore, the base 405 may contain one or more additional hardware elements such as, for example, one or more processors 202, memory 204, and / or communication interface(s) 212 to process and / or communicate the captured biopotential signals / data to a target system 110. The additional hardware elements may also include one or more sensors, such as one or more orientation sensors, microphones, and / or the like. The orientation sensors may be used to determine, for example, a head orientation (e.g., supine versus non-supine) when the device is attached to the user 102. Examples of such orientation sensors can include accelerometers, gyroscopes, magnetometers, and / or the like. In some implementations, these orientations sensors may be part of an IMU, MEMS, or NEMS circuitry. A microphone may additionally or alternatively be included to capture and record sounds and / or sound levels made during various sleep stages / states. In some cases, the placement of the base 405 on the user’s 102 forehead may allow for optimal measurement of sensor data from these sensors, such as for recording the orientation of the head and / or for recording sound with the microphone (e.g., snoring and the like). 5.3.2.3. Software Application

[0108] A biopotential measurement system in accordance with one form of the present technology may include hardware elements (e.g., one or more processors 202, memory 204, and / or the like) to execute, run, or otherwise operate a software application that derives EEG, EOG, and EMG data by filtering the raw biopotential data generated by the biopotential measurement device 400 comprising a single channel measurement device.

[0109] In some implementations, circuitry of the biopotential measurement device 400 operates the software application to process the sensor data / signals produced by the various sensors of the biopotential measurement device 400. For example, the biopotential measurement device 400 may derive EEG, EOG, and EMG data from the biopotential signal measured and captured by the biopotential measurement sensor(s), such as from a single channel measurement device, determine a head orientation based on the sensor data / signals produced or measured by orientation sensor(s), sleep state sounds based on audio data recorded by microphone(s), and / or the like. The data may be derived by filtering, such as by hardware and / or software filters, as discussed in more detail herein. This data may be stored in the memory 204 of the biopotential measurement device 400 and / or may be streamed directly to a target system 110. Here, the software application may control a local communication interface 212 to stream or otherwise transfer the processed data to the target system 110 for further processing and / or analysis.

[0110] Additionally or alternatively to processing the raw data on the device 400 itself, the software application may control the communication interface 212 to stream raw sensor data / signals to the target system 110 for processing and / or analysis in a similar manner as previously described. In either implementation, a corresponding software application on the targetsystem 110 receives the raw or processed data stream in real time via wired or wireless connection during a sleep session, or receives the data, for example, as a single data file or in data chunks, after the sleep session is completed. The corresponding software application on the target system 110 can also receive other raw or processed sensor data from other sensors / devices, such as the finger sensor lOlf and / or some other data gathering devices 101. The corresponding software application can correlate or otherwise process this other raw or processed sensor data, such as by the filtering techniques described herein, for determining sleep related events such as sleep staging (e.g., sleep stages including, e.g., REM, non-REM (e.g., stages Nl, N2 and / or N3), etc.) and / or for diagnosing sleep disorder events such as discussed herein. Sleep staging may be based on sleep staging events derived from the sensor data. Sleep staging events may refer to events related to one or more sleep stages and / or features within or derived from biosignals and / or biopotential signals that are indicative of one or more sleep stages. These can include, but are not limited to, EEG features such positive occipital sharp transients of sleep (POSTS) and vertex waves, sleep spindles and K complexes, delta waves, diffuse attenuation of signal amplitudes, and EMG and / or EOG features such as muscle tone and eye movements.5.3.3. Statistical Analysis and Algorithm5.3.3.1. Performance Endpoint Selection[OHl] Since there are no standardized performance targets established for sleep stage determination, endpoints that have been used in various studies of wearable biopotential-based sleep staging devices can be considered. Such studies can include, for example, Imtiaz, A Systematic Review of Sensing Technologies for Wearable Sleep Staging, SENSORS, vol. 21, no. 5, p. 1562 (24 Feb. 2021) and validation studies on wearable biopotential devices using at least EEG that were suited for at-home sleep staging analysis. Examples of such wearable biopotential devices include Zeo system provided by Zeo, Inc. (see Griessenberger et al., Assessment of a wireless headband for automatic sleep scoring, SLEEP AND BREATHING, vol. 17, no.2, pp. 747-752 (May 2013; published online: 21 Sep. 2012); and Shambroom et al., Validation of an automated wireless system to monitor sleep in healthy adults, J SLEEP RES, vol. 21, no. 2, pp. 221-230 (Apr. 2012)), The Dreem Headband™ provided by Beacon Biosignals, Inc. (see Amal et al., The Dreem Headband compared to polysomnography for electroencephalographic signal acquisition and sleep staging, SLEEP, vol. 43, no. 11, p. zsaa097 (Nov. 2020); and Arnal et al., The Dreem Headband as an Alternative to Polysomnography for EEG Signal Acquisition and Sleep Staging, BIORXIV (10 Jun. 2019), doi: 10.1101 / 662734)), SOMNOwatch® provided by SOMNOmedics AG (see Voinescu et al., Assessment of SOMNOwatch plus EEG for sleep monitoring in healthy individuals, PHYSIOLOGY & BEHAVIOR, vol. 132, pp. 73-78 (10 Jun. 2014); and Hof Zum Berge et al., Portable PSG for sleep stage monitoring in sports: Assessment of SOMNOwatch plus EEG,EUR J SPORT SCI, vol. 20, no. 6, pp. 713-721 (2 Jul. 2020)), SleepProfiler™ provided by Advanced Brain Monitoring, Inc. (see Finan et al., Validation of a Wireless, Self-Application, Ambulatory Electroencephalographic Sleep Monitoring Device in Healthy Volunteers, JOURNAL OF CLINICAL SLEEP MEDICINE, vol. 12, no. 11, pp. 1443-1451 (15 Nov. 2016); Lucey et al., Comparison of a single-channel EEG sleep study to polysomnography, J SLEEP RES, vol. 25, no. 6, pp. 625-635 (Dec. 2016; published online: 2 Jun. 2016); and Levendowski, Premarket Notification: Automatic Event Detection Software for Polysomnograph with Electroencephalograph, USFDA, 510(k) no. K120450, Vol 882.14002012 (2012)), the Zmachine® provided by General Sleep Corp, (see Kaplan et al., Evaluation of an automated single-channel sleep staging algorithm, NATURE AND SCIENCE OF SLEEP, pp. 101-111 (Sep. 2015)), and the Cognionics study device provided by Cognionics, Inc. (see Casciola et al, A Deep Learning Strategy for Automatic Sleep Staging Based on Two-Channel EEG Headband Data, SENSORS, vol. 21, no. 10, p. 3316 (11 May 2021)).

[0112] Informative and transparent endpoint parameters that are consistently reported in the aforementioned studies include stage-wise and overall accuracy, stage-wise and overall Cohen’s kappa, and stage-wise sensitivity and specificity. Another commonly reported outcome are the sleep staging confusion matrices. When the aforementioned parameters are not readily available, they can be calculated from the confusion matrices provided in their respective studies. The sleep staging outcomes can be considered using standard length epochs, such as 30 second epochs.

[0113] As sensitivity and specificity are 2-class parameters, they can only be calculated in a stagewise manner. Accuracy and Cohen’s Kappa on the other hand can be calculated both on a stage and on an overall level. Cohen's kappa coefficient (K) is a statistic that is used to measure interrater reliability for categorical items. It is generally considered to be a more robust measure than the simple percent agreement calculation, as K considers the possibility of the agreement occurring by chance.

[0114] Additionally or alternatively, parameters that are commonly derived from the sleep stages for PSG analysis can be considered such as, for example, Total Recording Time (TRT), TST, Sleep Efficiency (SE), Sleep Onset Latency (SOL), Wake After Sleep Onset (WASO), and time spent in each of the sleep stages. TRT is the total sum of non-rejected epochs, and TST is the total sum of those epochs spent in sleep. SE is calculated as TST / TRT x 100%. SOL is defined as the time from the start of a recording until the first n number of consecutive epochs of sleep (e.g., where n = 3 or some other number). WASO is the time spent in wake after the SOL point. For each of these parameters, and for the time spent in each of the sleep stages (Wake, Nl, N2, N3, REM), the mean, bias, and Pearson correlation coefficient can be calculated together with their 95% confidence intervals.5.3.3.2. Data Synchronization

[0115] In implementations where multiple data gathering devices 101 are used, such as the biopotential measurement device 400, finger sensor 10 If, and / or the like, the data from each of devices 101 can be algorithmically synchronized with one another. In one example, the synchronization may be accomplished by matching the instantaneous heart rate traces derived from an electrocardiogram trace, a pulse rate (PR) trace, and / or a peripheral arterial tone signal trace of the finger sensor 101 with the biopotential data / signals of the biopotential measurement device 400. Additionally or alternatively, the data collected and / or measured by each sensor 101 may be timestamped and correlated with one another during the processing stage. Additionally or alternatively, data epochs with high lead impedance and data epochs with disconnections or weak communication signals can be discarded or rejected from the analysis.5.3.3.3. EEG pre-processing

[0116] As mentioned previously, EEG, EOG, and EMG data may be derived by filtering raw biopotential signals / data generated by the biopotential measurement device 400 that may have at least one single channel measurement device. Specifically, the biopotential measurement device 400 may produce a single biopotential signal, but when filtered differently, a plurality of signals, such as at least three different signals may be generated from the sensed biopotential signal. For example, with different filters, such as a filter for each different signal, data representing a plurality of signals (e.g., EEG, EOG, and EMG) can be generated. That generated signals may then be classified by a trained ML model into sleep stages, such one of five sleep stages (e.g., wake, Nl, N2, N3, and REM), or sleep states, such as wake or sleep. For example, in some implementations, the biopotential measurement device 400 in accordance with the present technology may generate raw EEG data based on biopotential signal measurement. As an example, the raw EEG data may be preprocessed as follows.

[0117] The signal from the biopotential measurement device 400 may be bandpass filtered to a frequency range suitable for EEG. For example, the signal may be filtered to pass frequencies between, for example, 0.5 Hertz (Hz) and 35 Hz, and optionally resampled, such as to 100 Hz, to generate an EEG data trace / signal. Next, the EEG trace may be normalized by, for example, subtracting the median and dividing by the interquartile range. Since the Sleep Staging Model (SSM) may make use of a recurrent neural network (RNN) (discussed infra) that expects a two- dimensional matrix as input, the EEG signal may be transformed into a time-frequency image X by means of a Short-Time Fourier Transform (STFT). The STFT may include windowing, wherein the input signal is divided into segments, epochs, or frames. A Fourier transform may be applied to each segment / epoch / frame to analyze its frequency content. The Fourier transform converts the signal from the time domain to the frequency domain, providing information about the amplitude and phase of different frequency components present in the segment / epoch / frame. The STFTresults in a time-frequency representation of the signal, such as a spectrogram, which shows how the signal’s frequency content changes over time, for example, with time on the x-axis, frequency on the y-axis, and intensity or magnitude represented by color or grayscale. In one example, the input parameters for the STFT can include a Hamming window with a 50% overlap, and a window size of 2 seconds. To obtain the log-power spectrum, logarithmic scaling is applied as follows:Xz= 20 x logW\X\

[0118] These power spectrograms can then be epoched to serve as input data to the deep learning models. In some implementations, EOG and / or EMG data may also be preprocessed, filtered, and / or derived in a same or similar manner as the EEG data described previously. However, the frequency passband range of the filters may be chosen for producing EOG and EMG respectively. For instance, the EOG and / or EMG data may be preprocessed, filtered, and / or derived from the single electrode signal, along with the EEG signal based on the filtered frequencies. For example, the frequency ranges for the EOG, EMG, and EEG signals may be between 0.1 Hz and 35 Hz, between 70 Hz and 110 Hz, and between 0.5 Hz and 35 Hz, respectively. In one example, the signal may be filtered to pass frequencies between, for example, 0.1 Hz and 35 Hz to generate an EOG data trace / signal. In the same or another example, the signal may be filtered to pass frequencies between, for example, 70 Hz and 110 Hz to generate an EMG data trace / signal. In the same or another example, the signal may be filtered to pass frequencies between, for example, 0.5 Hz and 35 Hz to generate an EEG data trace / signal. In the respective frequency ranges, the EOG signal is mainly in the lower frequency portion of the ranges and of higher amplitude relatively to the EEG. An EOG signal fluctuation is typically approximately ten times the amplitude of an EEG fluctuation. An ML algorithm such as described herein, such as a deep learning algorithm, can learn EOG-related features and EEG-related features based on the relevant filtered frequency bands. The EOG and EMG traces may also be normalized and transformed into a time-frequency image by means of an STFT in a same or similar manner as discussed previous with respect to the EEG signal.5.3.3.4. Neural Network Architecture

[0119] A common multi-channel PSG system (e.g., a sleep lab EEG setup) study is usually derived from the standard 10-20 protocol proposed by Herbert Jasper. See e.g., Jasper H., Report of the committee on methods of clinical examination in electroencephalography, ELECTROENCEPHALOGR CLIN NEUROPHYSIOL, vol. 10, no. 2, pp. 370-375 (1958). In such a system, a set of anatomical landmarks are identified to guide electrode placement. The electrodes are placed at 10% or 20% intervals along the anatomical directions with the ground electrode. Different electrodes present distinct sleep signatures depending on their position. The 2007 AASM scoring rules recommend the use of frontal electrodes to detect K-complexes and delta wave activity, central electrodes todetect sleep spindles and occipital electrodes to detect alpha waves. See Silber et al., The visual scoring of sleep in adults, JOURNAL OF CLINICAL SLEEP MEDICINE (JCSM), vol. 3, no. 2, pp. 121- 131 (2007). These sleep signatures are used to identify sleep stages and are used for visual EEG sleep assessment.

[0120] Since the biopotential measurement device 400 of the present technology includes a single lead biopotential module, which can be considered to record a superposition of frontal EEG, EOG, and EMG, it does not easily unambiguously present each of the typical sleep signatures. Accordingly, advanced machine learning (ML) techniques, such as deep learning and / or the like, may be used to overcome this challenge by learning and interpreting sleep-related patterns that are visually less straightforward to discern. Deep learning mechanisms, such as deep neural networks (DNNs), have an advantage over traditional ML algorithms in that they do not require domain knowledge for manual feature extraction, and therefore, can self-adapt to different EEG setups. See Eldele et al., An Attention-Based Deep Learning Approach for Sleep Stage Classification with Single-Channel EEG, IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, vol. 29, pp. 809-818 (28 Apr. 2021) (fEldelefi. In various implementations, the biopotential measurement sensor system, which may include the biopotential measurement device 400 and / or at a target system 110, makes use of such deep learning network frameworks. In these implementations, a DNN includes a combination of epoch and sequence based RNN layers. RNNs are effective in capturing temporal dependencies in time-series such as EEG data. See Eldele. In other implementations, other DNN topologies, configurations, and / or arrangements may be used. An example of such DNNs may be considered in reference to Fig. 10.

[0121] Fig. 10 illustrates an example NN 1000, which may be suitable for use by one or more of the computing systems mentioned herein. The NN 1000 may be deep neural network (DNN) used as an artificial brain of a compute node or network of compute nodes to handle very large and complicated observation spaces. Additionally or alternatively, the NN 1000 can be some other type of topology (or combination of topologies), such as feed forward NN (FFN), deep FNN (DFF), convolutional NN (CNN), deep CNN (DCN), deconvolutional NN (DNN), a deep belief NN, a perception NN, recurrent NN (RNN) (e.g., including Long Short Term Memory (LSTM) algorithm, gated recurrent unit (GRU), echo state network (ESN), and the like), spiking NN (SNN), deep stacking network (DSN), Markov chain, perception NN, generative adversarial network (GAN), transformers, attention models, autoencoders, stochastic NNs (e.g., Bayesian Network (BN), Bayesian belief network (BBN), a Bayesian NN (BNN), Deep BNN (DBNN), Dynamic BN (DBN), probabilistic graphical model (PGM), Boltzmann machine, restricted Boltzmann machine (RBM), Hopfield network or Hopfield NN, convolutional deep belief network (CDBN), and the like), Linear Dynamical System (LDS), Switching LDS (SLDS),Optical NNs (ONNs), and / or the like. NNs are usually used for supervised learning, but can be used for unsupervised learning and / or reinforcement learning (RL) and / or deep RL (DRL).

[0122] The NN 1000 may encompass a variety of ML techniques where a collection of connected artificial neurons 1010 that (loosely) model neurons in a biological brain that transmit signals to other neurons / nodes 1010. The neurons 1010 may also be referred to as nodes 1010, processing elements (PEs) 1010, or the like. The connections 1020 (or edges 1020) between the nodes 1010 are (loosely) modeled on synapses of a biological brain and convey the signals between nodes 1010. Note that not all neurons 1010 and edges 1020 are labeled in Fig. 10 for the sake of clarity.

[0123] Each neuron 1010 has one or more inputs and produces an output, which can be sent to one or more other neurons 1010 (the inputs and outputs may be referred to as “signals”). Inputs to the neurons 1010 of the input layer Lxcan be feature values of a sample of external data (e.g., input variables x^. The input variables xtcan be set as a vector containing relevant data (e.g., observations, ML features, and the like). The inputs to hidden units 1010 of the hidden layers La, Lb, and Lcmay be based on the outputs of other neurons 1010. The outputs of the final output neurons 1010 of the output layer Ly(e.g., output variables y7) include predictions, inferences, and / or accomplish a desired / configured task. The output variables y7may be in the form of determinations, inferences, predictions, and / or assessments. Additionally or alternatively, the output variables y7can be set as a vector containing the relevant data (e.g., determinations, inferences, predictions, assessments, and / or the like).

[0124] In the context of ML, an “ML feature” (or simply “feature”) is an individual measurable property or characteristic of a phenomenon being observed. Features are usually represented using numbers / numerals (e.g., integers), strings, variables, ordinals, real-values, categories, and / or the like. Additionally or alternatively, ML features are individual variables, which may be independent variables, based on observable phenomenon that can be quantified and recorded. ML models use one or more features to make predictions or inferences. In some implementations, new features can be derived from old features.

[0125] Neurons 1010 may have a threshold such that a signal is sent only if the aggregate signal crosses that threshold. A node 1010 may include an activation function, which defines the output of that node 1010 given an input or set of inputs. Additionally or alternatively, a node 1010 may include a propagation function that computes the input to a neuron 1010 from the outputs of its predecessor neurons 1010 and their connections 1020 as a weighted sum. A bias term can also be added to the result of the propagation function. The NN 1000 also includes connections 1020, some of which provide the output of at least one neuron 1010 as an input to at least another neuron 1010. Each connection 1020 may be assigned a weight that represents its relative importance. Theweights may also be adjusted as learning proceeds. The weight increases or decreases the strength of the signal at a connection 1020.

[0126] The neurons 1010 can be aggregated or grouped into one or more layers L where different layers L may perform different transformations on their inputs. In Fig. 10, the NN 1000 comprises an input layer Lx, one or more hidden layers La, Lb, and Lc, and an output layer Ly(where a, b, c, x, and y may be numbers), where each layer L comprises one or more neurons 1010. Signals travel from the first layer (e.g., the input layer L- , to the last layer (e.g., the output layer Ly), possibly after traversing the hidden layers La, Lb, and Lcmultiple times. In Fig. 10, the input layer Lareceives data of input variables xt(where i = 1,where p is a number). Hidden layers La, Lb, and Lcprocesses the inputs xt, and eventually, output layer Lyprovides output variables y7(where j = 1, p', where p' is a number that is the same or different than p). In the example of Fig. 10, for simplicity of illustration, there are only three hidden layers La, Lb, and Lcin the NN 1000, however, the NN 1000 may include many more (or fewer) hidden layers La, Lb, and Lcthan are shown.

[0127] In various implementations, the NN 1000 can be implemented as an ML classifier such as an RNN, CNN, and / or the like. In these implementations, the ML classifier may classify data, such as biopotential data generated by the biopotential measurement device 400 and / or the signals derived therefrom by filtering as previously mentioned, as belonging to one of the five sleep stages (e.g., wake, Nl, N2, N3, and REM). In some examples, the classification may be done every 30 second epoch to produce a label for data in that epoch.

[0128] In one example implementation, the NN 1000 is an end-to-end hierarchical RNN adapted from Phan et al., SeqSleepNet: End-to-End Hierarchical Recurrent Neural Network for Sequence- to-Sequence Automatic Sleep Staging, IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, vol. 27, no. 3, pp. 400-410 (31 Jan. 2019), which is hereby incorporated by reference in its entirety. The layers of such an RNN can be summarized as follows: (1) a parallel filter bank designed to extract the most informative frequency sub-bands; (2) an epoch-level bidirectional attention-based RNN for sequential feature extraction and / or for shortterm (e.g., intra-epoch) sequential modelling; (3) a sequence-level bidirectional RNN for longer- term modelling of sequences of epoch-based features and / or for long-term (e.g., inter-epoch) sequential modelling; and (4) a softmax layer that predicts probabilities for each of the input classes. This approach enables an end-to-end learning strategy with a global optimization solution, while at the same time leveraging the sequential characteristics of the sleep data.

[0129] In these implementations, a sequence of epochs (e.g., 30 second epochs) are input to the end-to-end hierarchical RNN. An epoch in the input sequence includes a number of channels (e.g.,EEG, EOG, and EMG biosignal data) is transformed into corresponding time-frequency image. Each epoch (e.g., EEG, EOG, and EMG) is transformed into power spectra via STFT as discussed previously. The set of parallel filterbank layers (e.g., column-wise tied filterbank layers) are tailored to learn channel-specific frequency-domain filterbanks to preprocess the input image, such as the power spectra, for frequency smoothing and dimension reduction. After the channelspecific preprocessing, all image channels are concatenated in the frequency direction to form an image (e.g., the previously described time-frequency image X), which can be interpreted as a sequence of feature vectors. The epoch-level attention-based bidirectional RNN (e.g., including epoch-wise tied recurrent layers and epoch-wise tied attention layers) is then used to encode the feature vector sequence of the epoch into a fixed attentional feature vector. The sequence of attentional feature vectors obtained from the input epoch sequence are modelled by the sequencelevel bidirectional RNN (e.g., including sequence-wise recurrent layers) to encode long-term sequential information across epochs. A sequence of output vectors is obtained from the sequencelevel bidirectional RNN, and each output vector is presented to the softmax layer for classification. The softmax layer produces a sequence of classification outputs, where each classification output includes an output probability distribution over all sleep stages.

[0130] Other ML classifiers, such as NN-based classifiers, can be used in other implementations. Examples of such other ML classifiers can include DeepSleepNet (see Supratak et al., DeepSleepNet: A model for automatic sleep stage scoring based on raw single-channel EEG, IEEE TRANS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, vol. 25, no. 11, pp. 1998-2008 (2017)), LWSleepNet (see Yang et al., LWSleepNet: A lightweight attention-based deep learning model for sleep staging with singlechannel EEG, Digital Health SageJournals, vol. 9, January-December 2023 (27 Jul. 2023)), stacked autoencoder networks (e.g., Tsinalis et al., Automatic Sleep Stage Scoring Using Time-Frequency Analysis and Stacked Sparse Autoencoders, ANNALS OF BIOMEDICAL ENGINEERING, vol. 44, no. 5, pp. 1587-1597 (May 2016)), a CNN adapted to predict sleep staging events (e.g., Tsinalis et al., Automatic Sleep Stage Scoring with Single-Channel EEG Using Convolutional Neural Networks, arXiv: 1610.01683vl [stat.ML] (5 Oct. 2016), and Phan et al., Joint Classification and Prediction CNN Framework for Automatic Sleep Stage Classification, IEEE TRANS BIOMED ENG., vol. 66, no. 5, pp. 1285-1296 (May 2019)), among many other possibilities.5.3.3.5. Sleep Staging Model5.3.3.5.1. Transfer learning

[0131] Due to the high number of parameters to be trained and the variability in training data, building performant DNNs usually requires large amounts of data. Transfer learning solutions can help to improve the generalizability and the training efficiency of these networks. Generally,transfer learning involves transferring ML model knowledge from an already learned source task to the learning of a related target task. See Handbook of Research on Machine Learning Applications and Trends, IGI GLOBAL (Olivas ES, Guerrero JDM, Martinez-Sober M, Magdalena- Benedito JR, Serrano Lopez AJ, eds., 2010). To this end, a base ML model is first trained on a large source dataset. The trained model parameters then serve as starting point for further training and finetuning using the target dataset.

[0132] In some implementations, the Sleep Heart Health Study (SHHS), a multi-centric cohort study to investigate cardiovascular and other effects of sleep-disordered breathing, is adopted as the source dataset. See Quan et al., The Sleep Heart Health Study: design, rationale, and methods, SLEEP, vol. 20, no. 12, pp. 1077-1085 (01 Dec. 1997), and Zhang et al., The National Sleep Research Resource: towards a sleep data commons, JOURNAL OF THE AMERICAN MEDICAL INFORMATICS ASSOCIATION, vol. 25 issue 10, pp. 1351-1358 (oct. 2018). The dataset includes labelled PSG data of 6441 individuals, including C3 / A2 and C4 / A1 EEGs at 125 Hz. A transfer learning base model using the described NN architecture is trained on the C3 / A2 channel of 200 patients. Base model training input data may be extracted by transforming the EEG channel of each patient into its time-frequency spectrum using the SSM settings and dividing the obtained power spectrum into 30 second epochs.5.3.3.5.2. Training and Cross-Validation

[0133] The NN 1000 may be adapted from a deep learning framework that learns from biopotential data, such as data generated by the biopotential measurement device 400 as previously described, that is fed into it. In these implementations, the NN 1000 may be trained on training data of EEG data obtained from public databases. Additionally or alternatively, the NN 1000 may be trained on EEG, EOG, and / or EMG data captured by other sleep study devices such as, for example, data captured using one or more of the other sleep study devices mentioned previously (see e.g., section 5.3.3.1, supra). In either case, training data points may be selected from datasets collected using sleep study devices that include leads or electrodes placed in similar locations as the locations where the electrodes 410, 420 are placed (see e.g., Figs. 3, 4, and 5).

[0134] The training process can also involve the ML model, such as NN 1000, learning various features from the input data to make predictions or inferences, such as classifications. The classifications or classes may include the five sleep stages (e.g., wake / active, Nl, N2, N3, and REM). The specific features that are learned or derived from the learning network may depend on the type of classifier used (e.g., linear classifiers, decision trees, RNNs, and / or the like) and the nature of the dataset. Typically, a classifier NN 1000 may learn hierarchical features by stacking multiple layers of neurons 1010 where lower layers typically learn low-level features (e.g., edges, textures), while higher layers learn more abstract and complex features (e.g., shapes, patterns).

[0135] Additionally, the base ML model may be further finetuned using a target dataset patients that have used the biopotential measurement device 400. As alluded to previously, the pre- processed EEG channel of each patient is transformed into its power spectrum and divided in 30 second epochs. The power spectrum entries of rejected epochs are set to 0 and the corresponding labels are set to ‘Invalid’. By introducing this additional label, the model automatically learns to deal with rejected epochs while still preserving the sequential nature and time-dependencies of the data.

[0136] To maximize utilization of available data, training and evaluation are performed via a patient-wise 10-fold cross-validation. For each fold, the base model may be adapted using 90% of the available patients and is sub sequentially evaluated on the remaining 10%. This process may be repeated a number of times (e.g., 10 times) to obtain the full output for each patient independently. This final set of validation outcomes is then used to calculate the endpoint parameters.5.3.4. Example Methods

[0137] As has been shown in the previous examples, sleep staging events can be detected using a single-channel biopotential measurement device, such as the biopotential measurement device 400, and an ML classifier can be used to classify different segments of identified biosignals as belonging to different sleep stages. In this regard, such methodologies of the present technology as previously described may be further considered in relation to the methods depicted by Fig. 11. In Fig. 11, the process 1100 may begin at operation 1110, where one or more processors may obtain a facial biopotential signal measured between two electrodes connected to a user’s face, such as when positioned in the transverse-ocular measurement vector as previously discussed. The facial biopotential signal may be measured by a biopotential measurement device 400, for example. At operation 1120, the one or more processors may derive, based on the facial biopotential signal, a plurality of biosignals suggestive of sleep staging events. The plurality of biosignals may include any or all of electroencephalography (EEG) signals, electrooculography (EOG) signals, and / or electromyography (EMG) signals. At operation 1130, the one or more processors may segment each of the plurality of biosignals into a plurality of epochs. At operation 1140, the one or more processors may feed the plurality of epochs to a sleep staging model (SSM). The SSM may be a ML classifier, such as a hierarchical RNN as discussed previously. At operation 1150, the one or more processors may operate the SSM to predict a classification label for each epoch of the plurality of epochs. Each classification label may correspond to one or more sleep staging events. Optionally, at operation 1160, one or more processors of the system may generate an output with, or based on, the predicted classification labels of the plurality of epochs. Such an output may include, for example, a signal. Such a signal may include text or graphical outputs tobe displayed by a display device. Such an output may include, for example, a signal, such as control signal, with a setting or for setting (e.g., a pressure or flow rate setting) of an operation of a respiratory therapy. Such as signal may optionally be communicated to a respiratory therapy device from the one or more processors, such via a communications network (e.g., an internet) and / or other intermediary device(s) (e.g., one or more servers).5.4. GLOSSARY

[0138] For the purposes of the present disclosure, in certain forms of the present technology, one or more of the following definitions may apply. In other forms of the present technology, alternative definitions may apply.5.4.1. General

[0139] Patient'. A person, whether or not they undergo a sleep study or whether or not they have a sleep disorder.

[0140] Sleep Stages'. Sleep occurs in five stages, including wake / alert, Nl, N2, N3, and rapid eye movement (REM). Stages Nl to N3 are considered non-rapid eye movement (NREM) sleep, with each stage leading to progressively deeper sleep. Sleep staging events may refer to events related to one or more sleep stages and / or features within or derived from biosignals and / or biopotential signals that are indicative of one or more sleep stages.

[0141] Wake / alert stage', a sleep stage where EEG signals include beta waves, which have the highest frequency and lowest amplitude among the five sleep stages. Alpha waves can be seen during quiet / relaxed wakefulness.

[0142] Nl (stage 1) '. a sleep stage representing light sleep, where EEG signals include theta waves and a low voltage. Nl is the lightest stage of sleep and begins when more than 50% of the alpha waves are replaced with low-amplitude mixed-frequency (LAMF) activity. Muscle tone is present in the skeletal muscle, and breathing occurs regularly. This stage lasts around 1 to 5 minutes, comprising about 5% of total sleep time.

[0143] N2 (stage 2)'. a sleep stage representing deeper sleep than Nl, where EEG signals include sleep spindles and K complexes. N2 stage represents deeper sleep as the heart rate and body temperature drop. The presence of sleep spindles, K-complexes, or both characterizes it. Sleep spindles are brief, powerful bursts of neuronal firing in the superior temporal gyri, anterior cingulate, insular cortices, and thalamus, inducing calcium influx into cortical pyramidal cells. This mechanism is believed to be integral to synaptic plasticity. Numerous studies suggest that sleep spindles are essential in memory consolidation, specifically procedural and declarative memory. K-complexes are long delta waves that last approximately one second and are known to be the longest and most distinct of all brain waves. K-complexes are shown to function in maintaining sleep and memory consolidation. Stage 2 sleep lasts around 25 minutes in the firstcycle and lengthens with each successive cycle, eventually comprising about 45% of total sleep. This stage of sleep is when bruxism (teeth grinding) occurs.

[0144] N3 (stage 3): a sleep stage representing the deepest non-REM sleep, where EEG signals include delta waves, which have the lowest frequency and highest amplitude among the five sleep stages. N3 is also known as slow-wave sleep (SWS). This is considered the deepest stage of sleep and is characterized by signals with lower frequencies and higher amplitudes, known as delta waves. This stage is the most difficult to awaken from; for some people, loud noises (> 100 decibels) will not lead to an awake state. As people age, they spend less time in this slow, deltawave sleep and more time in stage N2 sleep. Although this stage has the greatest arousal threshold, if someone is awoken during this stage, they will have a transient phase of mental fogginess, known as sleep inertia. Cognitive testing shows that individuals awakened during this stage tend to have moderately impaired mental performance for 30 minutes to 1 hour. This is the stage when the body repairs and regrows tissues, builds bone and muscle, and strengthens the immune system. This is also the stage when sleepwalking, night terrors, and bedwetting occur.

[0145] Biosignal-. any signal in living beings that can be measured and monitored, and can include bioelectrical signals and / or non-electrical signals. Examples of biosignals include electroencephalogram (EEG), electrocardiogram (ECG), electromyogram (EMG), electrooculogram (EOG), electroretinogram (ERG), electrogastrogram (EGG), galvanic skin response (GSR) and / or electrodermal activity (EDA), photoplethysmography (PPG), peripheral arterial tone, and biopotential.

[0146] Biopotential: electrical signals (voltages) that are generated by physiological processes occurring within the body. A biopotential can also be considered as the difference of potentials between two points of tissue, which reflects its bioelectric activity. Sources of biopotential signals include action potentials generated by excitable cells, corneal-retinal (comeoretinal) potential, skin potential, electromechanical behavior of bone, electrodermal activity (EDA), cardiac impedance (stroke volume), thoracic impedance (respiration), optical blood volume density (PPG or BVP), and blood oxygen level (SpO2).

[0147] Apnea. Individuals with sleep apnea typically experience airway collapse in deeper sleep states, causing them to experience reduced time in stage N3 and REM sleep, leading to excessive daytime drowsiness. There are two types of sleep apnea: central and obstructive. Central sleep apnea occurs when the brain fails to signal respiratory muscles during sleep. Obstructive sleep apnea is a mechanical problem in which there is a partial or complete blockage of the upper airway. According to some definitions, an apnea is said to have occurred when respiratory flow rate falls below a predetermined threshold for a duration, for example, 10 seconds. An obstructive apnea will be said to have occurred when, despite patient effort, some obstruction of the airway does notallow air to flow. A central apnea will be said to have occurred when an apnea is detected that is due to a reduction in breathing effort, or the absence of breathing effort.

[0148] REM Sleep Disorder', a sleep disorder taking place during the REM stage where the temporary atonia of REM sleep is disturbed. Atonia refers to temporary muscle paralysis during REM sleep.5.4.2. Terms for Data Analytics and Machine Learning (ML)

[0149] Classifier', any system or methodology that performs classification, wherein classification involves categorizing or assigning objects or data points to specific classes or categories, which may sometimes be called “targets” or “labels.” Classifiers may include ML classifiers or non-ML classifiers. Non-ML classifiers typically use predefined rules, heuristics, or algorithms to perform classification. Examples of non-ML classifiers include rules-based classifiers, heuristics classifiers, pattern matching classifiers, knowledge-based classifiers, and hand-crafted or manually designed classifiers. ML classifiers typically learn patterns from data. More specifically, an ML classifier is an ML algorithm or ML model that takes input data, which can be in the form of features or attributes, and assigns each data point to one of several classes or categories. The classes or categories may be predefined or learned. The process of classification involves training the classifier on labeled training data, where each data point is associated with a known class label. During training, the classifier learns to identify patterns, trends, and decision boundaries in the input data that differentiate between different classes. Once trained, the classifier can then be used to predict the class labels of new, unseen data points (e.g., referred to as an “inference dataset” or the like) based on the learned patterns and relationships. Examples of ML classifiers include linear classifiers such as logistic regression and perceptrons, k-nearest neighbor (kNN), decision trees, random forests, support vector machines (SVMs), Bayesian classifiers, CNNs, RNNs, among many others (note that some of these algorithms can be used for other ML tasks as well).

[0150] Epoch'. One cycle through a full training dataset during an ML training process. An epoch may also be a full training pass over an entire training dataset such that each training example has been seen once. In some examples, an epoch represents N / batch size training iterations, where N is the total number of examples.

[0151] Feature'. A measurable and / or quantifiable property, and / or a characteristic of a phenomenon being observed. Additionally or alternatively, a feature may refer to an input variable used in making predictions / inferences. Features may be represented using numbers / numerals (e.g., integers), strings, variables, ordinals, real-values, categories, and / or the like.

[0152] Filterbank', a layer in a neural network that applies a set of filters or kernels to input data, usually in parallel. Each filter captures different aspects or features of the input, resulting in a multi-channel output. The filters in a filterbank layer can have different sizes, shapes, andproperties, allowing the network to capture diverse features at different spatial or frequency scales.

[0153] Hyperparameter-. Characteristics, properties, and / or parameters for an ML process that are not learned during a training process. Hyperparameters are usually set before training takes place, and may be used in processes to help estimate model parameters. Examples of hyperparameters include model size (e.g., in terms of memory space, bytes, number of layers, and the like); training data shuffling (e.g., whether to do so and by how much); number of evaluation instances, iterations, epochs (e.g., a number of iterations or passes over the training data), or episodes; number of passes over training data; regularization; learning rate (e.g., the speed at which the algorithm reaches (converges to) optimal weights); learning rate decay (or weight decay); momentum; number of hidden layers; size of individual hidden layers; weight initialization scheme; dropout and gradient clipping thresholds; the C value and sigma value for SVMs; the k in k-nearest neighbors; number of branches in a decision tree; number of clusters in a clustering algorithm; vector size; word vector size for NLP and NLU; and / or the like.

[0154] Model parameter-. Values, characteristics, and / or properties that are learnt during ML training. A model parameter may be a configuration variable that is internal to the model and whose value can be estimated from the given data. Model parameters are usually required by a model when making predictions or inferences, and their values define the skill of the model on a particular problem. Examples of such model parameters include weights, biases, constraints, classes or categories, support vectors in an SVM, coefficients in a linear regression and / or logistic regression, and / or the like.

[0155] Inference '. The process of utilizing a trained ML model to extract meaningful insights, make predictions, and / or take actions based on new data. For example, inference generation in supervised learning involves making predictions based on labeled data, inference generation in unsupervised learning involves uncovering patterns in unlabeled data, and inference generation in reinforcement learning involves decision-making based on learned policies and environmental feedback. For purposes of the present disclosure, the term “inference”, “inference generation”, or the like, refers to the process of using trained ML model(s) to generate statistical inferences, predictions, decisions, probabilities, probability distributions, actions, configurations, policies, data analytics, outcomes, optimizations, and / or the like based on new, unseen data (e.g., “input inference data”).

[0156] Iteration'. The repetition of a process in order to generate a sequence of outcomes, wherein each repetition of the process is a single iteration, and the outcome of each iteration is the starting point of the next iteration. Additionally or alternatively, an iteration may refer to a single update of a model’s weights during training.

[0157] Softmax'. a function that converts a vector of numerical values into a probabilitydistribution. The softmax function is often used in ML models, such as those involving classification tasks where the model needs to output probabilities for multiple classes. The softmax function may be used as the last activation function of a neural network, to normalize the output of the ML model to a probability distribution over predicted output classes.5.5. OTHER REMARKS

[0158] A portion of the disclosure of this patent document contains material which is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent document or the patent disclosure, as it appears in the Patent and Trademark Office patent file or records, but otherwise reserves all copyright rights whatsoever.

[0159] Unless the context clearly dictates otherwise and where a range of values is provided, it is understood that each intervening value, to the tenth of the unit of the lower limit, between the upper and lower limit of that range, and any other stated or intervening value in that stated range is encompassed within the technology. The upper and lower limits of these intervening ranges, which may be independently included in the intervening ranges, are also encompassed within the technology, subject to any specifically excluded limit in the stated range. Where the stated range includes one or both of the limits, ranges excluding either or both of those included limits are also included in the technology.

[0160] Furthermore, where a value or values are stated herein as being implemented as part of the technology, it is understood that such values may be approximated, unless otherwise stated, and such values may be utilized to any suitable significant digit to the extent that a practical technical implementation may permit or require it.

[0161] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this technology belongs. Although any methods and materials similar or equivalent to those described herein can also be used in the practice or testing of the present technology, a limited number of the exemplary methods and materials are described herein.

[0162] When a particular material is identified as being preferably used to construct a component, obvious alternative materials with similar properties may be used as a substitute. Furthermore, unless specified to the contrary, any and all components herein described are understood to be capable of being manufactured and, as such, may be manufactured together or separately.

[0163] It must be noted that as used herein and in the appended claims, the singular forms "a", "an", and "the" include their plural equivalents, unless the context clearly dictates otherwise.

[0164] All publications mentioned herein are incorporated by reference in their entireties to disclose and describe the methods and / or materials which are the subject of those publications. The publications discussed herein are provided solely for their disclosure prior to the filing date ofthe present application. Nothing herein is to be construed as an admission that the present technology is not entitled to antedate such publication by virtue of prior invention. Further, the dates of publication provided may be different from the actual publication dates, which may need to be independently confirmed.

[0165] Moreover, in interpreting the disclosure, all terms should be interpreted in the broadest reasonable manner consistent with the context. In particular, the terms "comprises" and "comprising" should be interpreted as referring to elements, components, or steps in a nonexclusive manner, indicating that the referenced elements, components, or steps may be present, or utilized, or combined with other elements, components, or steps that are not expressly referenced.

[0166] The subject headings used in the detailed description are included only for the ease of reference of the reader and should not be used to limit the subject matter found throughout the disclosure or the claims. The subject headings should not be used in construing the scope of the claims or the claim limitations.

[0167] Although the technology herein has been described with reference to particular embodiments, it is to be understood that these embodiments are merely illustrative of the principles and applications of the technology. In some instances, the terminology and symbols may imply specific details that are not required to practice the technology. For example, although the terms "first" and "second" may be used, unless otherwise specified, they are not intended to indicate any order but may be utilised to distinguish between distinct elements. Furthermore, although process steps in the methodologies may be described or illustrated in an order, such an ordering is not required. Those skilled in the art will recognize that such ordering may be modified and / or aspects thereof may be conducted concurrently or even synchronously.

[0168] It is therefore to be understood that numerous modifications may be made to the illustrative embodiments and that other arrangements may be devised without departing from the spirit and scope of the technology.| 1691 Although the present invention has been illustrated by reference to specific embodiments, it will be apparent to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present technology may be embodied with various changes and modifications without departing from the scope thereof. The present examples are therefore to be considered in all respects as illustrative and not restrictive, the scope of the technology being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. In other words, it is contemplated to cover any and all modifications, variations or equivalents that fall within the scope of the basic underlying principlesand whose essential attributes are claimed in this patent application. It will furthermore be understood by the reader of this patent application that the words "comprising" or "comprise" do not exclude other elements or steps, that the words "a" or "an" do not exclude a plurality, and that a single element, such as a computer system, a processor, or another integrated unit may fulfil the functions of several means recited in the claims. Any reference signs in the claims shall not be construed as limiting the respective claims concerned. The terms "first", "second", third", "a", "b", "c", and the like, when used in the description or in the claims are introduced to distinguish between similar elements or steps and are not necessarily describing a sequential or chronological order. Similarly, the terms "top", "bottom", "over", "under", and the like are introduced for descriptive purposes and not necessarily to denote relative positions. It is to be understood that the terms so used are interchangeable under appropriate circumstances and embodiments of the technology are capable of operating according to the present technology in other sequences, or in orientations different from the one(s) described or illustrated above.

Claims

CLAIMS:

1. A processor-implemented method for detecting sleep staging events, the method comprising: obtaining a facial biopotential signal measured between two electrodes connected to a user’s face; deriving, from the facial biopotential signal, a plurality of biosignals suggestive of sleep staging events; and classifying individual segments of the plurality of biosignals as belonging to one of a plurality of sleep staging events.

2. The method of claim 1, wherein the two electrodes form a transverse-ocular measurement vector when the facial biopotential signal is measured between the two electrodes.

3. The method of any one of claims 1 to 2, wherein the plurality of biosignals is derived by filtering.

4. The method of claim 3, wherein the filtering produces electroencephalography (EEG) data, electrooculography (EOG) data and electromyography (EMG) data.

5. The method of claim 3 or 4, wherein the filtering includes applying frequency-based filtering to the facial biopotential signal.

6. The method of claim 5, wherein the applying frequency-based filtering to the facial biopotential signal comprises any one or more of: filtering a frequency range between 0.5 Hertz (Hz) and 35 Hz from the facial biopotential signal to produce the EEG data; filtering a frequency range between 0.1 Hz and 35 Hz from the facial biopotential signal to produce the EOG data; and filtering a frequency range between 70 Hz and 110 Hz from the facial biopotential signal to produce the EMG data.

7. The method of any one of claims 1 to 6, wherein the facial biopotential signal is measured based on a voltage difference between the two electrodes.

8. The method of any one of claims 1 to 7, wherein the facial biopotential signal is measured based on a current between the two electrodes.- 47 -9. The method of any one of claims 1 to 8, wherein the obtaining includes: obtaining the facial biopotential signal from a biopotential measurement device comprising the two electrodes.

10. The method of claim 9, wherein the biopotential measurement device is a single-channel biopotential measurement device comprising two electrodes.

11. The method of any one of claims 9 to 10, wherein the biopotential measurement device is a single-channel biopotential measurement device comprising only two electrodes.

12. The method of any one of claims 1 to 11, wherein the two electrodes are skin contact electrodes including respective adhesive patches.

13. The method of any one of claims 1 to 12, wherein the two electrodes include a sub-ocular facial electrode and a supra-ocular facial electrode.

14. The method of claim 13, wherein the sub-ocular facial electrode is a sagittal -right facial electrode or a sagittal-left facial electrode.

15. The method of any one of claims 13 to 14, wherein the supra-ocular facial electrode is a sagittal -neutral facial electrode.

16. The method of any one of claims 13 to 15, wherein the sub-ocular facial electrode is a measurement electrode and the supra-ocular facial electrode is a reference electrode.

17. The method of claim 16, wherein the reference electrode is configured to be attached to a center of the user’s forehead, and the measurement electrode is configured to be attached sub-ocular to an eye of the user.

18. The method of any one of claims 13 to 15, wherein the supra-ocular facial electrode is a measurement electrode and the sub-ocular facial electrode is a reference electrode.

19. The method of claim 18, wherein the measurement electrode is configured to be attached to a center of the user’s forehead, and the reference electrode is configured to be attached below the user’s eye.- 48 -20. The method of any one of claims 13 to 19, wherein the reference electrode is a ground electrode.

21. The method of any one of claims 13 to 20, wherein the measurement electrode is connected to the reference electrode via an electric wire.

22. The method of any one of claims 1 to 21, wherein the biopotential measurement device further comprises a base, and the base comprises a controller with at least one processor configured to measure the biopotential signal.

23. The method of claim 22, wherein the base is directly connected to the reference electrode.

24. The method of any one of claims 22 to 23, wherein the base further comprises a communication interface connected to the at least one processor.

25. The method of claim 24, wherein the method is performed by the controller in the base, and the method further comprises: causing transmission, using the communication interface, of any one or more of: (a) the classified individual segments, (b) the measured facial biopotential signal, and (c) the derived plurality of biosignals, to an external computing system over a wired or wireless connection between the communication interface and the external computing system.

26. The method of claim 24, wherein the method is performed by a computing system external to the biopotential measurement device, and the obtaining comprises: receiving the biopotential signal from the biopotential measurement device over a wired or wireless connection with the biopotential measurement device, wherein the biopotential signal is transmitted by the biopotential measurement device using the communication interface.

27. The method of any one of claims 1 to 26, wherein the method further comprises: segmenting each of the plurality of biosignals into the individual segments, each individual segment comprising one of a plurality of epochs.

28. The method of claim 27, wherein the classifying comprises:- 49 -feeding the plurality of epochs to a sleep staging model (SSM) to predict a classification label for each epoch of the plurality of epochs, wherein each classification label corresponds to one or more sleep staging events.

29. The method of claim 28, wherein the SSM is a trained machine learning (ML) classifier model.

30. The method of claim 29, wherein the ML classifier model is a recurrent neural network (RNN).

31. The method of any one of claims 28 to 30, wherein the SSM comprises: a filterbank configured to extract frequency sub-bands; an epoch-level bidirectional attention-based SSM configured to perform sequential feature extraction based on the extracted frequency sub-bands; a sequence-level bidirectional SSM configured to perform modelling of sequential features extracted by the epoch-level bidirectional attention-based SSM; and a softmax layer configured to predict probabilities for each classification label.

32. The method of any one of claims 28 to 31, wherein the method further comprises: transforming each epoch of the plurality of epochs into a time-frequency image via short-time Fourier transform.

33. The method of any one of claims 29 to 32, wherein the method further comprises: outputting the predicted classification labels for each epoch.

34. A controller comprising at least one processor and at least one memory including processor control instructions, the at least one memory and processor control instructions configured to, with the at least one processor, cause the controller to perform a method of any one of claims 1 to 33.

35. A processor-readable storage medium comprising processor-executable instructions, wherein execution of the processor-executable instructions by one or more processors of a computing system is to cause the computing system to perform the method of any one of claims 1 to 33.

36. A biopotential measurement device, comprising: a reference electrode; a measurement electrode;- 50 -a positioning structure connecting the reference electrode and the measurement electrode, the positioning structure configured to, when in use on a patient's face, enable positioning of the reference electrode with respect to the measurement electrode so that the electrodes form a transverse-ocular measurement vector; and a base comprising at least one processor configured to: measure a biopotential between the measurement electrode and the reference electrode, and generate a biopotential signal based on the measured biopotential.

37. The biopotential measurement device of claim 36, wherein the positioning structure comprises an electric wire that electrically couples one of the reference electrode and the measurement electrode to the base.

38. The biopotential measurement device of any one of claims 36 to 37, wherein the positioning structure is configured to connect the reference electrode and the measurement electrode so that, when in use on a patient's face, one of the reference electrode and the measurement electrode is positioned in a central forehead location of the patient's head, and the other one of the reference electrode and the measurement electrode is positioned on a cheek location of the patient's face.

39. The device of any one of claims 36 to 38, wherein the facial biopotential signal is measured based on a voltage difference between the reference electrode and the measurement electrode.

40. The device of any one of claims 36 to 39, wherein the facial biopotential signal is measured based on a current between the reference electrode and the measurement electrode.

41. The device of any one of claims 36 to 40, wherein the reference electrode and the measurement electrode are skin contact electrodes including respective adhesive patches.

42. The device of any one of claims 36 to 41, wherein the two electrodes include a sub-ocular facial electrode and a supra-ocular facial electrode.

43. The device of claim 42, wherein the sub-ocular facial electrode is a sagittal -right facial electrode or a sagittal-left facial electrode.

44. The device of any one of claims 42 to 43, wherein the supra-ocular facial electrode is a sagittal- neutral facial electrode.

45. The device of any one of claims 42 to 44, wherein the sub-ocular facial electrode is a measurement electrode and the supra-ocular facial electrode is a reference electrode.

46. The device of claim 45, wherein the reference electrode is configured to be attached to a center of the user’s forehead, and the measurement electrode is configured to be attached sub-ocular to an eye of the user.

47. The device of any one of claims 42 to 44, wherein the supra-ocular facial electrode is a measurement electrode and the sub-ocular facial electrode is a reference electrode.

48. The device of claim 47, wherein the measurement electrode is configured to be attached to a center of the user’s forehead, and the reference electrode is configured to be attached below the user’s eye.

49. The device of any one of claims 36 to 48, wherein the reference electrode is a ground electrode.

50. The device of any one of claims 36 to 49, wherein the base comprises the reference electrode.

51. The device of any one of claims 36 to 50, wherein the base further comprises a communication interface connected to the at least one processor.

52. The device of claim 51, wherein the at least one processor is configured to: cause the communication interface to transmit the biopotential signal to an external computing system over a wired or wireless connection between the communication interface and the external computing system.

53. The device of claim 52, wherein the at least one processor is configured to: derive, from the biopotential signal, a plurality of biosignals indicative of sleep staging events; and classify individual segments of the plurality of biosignals as belonging to one of a plurality of sleep staging events.

54. The device of claim 53, wherein the plurality of biosignals comprises electroencephalography (EEG) signals, electrooculography (EOG) signals, and electromyography (EMG) signals.

55. The device of any one of claims 36 to 54, wherein the at least one processor is configured to: cause the communication interface to transmit any one or more of: (a) classified individual segments of a plurality of biosignals derived from the measured facial biopotential signal, (b) the measured facial biopotential signal, and (c) the derived plurality of biosignals, to an external computing system over a wired or wireless connection between the communication interface and the external computing system.

56. The device of any one of claims 53 to 55, wherein the at least one processor is configured to: segment each of the plurality of biosignals into a plurality of epochs.

57. The device of claim 56, wherein the at least one processor is configured to: operate a sleep staging model (SSM) to predict a classification label for each epoch of the plurality of epochs, wherein each classification label corresponds to one or more sleep staging events.

58. The device of claim 57, wherein the SSM is a trained recurrent neural network (RNN) classifier model.

59. The device of claim 57 or 58, wherein the SSM classifier model comprises: a filterbank configured to extract frequency sub-bands; an epoch-level bidirectional attention-based SSM configured to perform sequential feature extraction based on the extracted frequency sub-bands; a sequence-level bidirectional SSM configured to perform modelling of sequential features extracted by the epoch-level bidirectional attention-based SSM; and a softmax layer configured to predict probabilities for each classification label.

60. The device of claim 59, wherein the filterbank is configured to apply frequency -based filtering to the facial biopotential signal, wherein the frequency sub-bands comprise any one or more of: a frequency range between 0.5 Hertz (Hz) and 35 Hz from the facial biopotential signal to produce the EEG data; a frequency range between 0.1 Hz and 35 Hz from the facial biopotential signal to produce the EOG data; and- 53 -a frequency range between 70 Hz and 110 Hz from the facial biopotential signal to produce the EMG data.

61. The device of any one of claims 36 to 58, further comprising a filterbank that is configured to apply frequency-based filtering to the facial biopotential signal to derive one or more biosignals, wherein one or more frequency bands of the filterbank comprise any one or more of: a frequency range between 0.5 Hertz (Hz) and 35 Hz from the facial biopotential signal to produce the EEG data; a frequency range between 0.1 Hz and 35 Hz from the facial biopotential signal to produce the EOG data; and a frequency range between 70 Hz and 110 Hz from the facial biopotential signal to produce the EMG data.

62. The device of any one of claims 56 to 61, wherein the at least one processor is configured to: transform each epoch of the plurality of epochs into a time-frequency image via short-time Fourier transform.

63. The device of any one of claims 36 to 62, further comprising a microphone, and wherein the at least one processor is configured to detect snoring sounds.

64. The device of any one of claims 36 to 63, further comprising an accelerometer, and wherein the at least one processor is configured to detect body position and / or user motion.

65. A computing system, comprising: a communication interface configured to receive, over a wired or wireless connection, a facial biopotential signal from a biopotential measurement device comprising two electrodes, wherein the facial biopotential signal is measured between the two electrodes connected to a user’s face and configured in a transverse-ocular measurement vector; one or more memories; and one or more processors connected to the one or more memories, wherein the one or more processors are configured to: obtain the facial biopotential signal; derive, from the facial biopotential signal, a plurality of biosignals suggestive of sleep staging events; and- 54 -classify individual segments of the plurality of biosignals as belonging to one of a plurality of sleep stating events.

66. The computing system of claim 65, wherein the facial biopotential signal is measured based on a voltage difference between the two electrodes or based on a current between the two electrodes.

67. The computing system of any one of claims 67 to 68, wherein the two electrodes include a reference electrode and a measurement electrode.

68. The computing system of claim 69, wherein the measurement electrode is a sub-ocular facial electrode and one of a sagittal-right facial electrode or a sagittal-left facial electrode.

69. The computing system of any one of claims 69 to 70, wherein the reference electrode is a supra-ocular facial electrode and a sagittal-neutral facial electrode.

70. The computing system of any one of claims 69 to 71, wherein the reference electrode is configured to be attached to a center of the user’s forehead, and the measurement electrode is configured to be attached to a portion of the user’s face underneath the user’s eye.

71. The computing system of claim 69, wherein the reference electrode is a sub-ocular facial electrode and one of a sagittal-right facial electrode or a sagittal-left facial electrode.

72. The computing system of claim 73, wherein the measurement electrode is a supra-ocular facial electrode and a sagittal-neutral facial electrode.

73. The computing system of any one of claims 73 to 74, wherein the measurement electrode is configured to be attached to a center of the user’s forehead, and the reference electrode is configured to be attached to a portion of the user’s face underneath the user’s eye.

74. The computing system of any one of claims 69 to 75, wherein the reference electrode is a ground electrode.

75. The computing system of any one of claims 69 to 76, wherein the measurement electrode is connected to the reference electrode via an electric wire.- 55 -76. The computing system of any one of claims 67 to 77, wherein the plurality of biosignals comprises electroencephalography (EEG) signals, electrooculography (EOG) signals, and electromyography (EMG) signals.

77. The computing system of any one of claims 67 to 78, wherein the one or more processors are configured to operate software to: segment each of the plurality of biosignals into a plurality of epochs.

78. The computing system of claim 79, wherein the classifying comprises: operate a sleep staging model (SSM) to predict a classification label for each epoch of the plurality of epochs, wherein each classification label corresponds to one or more sleep staging events.

79. The computing system of claim 80, wherein the SSM is a trained recurrent neural network (RNN) classifier model comprising: a filterbank configured to extract frequency sub-bands; an epoch-level bidirectional attention-based RNN configured to perform sequential feature extraction based on the extracted frequency sub-bands; a sequence-level bidirectional RNN configured to perform modelling of sequential features extracted by the epoch-level bidirectional attention-based RNN; and a softmax layer configured to predict probabilities for each classification label.

80. The computing system of any one of claims 80 to 81, wherein the one or more processors are configured to operate with software to: transform each epoch of the plurality of epochs into a time-frequency image via short-time Fourier transform.

81. The computing system of any one of claims 80 to 82, wherein the one or more processors are configured to operate with software to: output the predicted classification labels for each epoch.

82. The computing system of any one of claims 67 to 83, wherein the computing system is a server or a mobile device.- 56 -83. The computing system of any one of claims 67 to 82, wherein the one or more processors comprise a filterbank configured to apply frequency -based filtering to the facial biopotential signal to derive one or more biosignals, wherein one or more frequency bands of the filterbank comprise any one or more of: a frequency range between 0.5 Hertz (Hz) and 35 Hz from the facial biopotential signal to produce the EEG data; a frequency range between 0.1 Hz and 35 Hz from the facial biopotential signal to produce the EOG data; and a frequency range between 70 Hz and 110 Hz from the facial biopotential signal to produce the EMG data.

Images