Closed-loop noninvasive brain stimulation with simultaneous imaging to control cerebrospinal fluid flow in humans

Abstract

A method and device for stimulating cerebrospinal fluid flow is disclosed. The method may include acquiring an electroencephalography (EEG) signal from a human subject, predicting a target feature of the EEG based on the EEG signal, and playing an auditory stimulus to coincide with the target feature of the EEG learned by a neural network. The target feature may include a peak of a slow wave. The method may include using a recurrent neural network to predict the target feature for a human subject.

Description

CLOSED-LOOP NONINVASIVE BRAIN STIMULATION WITH SIMULTANEOUSIMAGING TO CONTROL CEREBROSPINAL FLUID FLOW IN HUMANSCROSS-REFERENCE TO RELATED APPLICATION(S)

[0001] This application claims priority to U.S. Application No. 63 / 559,435, filed February 29, 2024, which is incorporated herein by reference in its entirety for all purposes.GOVERNMENT SUPPORT

[0002] This invention was made with government support under AG070135 awarded by the National Institutes of Health. The government has certain rights in the invention.BACKGROUND

[0003] Humans spend roughly one third of their lives sleeping, and adequate sleep is critical for health and wellness. Sleep also plays a key role in learning and memory consolidation. People who have disrupted sleep perform less well on tests designed to measure memory. Disrupted sleep is linked to worse memory, and sleep loss is a predictor of future onset of neurodegenerative disorders such as Alzheimer's disease. Meanwhile interventions that improve sleep may also improve recall. Sleep helps to preserve brain health and function.

[0004] Sleep quality is often measured clinically using an electroencephalogram (EEG), a tool which non-invasively measures electrical activity in the brain. Non-rapid eye movement (NREM) sleep is a phase of sleep where the eyes do not move rapidly and may be considered a restful sleep stage. One feature of NREM sleep is the appearance of slow waves in the EEG, marked by episodes of high amplitude (approximately > 50 pV, for example, > 75 pV), low frequency (approximately 1 Hz) oscillations in electrical activity. Slow waves have a particular importance for memory, as research has shown that memory re-activation during NREM sleep may have a beneficial effect on recall. Additionally, interventions that increase the number of slow waves may have a beneficial effect on memory.

[0005] One mechanism by which sleep is thought to preserve brain health is by modulating waste transport out of the brain. Cerebrospinal fluid (CSF) is a liquid that surrounds the brain and is the main transport system for waste clearance out of the brain. During sleep, large waves of CSF flow appear. Diminished CSF flow impedes waste clearance and may contribute to pathological buildup of proteins in the brain in diseases such as Alzheimer’s Disease.

[0006] A challenge in developing ways to enhance CSF flow in humans is that CSF flow has been technically difficult to measure, particularly when trying to also measure EEG. Measuring CSF flow requires the use of functional magnetic resonance imaging (fMRI), in addition to the EEG needed to monitor sleep in a patient. Thus, measuring CSF flow is expensive, difficult to access due to the requirements of fMRI, and may complicate the experimental design particularly when also trying to measure EEG.

[0007] Simultaneous EEG-fMRI is a powerful tool for studying the brain, as it offers both the high temporal resolution of EEG and the high spatial resolution of fMRI. However, its use has been limited due to the technically challenging nature of this kind of data collection and processing. Particularly, fMRI causes persistent high-amplitude artifacts in EEG signals. These artifacts must be removed before the EEG data can be useful. While techniques for removing these artifacts have existed for many years, until recently, there have not been any artifact removal techniques with both low latency and highly effective. This has been a substantial barrier to EEG-fMRI experiments that demanded low latency, such as EEG neurofeedback experiments which must have fast access to cleaned EEG data in order to control stimulus presentation. More recently, low-latency methods for EEG-fMRI artifact removal have been developed, such as EEG-LLAMAS (Low Latency Artifact Mitigation Acquisition Software), which have made these neurofeedback experiments more practical.SUMMARY

[0008] In some aspects, the techniques described herein relate to a method of stimulating cerebrospinal fluid flow, the method including acquiring an electroencephalography (EEG) signal from a human subject while the human subject is asleep, predicting a target feature of the EEG signal, and delivering a stimulus to the human subject to coincide with the target feature of the EEG signal, the stimulus stimulating cerebrospinal fluid flow.

[0009] In some aspects, the techniques described herein relate to a method, wherein the target feature of the EEG signal includes at least one of a peak of a slow wave, a slow wave coupled to another frequency band, a phase of EEG bandpower, an amplitude of EEG bandpower, or a wave-linked increase in spindle power.

[0010] In some aspects, the techniques described herein relate to a method, wherein the target feature of the EEG signal includes a peak of a slow wave.

[0011] In some aspects, the techniques described herein relate to a method, wherein predicting the peak of the slow wave is based on an age of the human subject.

[0012] In some aspects, the techniques described herein relate to a method, wherein the stimulus includes an auditory stimulus.

[0013] In some aspects, the techniques described herein relate to a method, wherein the auditory stimulus includes pink noise and delivering the auditory stimulus includes playing about 50 milliseconds of the pink noise.

[0014] In some aspects, the techniques described herein relate to a method, wherein the stimulus includes a vibration.

[0015] In some aspects, the techniques described herein relate to a method, wherein the vibration is a bone conduction vibration.

[0016] In some aspects, the techniques described herein relate to a method, wherein predicting the target feature of the EEG signal includes using a recurrent neural network trained on previously collected EEG data.

[0017] In some aspects, the techniques described herein relate to a method, wherein the previously collected EEG data includes previously collected EEG data from the human subject.

[0018] In some aspects, the techniques described herein relate to a method, wherein the previously collected EEG data includes previously collected EEG data from human subjects with ages within 10 years of the age of the human subject.

[0019] In some aspects, the techniques described herein relate to a method, wherein the delivering the stimulus to coincide with the target feature of the EEG signal includes starting to deliver the stimulus within about 100 milliseconds of the target feature of the EEG signal.

[0020] In some aspects, the techniques described herein relate to a method, wherein the delivering the stimulus to coincide with the target feature of the EEG signal includes starting to deliver the stimulus within about 70 milliseconds of the target feature of the EEG signal.

[0021] In some aspects, the techniques described herein relate to a method, wherein delivering the stimulus includes delivering the stimulus for a duration of the target feature of the EEG signal.

[0022] In some aspects, the techniques described herein relate to a method, wherein acquiring the EEG signal from the human subject while the subject is asleep includes acquiring the EEG signal while the human subject is in non-rapid eye movement (NREM) sleep.

[0023] In some aspects, the techniques described herein relate to a device for stimulating cerebrospinal fluid flow in a subject, the device including an electroencephalography (EEG) headset configured to be worn by the subject and to record an EEG signal from the subject while the subject is sleeping, an EEG recording machine operably coupled to the EEGheadset and configured to receive the EEG signal from the EEG headset, a stimulus emitting device configured to emit a stimulus to the subject to coincide with a target feature of the EEG signal, and at least one processor operably coupled to the EEG recording machine and the stimulus emitting device and configured to control the stimulus emitting device using a recurrent neural network trained on EEG data acquired previously from the subject.

[0024] In some aspects, the techniques described herein relate to a device, wherein the wherein the stimulus emitting device includes at least one headphone and the stimulus includes an auditory stimulus.

[0025] In some aspects, the techniques described herein relate to a device, wherein the auditory stimulus includes pink noise.

[0026] In some aspects, the techniques described herein relate to a device, wherein the stimulus includes a vibration.

[0027] In some aspects, the techniques described herein relate to a method of stimulating cerebrospinal fluid flow, the method including acquiring an electroencephalography (EEG) signal from a human subject while the human subject is asleep, predicting a peak of a slow wave of the EEG signal using a recurrent neural network trained on previously collected EEG data, and playing an auditory stimulus to the human subject within about 100 milliseconds of the peak of the slow wave of the EEG signal, the auditory stimulus stimulating cerebrospinal fluid flow.

[0028] All combinations of the foregoing concepts and additional concepts discussed in greater detail below (provided such concepts are not mutually inconsistent) are part of the inventive subject matter disclosed herein. In particular, all combinations of claimed subject matter appearing at the end of this disclosure are part of the inventive subject matter disclosed herein. The terminology used herein that also may appear in any disclosure incorporated by reference should be accorded a meaning most consistent with the particular concepts disclosed herein.BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The skilled artisan will understand that the drawings primarily are for illustrative purposes and are not intended to limit the scope of the inventive subject matter described herein. The drawings are not necessarily to scale; in some instances, various aspects of the inventive subject matter disclosed herein may be shown exaggerated or enlarged in the drawings to facilitate an understanding of different features. In the drawings, like referencecharacters generally refer to like features (e.g., functionally and / or structurally similar elements).

[0030] FIG. 1 A shows a schematic of an example device to stimulate CSF flow in a subject.

[0031] FIG. IB shows a schematic representation of using the device of FIG. lAto stimulate cerebrospinal fluid (CSF) flow in a subject.

[0032] FIG. 2A shows stimuli (vertical lines) delivered to coincide with the peaks of electroencephalography (EEG) slow waves for phase-targeted stimulation.

[0033] FIG. 2B shows the slow wave phase at the time of stimulation (N = 976 stimuli; 0 degrees corresponds with the slow wave peaks).

[0034] FIG. 2C shows the event-related potentials (ERPs) of the stimulus (lower trace, N = 976) and sham (upper trace, N = 1025) events. The shaded area shows the boot strap 95% confidence interval.

[0035] FIG. 3 A shows the difference between the time-frequency evoked response of the stimulus and sham conditions.

[0036] FIG. 3B shows the slow wave power following stimulus (left, N = 976) and sham events (right, N = 1025). The lower bars in each group show the group means and the upper central horizontal bar indicates significance.

[0037] FIG. 3C shows the spindle power following stimulus (left, N = 976) and sham events (right, N = 1025). The lower bars in each group show the group means and the upper central horizontal bar indicates significance.

[0038] FIG. 3D shows the evoked response of the normalized CSF signal to stimulus (upper trace) and sham events (lower trace). Shaded areas show bootstrap 95% confidence intervals, and vertical line shows stimulus timing.

[0039] FIG. 4 shows example slow waves with different amplitudes and / or shapes for two individuals, Person A and Person B.

[0040] FIG. 5 shows an illustration of a raw EEG signal and example EEG target features that can be acquired from the EEG signal.

[0041] FIG. 6A shows the ERP, CSF, and ERP-spectrum of stimulation at the pre-peak of a slow wave phase.

[0042] FIG. 6B shows the ERP, CSF, and ERP-spectrum of stimulation at the post-peak of a slow wave phase

[0043] FIG. 6C shows the ERP, CSF, and ERP-spectrum of stimulation at the pre-trough of a slow wave phase

[0044] FIG. 6D shows the ERP, CSF, and ERP-spectrum of stimulation at the post-trough of a slow wave phaseDETAILED DESCRIPTION

[0045] Since the waves of CSF flow tend to follow EEG slow waves, the inventors recognized that the electrical activity of the EEG and CSF flow in the brain could be coupled. The inventors further recognized that modulating CSF flow through the coupling of EEG activity may therefore be a promising target for enhancing brain fluid transport and maintaining brain health. Thus, the inventors recognized that phase-targeted EEG neurofeedback may stimulus CSF flow.

[0046] Disclosed herein are neurofeedback devices and methods for stimulating CSF flow in a subject. One technique that may increase the amount of slow- wave activity is called closed loop auditory stimulation (CLAS). During CLAS, brief pulses of auditory noise are played while the subject is sleeping such that the sounds coincide with the peaks of the subject’s slow waves. CLAS may increase the presence of EEG slow waves and improve performance in memory tasks upon waking. Slow wave sleep is a portion of non-rapid eye movement (NREM) sleep characterized by high-amplitude (e.g., approximately > 50pV), low-frequency (e.g., approximately 0.5-2Hz) brain waves. FIG. 2A illustrates an example of a slow wave. However, due to the artifact, latency, and quality issues with data collection for EEG-fMRI disclosed above, the use of EEG-fMRI to image CLAS has not previously been disclosed.

[0047] The methods and devices disclosed herein may include using CLAS to alter the flow of CSF during sleep by providing precisely timed auditory feedback through the use of a neural network to predict the optimal stimulus timing. The methods and devices disclosed herein may demonstrate the potential of low-latency EEG-fMRI neurofeedback, while also shedding new light on the phenomenon of CLAS. Experimentation with CLAS has shown that by delivering auditory stimuli in-phase with EEG slow waves during sleep (see FIG. 2A), performance of learning and memory tasks upon waking may be improved and slow wave duration and amplitude may be increased.

[0048] The methods and devices disclosed herein demonstrate a method for performing low- latency neurofeedback inside an MR scanner including artifact removal and preprocessing. For example, by performing a CLAS experiment inside a magnetic resonance (MR) scanner, high resolution spatial data can be collected to develop a new type of CLAS designed to increase CSF flow and provide a tool to personalize individual therapies involving CLAS(e.g., by providing a way to time the delivery of the auditory stimuli in-phase with EEG slow waves during sleep). The inventive devices and methods disclosed herein can be used to study and use the relationship between CSF and slow waves, which have been shown to be temporally linked. The methods and devices disclosed herein provide a mechanism for phase- targeted EEG neurofeedback (optionally inside an MR scanner), eliminating the need for artifact removal and providing lower latency.

[0049] The methods and devices disclosed herein also include predicting and learning when to stimulate a subject using a trained neural network, eliminating the need for a MR scanner. In one embodiment, an MR scanner may be used to tune the personalization of the delivery of the stimulus. Alternatively, the personalization of the delivery of the stimulus may be determined using prior MR studies and / or data. For example, the use of EEG-fMRI to image CLAS may allow for the optimization of neurofeedback and for the training of a neural network. The trained neural network disclosed herein may be used to predict slow waves without the use of an MR scanner. Thus, the methods and devices disclosed herein may also be used outside of a MR scanner.

[0050] The devices and methods disclosed herein demonstrate the use of EEG-fMRI neurofeedback to enhance CSF flow in the brain and may also provide an insight into the relationship between slow waves, CSF, and CLAS. Thus, the devices disclosed herein may have applications in healthcare and wellness. By increasing the flow of CSF into and out of the brain, the methods and devices disclosed herein may be used to modulate waste clearance function and physiology in the brain, potentially helping to reduce waste accumulation linked to aging and cognitive decline, such as amyloid-beta and tau proteins that are characteristic of neurodegenerative disorders including Alzheimer’s Disease.

[0051] Closed-Loop Noninvasive Brain Stimulation Device

[0052] FIG. 1 A illustrates a schematic of a device 100 that can be used to stimulate CSF flow in a subject 101. The subject may be a mammal, for example, a human. The device 100 may include a stimulus emitting device 110, an EEG headset 120, a sensor 130, an EEG recording machine 140, a processor 150 with memory 160 configured to run a trained neural network 170, and a communication interface 180 as shown in FIG. 1A. The device 100 may also include a transceiver and / or antenna 151 (e.g., a Bluetooth, Wi-Fi, or cellular transceiver) to communicate the results (e.g., the EEG signals) obtained from the device 100 to an external device (e.g., an external computer). The device 100 may also include a rechargeable battery or other power source. For example, the device may be connected via a wire to an external power source 190 as shown in FIG. 1 A.

[0053] The stimulus emitting device 110, EEG headset 120, EEG recording machine 140, processor 150, memory 160, neural network 170, communication interface 180, antenna 151, and / or power source 190 may be implemented in separate (purpose-built) devices, different combinations of devices, or a single device, such as an appropriately programmed computer. If implemented as separate devices or combinations of devices, the stimulus emitting device 110, EEG headset 120, EEG recording machine 140, processor 150, memory 160, neural network 170, communication interface 180, antenna 151, and / or power source 190 may be connected through one or more wires and / or wireless connections (e.g., through the use of Wi-Fi connections, Bluetooth connections, cellular connections, satellite links, and / or a local area network). The EEG recording machine 140, processor 150, memory 160, neural network 170, communication interface 180, and antenna 151 may be combined into a single physical device (e.g., a computing device such as a computer). Alternatively, the EEG recording machine 140, processor 150, memory 160, neural network 170, communication interface 180, and antenna 151 may be contained in one or more separate devices (e.g., multiple computing devices, such as multiple computers).

[0054] The stimulus emitting device 110 may include one or more headphones (as shown in FIG. 1 A), bone conduction devices (e.g., bone conduction headphones), and / or speakers. The headphones may be worn by the subject 101 and may be used to deliver an auditory stimulus, which can be generated beforehand and stored in the memory 160 or generated in real time by the processor 150 or the stimulus emitting device 110 itself. The auditory stimulus may be pink noise, which is random noise having equal energy per octave. The pink noise may range from a frequency of about 20 Hz to about 20,000 Hz. The amplitude of the pink noise may be between about -1 and about 1. The average amplitude of the pink noise may be around zero. Instead of, or in addition to pink noise, the auditory stimulus may include another suitable auditory stimulus (e.g., white noise, which is random noise having many frequencies with equal intensities, with a frequency range of 20 Hz to about 20,000 Hz). For example, the audio stimulus may include a spoken sound, such as a recording of someone saying a consonant (e.g., “b,” “c,” “d,” “f,” “g,” etc.) or vowel (e.g., “a,” “e,” “i,” “o,” “u,” or “y”), and / or a tone (e.g., a musical note such as a B tone (a frequency of about 493.88 Hz) and / or a C tone (a frequency of about 261.6 Hz)). The audio stimulus may include a variable sequence of pink noise, white noise, spoken sound(s), and / or tone(s). Alternatively, the audio stimulus may include a mixture of pink noise, white noise, spoken letter(s), vowel(s), and / or tone(s). Preferably the auditory stimulus is pink noise.

[0055] The stimulus emitting device may also provide other kinds of sensory stimulation, such as haptic vibrations and / or visual stimulus (e.g., flashes of light). The vibration may include, but is not limited to, a bone conduction vibration. For example, the stimulus emitting device 110 may include a motor 111 to deliver a haptic vibration. Instead of, or in addition to, motor 111, the stimulus emitting device 110 may include a light source 112 (e.g., a light emitting diode (LED) to deliver a visual stimulus. The stimulus emitting device 110 may be operably coupled to the processor 150 through one or more wires and / or wireless connections (e.g., e.g., through the use of Wi-Fi connections, Bluetooth connections, cellular connections, satellite links, and / or a local area network).

[0056] The length of the audio stimulus may range from about 5 milliseconds to about 500 milliseconds, for example, about 20 milliseconds to about 100 milliseconds, or about 30 milliseconds to about 80 milliseconds, or any duration in between (e.g., 15, 25, 30, 40, 50, 60, 70, 75, 85, 95, 100, 110, 120, 130, 140, 150, 160, 170, 180, 190, 200, 210, 220, 230, 240, 250, 260, 270, 280, 290, 300, 325, 350, 375, 400, 425, 450, or 475 milliseconds). The duration of the audio stimulus may be less than the duration of the EEG target feature (e.g., a slow wave peak, which may be about 100 milliseconds in length, and / or a slow wave, which may be about 1 second in length). Alternatively, the length of the audio stimulus may be for the duration of the EEG target feature (e.g., the length of the slow wave peak).

[0057] The EEG headset 120 may be worn by the subject 101 during sleep. The EEG headset 120 may include one or more EEG electrodes 121 positioned in contact with the subject’s scalp. For example, the EEG electrodes 121 may be positioned on the subject’s forehead. The EEG electrodes 121 may be conductive electrodes in the form of single electrodes (e.g., a reference electrode, a ground electrode, and a measurement electrode), strips (e.g., six electrodes), and / or a cap (e.g., a whole head cap including at least 32 electrodes with a plurality of channels). The EEG electrodes 121 can be placed in a variety of locations on the patient, such as with the data collection electrode on the center of the forehead, (called ‘Fpz’ in EEG terms), which is where slow waves are relatively easy to record, or elsewhere on scalp.

[0058] The EEG headset 120 may be connected to an EEG recording machine 140 through one or more wires and / or wireless connections (e.g., Wi-Fi connections, Bluetooth connections, cellular connections, satellite links, and / or local area network connections). The EEG recording machine 140 may include an amplifier to amplify the signals from the EEG headset 120. The EEG headset 120 and EEG recording machine 140 may be separate devices(e.g., as shown in FIG. 1 A). Alternatively, the EEG headset 120 and EEG recording machine 140 may be combined into one device.

[0059] The EEG recording machine 140 may measure signals from the EEG headset 120. The EEG recording machine 140 may be operably coupled to the processor 150 which may pass the EEG signals obtained from the EEG recording machine 140 through the neural network 170 to predict a slow- wave phase of the subject 101. The EEG recording machine 140 may filter the raw EEG data prior to providing the EEG data to the neural network 170. For example, the EEG recording machine 140 may process the EEG data to remove any scanner artifacts. The artifact filtering may include using an adaptive filter (e.g., a Kalman filter) and a reference signal (such as from an EEG reference layer or a carbon wire loop) to reduce latency for stimulus triggering (the time between the EEG feature being detected and the EEG feature being identified by the processor 150), and noise. The EEG recording machine may amplify the raw EEG data prior to providing the EEG data to the neural network 170.

[0060] The device 100 may also include an optional sensor 130 to measure a physiological state of the subject 101, for example, to assist in determining the sleep stage and / or autonomic state of the subject. For example, the sensor 130 may include, but is not limited to, a finger motion sensor (e.g., a photoplethysmography (PPG) sensor), a pulse oximeter, an accelerometer, and / or a galvanic skin response sensor. The sensor 130 may be positioned on a finger(s) of the subject 101. The sensor 130 may be used to determine what stage of sleep the subject is in (e.g., NREM sleep vs. rapid eye movement (REM) sleep). The sensor 130 may provide data to the neural network 170 (e.g., to improve the neural network 170 to assist in predicting when to play the stimulus).

[0061] The device 100 may also include a processor 150 that is operably coupled to the EEG headset 120, sensor 130, stimulus emitting device 110, EEG recording machine 140, and / or communications interface 180. The processor 150 may be connected to the EEG headset 120, sensor 130, stimulus emitting device 110, EEG recording machine 140, and / or communications interface 180 through one or more wires and / or wireless connections (e.g., e.g., through the use of Wi-Fi connections, Bluetooth connections, cellular connections, satellite links, and / or a local area network). The processor 150 may generate the stimulus, control emission of the stimulus, monitor emission of the stimulus and / or EEG signals obtained from the EEG headset 120 and / or EEG recording machine 140, and / or process feedback regarding the stimulus and EEG signals obtained from the EEG headset 120 and / or EEG recording machine 140. The processor 150 may also monitor data obtained from thesensor 130. The processor 150 may also control the emission of the stimulus using the neural network 170 (e.g., through processor-executable instructions stored in memory 160). The processor 150 may also control the amplitude or power level of the auditory stimulus emitted by the device 100.

[0062] The device 100 may also include one or more forms of memory 160 coupled to the processor 150. The memory 160 may store processor-executable instructions, the neural network 170, a stimulus generation policy, and / or data acquired by the EEG headset 120 and optional sensor 130. For example, the memory may store processor-executable instructions for the stimulus (e.g., a stimulus generation policy) that may be sent to the processor 150 to generate the stimulus. Depending on the exact configuration and type of device 100, the memory 160 may be volatile (such as Random Access Memory (RAM)), non-volatile (such as Read-Only Memory (ROM), flash memory, etc.) or some combination of the two.

[0063] The neural network 170 may include but is not limited to a recurrent neural network (RNN), a convolutional neural network (CNN), a transformer, an attention network, a support vector machine (SVM), a random forest model, and / or another suitable machine learning technique. The neural network 170 may receive input from the EEG recording machine 140, which is coupled to the processor 150, and provide output via the stimulus emitting device 110.

[0064] The neural network 170 may predict when to deliver the auditory stimulus for a given subject 101 based on EEG data collected with the EEG headset 120 and EEG recording machine 140. The neural network 170 may provide the processor 150 with instructions on when to deliver the stimulus for the subject 101. Preferably the stimulus is delivered in combination with a target EEG feature. The target EEG feature may include a peak of the slow wave, a different phase of the slow wave (e.g., a trough of the slow wave, a positive and / or negative zero-crossing of the slow wave, a pre-peak of the slow wave, a post-peak of the slow wave, a pre-trough of the slow wave, a post-trough of the slow wave), a slow wave coupled to another frequency band (e.g., a gamma band, which is a > 30 Hz wave, and / or a delta, theta, alpha, and / or beta band), a phase of EEG bandpower (e.g., a phase calculated from the envelope of a slow wave, a delta band, a theta band, an alpha band, and / or a beta band), an amplitude of EEG bandpower (e.g., a high and / or low amplitude of a slow wave and / or a different EEG band, such as a delta, theta, alpha, beta, and / or gamma band), a change in spindle power (e.g., an increase in spindle power), and / or any other EEG feature which can be identified to drive cerebrospinal fluid flow in the subject 10E For example, thetarget EEG feature may include a cross-frequency coupling, which is an association between different brain wave frequences in an EEG (e.g., a slow wave and a gamma band).

[0065] FIG. 5 shows a raw EEG signal and example target EEG feature that may be obtained from the EEG signal and delivery of the stimulus in combination with the target EEG feature. For example, the stimulus may be targeted to be delivered in combination with a slow wave peak (top panel), an elevated band power (middle panel), and / or a high amplitude event (bottom panel). The intensity (e.g., amplitude) of a slow wave (and / or another EEG band) can be used to target the playing of the stimulus with a high amplitude event. To determine elevated band power, the phase of an EEG signal may also be calculated from the envelope of a signal filtered into a band of interest (e.g., a slow wave). For example, a signal may be filtered into a slow-wave band, or the beta band (or any other EEG band), the envelope of that signal may be calculated using Hilbert’s transform, then the phase of that envelope may be calculated, and the stimulus may be delivered to target that phase.

[0066] The neural network 170 may accept one or more inputs. In one embodiment, the neural network 170 may accept three inputs derived from the EEG signal: 1) the raw signal amplitude (i.e., the EEG signal amplitude obtained from the EEG headset 120 and / or the EEG recording machine 140), 2) the infinite impulse response (IIR) filtered signal amplitude, and 3) the backwards difference of the raw signal (e.g., the difference between one sample point (sample tn) and the immediately preceding sample point (t„-i)). The parameters of the HR filter may include a passband of about 0.4-2 Hz, stopbands of 0.01 Hz and 6 Hz, stopband attenuations of about 20 dB, and a passband ripple of about 1. The neural network 170 may also contain one or more layers, including but not limited to the following ten layers: 1) a 3 * 1 sequence input layer, 2) a 50 * 1 long short-term memory (LSTM) layer, 3) a 100 x 1 LSTM layer, 4) a 50 * 1 fully connected layer, 5) a rectified linear unit (ReLu) layer, 6) a 25 * 1 fully connected layer, 7) a ReLu layer, 8) a 25% dropout layer, 9) a 2 * 1 fully connected layer, and / or 10) a regression layer with L2 norm loss. The output of the neural network 170 may be 2 * 1, representing the real and imaginary components of the Hilbert transform of the forward-and-reverse band-pass filtered EEG signal, from which phase is calculated.

[0067] The neural network 170 may be trained on previously collected EEG data. The previously collected EEG data may be previously collected EEG data from the subject 101. Alternatively, the previously collected EEG data may be data from a different subject of a similar age (e.g., an age range of about 10 years, for example, about 1 year, about 2 year, about 3 year, about 4 years, about 5 years, about 6 years, about 8 years, about 9 years, orabout 10 years, including all ages in between) and / or gender, or from multiple subjects. The multiple subjects may be of a similar age and / or gender to the subject 101. Alternatively, the multiple subjects may be a range of ages and / or genders. The EEG data from a different subject may also be from a subject with a similar condition (e.g., a subject with dementia, Alzheimer's disease, insomnia, and / or schizophrenia). The neural network 170 may be trained with previously collected EEG recordings that are divided into 4-second-long segments, and / or segments of another duration (e.g., 1-second-long segments, 2-second-long segments, 3-second-long segments, 4-second-long segments, 5-second-long segments, 6-second-long segments, 7-second-long segments, 8-second-long segments, 9-second-long segments, 10- second-long segments, 20-second-long segments 1-second-long segments, 30-second-long segments, 40-second-long segments, 50-second-long segments, 60-second-long segments, 70-second-long segments, 80-second-long segments, 90-second-long segments, 100-second- long segments, or any other suitable length, including all ranges in between). For example, the neural network 170 may be trained on at least 1 hour, at least two hours, at least three hours, at least four hours, at least five hours, at least six hours, at least seven hours, at least eight hours, at least nine hours, and / or at least ten hours or more of previously collected EEG data. The training labels (e.g., the real and imaginary components of the Hilbert transform) may be shifted, e.g., by about 75 milliseconds, to account for delays in computation and stimulus delivery. Preferably, the training labels are shifted forward in time such that the neural network 170 may predict a future EEG feature / state (e.g., a future peak of a slow wave). During training, a minibatch size of 128 samples may be used. The neural network 170 may undergo one or more training epochs, for example, 30 training epochs.

[0068] The neural network 170 may be trained on when to deliver the stimulus for the subject 101. Preferably the neural network 170 is trained to deliver the stimulus to coincide with a target EEG feature. The target EEG feature may include, but is not limited to, a peak of a slow wave, a different phase of the slow wave, a slow wave coupled to another frequency band, a phase of EEG bandpower, an amplitude of EEG bandpower, a wave-linked increase in spindle power. Preferably the target EEG feature is an EEG feature that may stimulate CSF flow in the subject.

[0069] For example, when the target EEG feature is the peak of a slow wave, the neural network 170 may be trained such that the auditory stimulus is delivered at or near the peak of a slow wave. The neural network 170 may be able to predict the slow-wave phase with a latency of less than about 1 second, preferably less than about 500 milliseconds, more preferably less than about 200 milliseconds, more preferably less than about 150milliseconds, or more preferably less than about 100 milliseconds. The latency may refer to the total computational and / or processing time of the neural network 170 — i.e., the time from when something happens in the brain to when the neural network 170 finishes making a prediction about it. For example, the auditory stimulus may be delivered within about 100 milliseconds, preferably within about 70 milliseconds, more preferably within about 50 milliseconds of the peak of a slow wave, as illustrated in FIG. IB. The peak of the slow wave may last for approximately 100 milliseconds to approximately 200 milliseconds, depending on an individual’s age, drug / medication use, the presence of any sleep disorders, and / or how well rested the individual is.

[0070] The neural network 170 may also be trained to target another feature of an EEG (e.g., a different phase of the slow wave, a slow wave coupled to another frequency band, a phase of EEG bandpower, an amplitude of EEG bandpower, a wave-linked increase in spindle power) and / or trained to deliver the stimulus to coincide with another feature of an EEG if a different feature is determined to be more effective for a given subject 101 based on their own prior EEG data. The neural network 170 may be able to predict the target EEG feature a latency of less than about 1 second, preferably less than about 500 milliseconds, more preferably less than about 200 milliseconds, more preferably less than about 150 milliseconds, or more preferably less than about 100 milliseconds. For example, the auditory stimulus may be delivered within about 100 milliseconds, preferably within about 70 milliseconds, more preferably within about 50 milliseconds of the target EEG feature.

[0071] The device 100 may also include at least one communication interface 180 to communication with the subject 101 and / or a third party (e.g., a healthcare provider, a caretaker, a clinical research investigator, a database, a monitoring application, etc.). The communication interface 180 may allow the subject 101 and / or third party to monitor and / or control the device 100. For example, the communication interface 180 may allow the subject 101 and / or third party to view the EEG signals obtained by the device 100. The communication interface 180 may also allow the subject 101 and / or third party to view the stimulus generating policy. The communication interface 180 can be implemented in a display with keyboard, mouse, or other input device(s) or in a touchscreen or other suitable device. The communication interface 180 may allow the subject 101 and / or third party to view the status of the device 100 (e.g., power level), how many times the stimulus has been provided in previous sessions, basic information about the subject’s sleep patterns (e.g., how long the subject 101 slept, when the subject 101 experienced various sleep stages, and / or a hypnogram). The communication interface 180 may also provide information about the EEGelectrode 121 impedance and / or signal quality (e.g., to ensure that the EEG electrodes 121 are in proper contact with the subject’s 101 scalp). The communication interface 180 may also allow the subject 101 and / or third party to view and / or control the volume of the stimulus and / or the type of stimulus emitted by the device (e.g., pink noise, white noise, spoken sound(s), tone(s), and / or stimulus combinations as disclosed herein).

[0072] FIG. IB is a schematic illustrating operation of the device 100 to stimulate CSF flow in the subject 101. From left to right, the subject 101 wears the EEG headset 120 to sleep. The signals from the EEG headset 120 are passed through the device 100 and the neural network 170 to predict a target feature of the EEG (e.g., a slow- wave phase of the subject 101). The processor uses these predictions to time an auditory stimulus played via the stimulus emitting device 110 (e.g., headphones), such that the playing of the auditory stimulus coincides with desired features of neurological activity (e.g., a peak of a slow wave). This auditory stimulus provokes increased CSF flow in the subject 101.

[0073] As shown in FIG. IB, the device 100 may deliver the auditory stimulus at or near a peak of the slow wave. Alternatively, the device 100 may deliver the auditory stimulus at another timing that is optimal for the subject 101 as described above. For example, the device 100 may deliver the auditory stimulus at a different phase of the slow wave, a slow wave coupled to another frequency band, a phase of EEG bandpower, an amplitude of EEG bandpower, a wave-linked increase in spindle power, and / or any other EEG feature which can be identified to drive CSF flow in the subject 101. The device 100 may determine when to deliver the auditory stimulus for a given subject 101 using the neural network 170. The device 100 may be used to increase CSF flow in the subject 101 due to playing of the auditory stimulus at or near the peak of the slow wave and / or at another phase as trained in the subject-specific timing.

[0074] For example, as illustrated in FIG. 4, the slow waves for two different individuals, Person A and Person B, may have different amplitudes and / or different shapes. The amplitude and / or shape of the slow wave may affect the peak of the slow wave for a given individual. As a result, the optimal timing for delivering the auditory stimulus for Person A may be different than for Person B.

[0075] The peak of a slow wave and / or the shape of the slow wave may vary based on an individual’s age, gender, and / or clinical conditions such as Alzheimer's disease, insomnia, and / or schizophrenia. For example, an older individual may spend less time in slow wave sleep, potentially resulting in a different slow wave shape and / or a difference in the occurrence and duration of slow wave sleep. Thus, the neural network 170 may take intoaccount an individual’s age to determine when to deliver the auditory stimulus based on the shape of a slow wave for an individual of a similar age.

[0076] Thus, the neural network 170 may be trained to personalize the timing of delivering the stimulus for a given subject 101. The neural network 170 may be trained and / or re-trained in real-time (e.g., within a few seconds). For example, the subject 101 may wear the device 100 to sleep and the neural network 170 may be trained on the EEG data of the subject 101 such that the neural network 170 may then be able to predict a target EEG feature of the subject 101 and the device 100 may be able to deliver the stimulus to the subject 101 for the rest of the night. In this embodiment, the ground truth may be available within seconds (e.g., less than 10 seconds) of the neural network 170 predicting a target EEG feature for the subject. The neural network 170 may compare this predict! on(s) and any errors in the prediction(s) to the ground truth enabling the neural network 170 to be continuously trained / re-trained in near-real-time.

[0077] In one embodiment, the device 100 may be used to learn the optimal timing for delivering the stimulus for a given subject 101 (e.g., to coincide with the peak of a slow wave or another target EEG feature as described above). The neural network 170 may be trained using the device 100 without an MR scanner. Instead of, or in addition to, training on the device 100, the neural network 170 may also be trained on the device 100 used in an MR scanner. The EEG data obtained with the device 100 in an MR scanner may contain more noise than EEG data obtained with the device 100 alone. Thus, the neural network 170 may further be trained to compensate for the noise induced by the MR scanner. The timing for delivering the stimulus may also be validated using an MR scanner. For example, the training of the neural network 170 may include validation of the timing for delivering the stimulus using an MR scanner. The device 100 may then be used to allow for personalized timing of the delivery of the stimulus without the use of the MR scanner.

[0078] The device 100 and methods disclosed herein provide several advantages over existing methods for using CLAS to improve memory. For example, previous methods using CLAS to improve memory relied upon a threshold value, wherein an auditory stimulus would be played whenever an EEG signal exceeded a pre-determined threshold value. However, these methods may result in administering the auditory stimulus too early (e.g., before the peak of the target EEG feature, such as the actual peak of the slow wave), too late (e.g., for any peaks that are below the threshold value), and / or missing an EEG feature (e.g., a slow wave event) entirely for individuals with smaller EEG features, such as smaller slow waves. Thus, the neural network 170 disclosed herein may provide a mechanism to more accuratelytime the delivery of an auditory stimulus based on the shape and timing of a slow wave for a given subject 101 and / or based on that individual’s prior EEG responses to a stimulus.

[0079] Methods

[0080] Eleven healthy adults participated in an EEG-fMRI nap study, and each completed two 25-minute sleep runs. During each run, they were instructed to press a button with each breath in until they fell asleep. While they were doing this, a previously trained recurrent neural network (e.g., neural network 170) was used to predict upcoming slow wave phase of channel FpZ. When the phase was predicted to fall within a desired range, corresponding to the slow- wave peak, the subject 101 randomly received either a real audio stimulus (about 50 milliseconds of pink noise; 50% chance) or a sham stimulus (no audio stimulus; 50% chance).

[0081] EEG data was collected using a 32-channel BrainCap MR headset with carbon wire loops and BrainAmp MR amplifiers and was preprocessed in real time using Low Latency Artifact Mitigation Acquisition Software (LLAMAS) to remove scanner artifacts. LLAMAS was also used to implement the neural network 170 and stimulus delivery via the stimulus emitting device 110. The artifact filtering may include using an adaptive filter (e.g., a Kalman filter) and a reference signal (such as from an EEG reference layer or a carbon wire loop) to offer a lower latency and noise reduction. Stimuli were delivered in phase with slow waves using the neural network 170 disclosed herein.

[0082] MR data was collected with a 3T Siemens Prisma scanner with 64 channel head coil. Functional scans consisted of single-shot gradient echo multi-band echo planar imaging (EPI). Scans used 2.5mm isotropic resolution, 40 slices, and a TR of 378ms, calling upon recent advances in fast fMRI. Volumes were positioned with the bottom slice at the lower entrance to the 4thventricle, so that CSF flow could be measured.

[0083] Results

[0084] The evoked response potential was calculated for each stimulus event and each sham event, as was the slow-wave phase at the time of stimulus delivery. To exclude stimuli delivered while the subject 101 was still awake, we removed all epochs in which the subject 101 was performing the breathing task. In addition, to exclude stimuli delivered while the subject 101 was asleep, but not experiencing a slow wave, we excluded all stimuli for which the delta band (e.g., about 1-3 Hz) power in the previous 10 seconds was below 5 pV2. The slow wave phase at the time of stimulus for the remaining events is shown in FIG. 2B. In FIG. 2B, zero degrees corresponds to slow wave peaks. A random-shuffle procedure was used to assess the statistical difference between the stimulus and sham event-related potentials(ERPs). The labels of each sample were shuffled between 5,000 and 10000 times, and the maximum absolute difference between the means of shuffled stim and sham groups was calculated. This distribution was used to assess the significance of the difference between the true stimulus and sham groups. A significant difference between stimulus and sham was found (P < 0.0001, FIG. 2C).

[0085] Time-frequency ERPs were also calculated for each event. To assess the effect of the effect of the stimulus on spindle power, the mean power in the frequency range of about 13- 16 Hz and time range of about 0.5-1.25 seconds was calculated from each sample, and an unpaired t-test was used to compare stimulus and sham. Spindles may play a role in memory consolidation and / or blocking sensory input to the brain during sleep. For example, an increase in spindle power may be related to an increase in memory consolidation. The spindle power may be used to validate that the stimulus is affecting the patient during sleep. To assess the impact of the stimulus on slow waves, the same comparison was made for the frequency range of about 0.4-3Hz and time range of about 0.4-0.7 seconds. The data show a statistically significant increase in both slow wave power (P < 0.0001) and spindle power (P < 0.0001) in the stimulus group (FIGS. 3A-3C).

[0086] An increase in spindle power following the stimulus was also observed, which demonstrates that the stimulus may effectively engage memory consolidation processes. Thus, the device 100 may cause an increase spindle power and / or CSF flow in the subject 101.

[0087] As illustrated in FIGS. 2C and 3 A-3B, the stimulus may increase power in the slow- wave frequency band and may also increase the amplitude of the subsequent slow wave. Additionally, the stimulus may cause a slow wave to occur (or continue to occur) when otherwise the slow wave may have ended or not have occurred at all. These results demonstrate that slow-wave power may increase following delivery of a timed stimulus. The device 100 may generate a neural change that increases occurrences of slow waves and / or lengthens slow waves in the subject 101, which in turn may generate an increase in CSF flow. For example, the device 100 may cause any ongoing slow waves in the subject 101 to continue for longer compared to a sham stimulus. The device 100 may also cause an increase in CSF flow. The stimulus may also cause an increase in slow wave duration in the subject 101, which may increase CSF flow downstream in the subject 101.

[0088] Evoked responses of the CSF flow to each included stimulus were calculated and resampled to 4 Hz. The significance of the difference between the evoked responses of stimulus and sham were assessed using the same shuffle procedure as described above (FIG.3D). We found a significant increase in CSF signal in the stimulus condition compared to the sham condition (P < 0.0001). These results show that CSF flow into the 4thventricle may be driven by the timed delivery of an auditory stimulus using the device 100. Furthermore, an increase in CSF flow following slow wave peaks may be increased with the presentation of an audio stimulus using the device 100. This suggests that delivery of the auditory stimulus at or near the peak of a slow wave may increase the flow of CSF.

[0089] FIGS. 6A-6D show how stimulation at various slow wave phases including pre-peak (FIG. 6 A), post-peak (FIG. 6B), pre-trough (FIG. 6C), and post-trough (FIG. 6D) may also affect CSF flow and electrophysiology.

[0090] Conclusion

[0091] While various inventive embodiments have been described and illustrated herein, those of ordinary skill in the art will readily envision a variety of other means and / or structures for performing the function and / or obtaining the results and / or one or more of the advantages described herein, and each of such variations and / or modifications is deemed to be within the scope of the inventive embodiments described herein. More generally, those skilled in the art will readily appreciate that all parameters, dimensions, materials, and configurations described herein are meant to be exemplary and that the actual parameters, dimensions, materials, and / or configurations will depend upon the specific application or applications for which the inventive teachings is / are used. Those skilled in the art will recognize or be able to ascertain, using no more than routine experimentation, many equivalents to the specific inventive embodiments described herein. It is, therefore, to be understood that the foregoing embodiments are presented by way of example only and that, within the scope of the appended claims and equivalents thereto, inventive embodiments may be practiced otherwise than as specifically described and claimed. Inventive embodiments of the present disclosure are directed to each individual feature, system, article, material, kit, and / or method described herein. In addition, any combination of two or more such features, systems, articles, materials, kits, and / or methods, if such features, systems, articles, materials, kits, and / or methods are not mutually inconsistent, is included within the inventive scope of the present disclosure.

[0092] Also, various inventive concepts may be embodied as one or more methods, of which an example has been provided. The acts performed as part of the method may be ordered in any suitable way. Accordingly, embodiments may be constructed in which acts are performed in an order different than illustrated, which may include performing some acts simultaneously, even though shown as sequential acts in illustrative embodiments.

[0093] All definitions, as defined and used herein, should be understood to control over dictionary definitions, definitions in documents incorporated by reference, and / or ordinary meanings of the defined terms.

[0094] The indefinite articles “a” and “an,” as used herein in the specification and in the claims, unless clearly indicated to the contrary, should be understood to mean “at least one.”

[0095] The phrase “and / or,” as used herein in the specification and in the claims, should be understood to mean “either or both” of the components so conjoined, i.e., components that are conjunctively present in some cases and disjunctively present in other cases. Multiple components listed with “and / or” should be construed in the same fashion, i.e., “one or more” of the components so conjoined. Other components may optionally be present other than the components specifically identified by the “and / or” clause, whether related or unrelated to those components specifically identified. Thus, as a non-limiting example, a reference to “A and / or B”, when used in conjunction with open-ended language such as “comprising” can refer, in one embodiment, to A only (optionally including components other than B); in another embodiment, to B only (optionally including components other than A); in yet another embodiment, to both A and B (optionally including other components); etc.

[0096] As used herein in the specification and in the claims, “or” should be understood to have the same meaning as “and / or” as defined above. For example, when separating items in a list, “or” or “and / or” shall be interpreted as being inclusive, i.e., the inclusion of at least one, but also including more than one, of a number or list of components, and, optionally, additional unlisted items. Only terms clearly indicated to the contrary, such as “only one of’ or “exactly one of,” or, when used in the claims, “consisting of,” will refer to the inclusion of exactly one component of a number or list of components. In general, the term “or” as used herein shall only be interpreted as indicating exclusive alternatives (i.e. “one or the other but not both”) when preceded by terms of exclusivity, such as “either,” “one of,” “only one of,” or “exactly one of.” “Consisting essentially of,” when used in the claims, shall have its ordinary meaning as used in the field of patent law.

[0097] As used herein in the specification and in the claims, the phrase “at least one,” in reference to a list of one or more components, should be understood to mean at least one component selected from any one or more of the components in the list of components, but not necessarily including at least one of each and every component specifically listed within the list of components and not excluding any combinations of components in the list of components. This definition also allows that components may optionally be present other than the components specifically identified within the list of components to which the phrase“at least one” refers, whether related or unrelated to those components specifically identified. Thus, as a non-limiting example, “at least one of A and B” (or, equivalently, “at least one of A or B,” or, equivalently “at least one of A and / or B”) can refer, in one embodiment, to at least one, optionally including more than one, A, with no B present (and optionally including components other than B); in another embodiment, to at least one, optionally including more than one, B, with no A present (and optionally including components other than A); in yet another embodiment, to at least one, optionally including more than one, A, and at least one, optionally including more than one, B (and optionally including other components); etc.

[0098] In the claims, as well as in the specification above, all transitional phrases such as “comprising,” “including,” “carrying,” “having,” “containing,” “involving,” “holding,” “composed of,” and the like are to be understood to be open-ended, i.e., to mean including but not limited to. Only the transitional phrases “consisting of’ and “consisting essentially of’ shall be closed or semi-closed transitional phrases, respectively, as set forth in the United States Patent Office Manual of Patent Examining Procedures, Section 2111.03.

Claims

CLAIMS1. A method of stimulating cerebrospinal fluid flow, the method comprising: acquiring an electroencephalography (EEG) signal from a human subject while the human subject is asleep; predicting a target feature of the EEG signal; and delivering a stimulus to the human subject to coincide with the target feature of the EEG signal, the stimulus stimulating cerebrospinal fluid flow.

2. The method of claim 1, wherein the target feature of the EEG signal comprises at least one of: a peak of a slow wave; a slow wave coupled to another frequency band; a phase of EEG bandpower; an amplitude of EEG bandpower; or a wave-linked increase in spindle power.

3. The method of claim 2, wherein the target feature of the EEG signal comprises a peak of a slow wave.

4. The method of claim 3, wherein predicting the peak of the slow wave is based on an age of the human subject.

5. The method of claim 1, wherein the stimulus comprises an auditory stimulus.

6. The method of claim 5, wherein the auditory stimulus comprises pink noise and delivering the auditory stimulus comprises playing about 50 milliseconds of the pink noise.

7. The method of claim 1, wherein the stimulus comprises a vibration.

8. The method of claim 7, wherein the vibration is a bone conduction vibration.

9. The method of claim 1, wherein predicting the target feature of the EEG signal comprises using a recurrent neural network trained on previously collected EEG data.

10. The method of claim 9, wherein the previously collected EEG data comprises previously collected EEG data from the human subject.

11. The method of claim 9, wherein the previously collected EEG data comprises previously collected EEG data from human subjects with ages within 10 years of the age of the human subject.

12. The method of claim 1, wherein the delivering the stimulus to coincide with the target feature of the EEG signal comprises starting to deliver the stimulus within about 100 milliseconds of the target feature of the EEG signal.

13. The method of claim 12, wherein the delivering the stimulus to coincide with the target feature of the EEG signal comprises starting to deliver the stimulus within about 70 milliseconds of the target feature of the EEG signal.

14. The method of claim 1, wherein delivering the stimulus comprises delivering the stimulus for a duration of the target feature of the EEG signal.

15. The method of claim 1, wherein acquiring the EEG signal from the human subject while the subject is asleep comprises acquiring the EEG signal while the human subject is in non-rapid eye movement (NREM) sleep.

16. A device for stimulating cerebrospinal fluid flow in a subject, the device comprising: an electroencephalography (EEG) headset configured to be worn by the subject and to record an EEG signal from the subject while the subject is sleeping; an EEG recording machine operably coupled to the EEG headset and configured to receive the EEG signal from the EEG headset; a stimulus emitting device configured to emit a stimulus to the subject to coincide with a target feature of the EEG signal; and at least one processor operably coupled to the EEG recording machine and the stimulus emitting device and configured to control the stimulus emitting device using a recurrent neural network trained on EEG data acquired previously from the subject.

17. The device of claim 16, wherein the wherein the stimulus emitting device comprises at least one headphone and the stimulus comprises an auditory stimulus.

18. The device of claim 17, wherein the auditory stimulus comprises pink noise.

19. The device of claim 16, wherein the stimulus comprises a vibration.

20. A method of stimulating cerebrospinal fluid flow, the method comprising: acquiring an electroencephalography (EEG) signal from a human subject while the human subject is asleep; predicting a peak of a slow wave of the EEG signal using a recurrent neural network trained on previously collected EEG data; and playing an auditory stimulus to the human subject within about 100 milliseconds of the peak of the slow wave of the EEG signal, the auditory stimulus stimulating cerebrospinal fluid flow.

Images