Systems and methods for controlling a device based on detection of transitional vibration bursts or pseudo vibration bursts

Abstract

A system, method, and apparatus for controlling a device based on detection of transitional oscillation bursts or pseudo-oscillation bursts are disclosed. For example, the method can include detecting one or more transitional oscillation bursts or pseudo-oscillation bursts from ongoing neural signal recordings of a subject. The method can also include extracting one or more burst features from the one or more transitional oscillation bursts or pseudo-oscillation bursts detected during a detection period. The method can further include predicting a thought generated or recalled by the subject, or a change in a mental state induced by the subject, by applying at least one of a machine learning algorithm and a feature threshold to the one or more burst features extracted during the detection period. To control the device, an input command associated with the prediction can be sent to the device.

Description

Technical Field

[0001] Cross - Reference to Related Applications This application claims the benefit of U.S. Provisional Patent Application No. 63 / 480,746, filed on January 20, 2023, and U.S. Provisional Patent Application No. 63 / 365,999, filed on June 7, 2022, the entire contents of which are incorporated herein by reference.

[0002] The present disclosure generally relates to brain - computer interfaces, and more specifically to systems and methods for controlling a device based on the detection of transient oscillation bursts or pseudo - oscillation bursts.

Background Art

[0003] It has been found that people with movement limitations can use a brain - computer interface (BCI) to control personal electronic devices or computing devices, Internet of Things (IoT) devices, software, and moving vehicles. An effective BCI should enable people with all ranges of movement limitations, including those with severe movement limitations such as locked - in patients who can only control certain neural functions such as generating or recalling thoughts, to effectively control such devices. The goal of an effective BCI is to be able to extract features from neural signal recordings that correlate with the subject's distinct mental and / or physical thoughts, states, or physical conditions and to identify them for use in utilizing those thoughts, states, or physical conditions. However, there are several problems related to this objective.

[0004] First, neural activity sampled by sensors often results in noisy signals, where changes in such signals that correlate with a subject's mental or physical state are better observed when the signals are averaged over many trials of the subject generating or achieving the same mental or physical state. This technique is sometimes called "averaging." Repeating a task for averaging is also commonly used to create a template of the signal based on the average, and this template can later be compared to ongoing signals to achieve real-time identification and classification of the signals. The drawback of analyzing neural signals by averaging is that subjects are often required to repeatedly achieve the same mental or physical state in order for the BCI system to overcome the low signal-to-noise ratio. However, for any BCI system to be truly useful to a subject, the system must be able to detect characteristic changes in the subject's neural signals in real time or in ongoing recordings.

[0005] Second, the overall power of the vibration signal is often calculated over a detection period of several seconds to emphasize changes in the signal when the subject enters and exits a mental and / or physical state. However, since the power is calculated over this detection period, the BCI system often misses more subtle transient changes in the signal that may be characteristic of a subject's particular mental and / or physical thoughts, states, or situations.

[0006] Third, most conventional BCI systems can utilize only a few features extracted from the subject's neural signal recordings. A more robust BCI system should be able to access multiple features, and these multiple features can then be used to associate such detected features with the subject's various mental and / or physical states. Further, some conventional BCI systems can only make classification decisions based on a longer time period by considering the characteristics of the signal over a longer data window, and as a result, there are fewer examples for training the learning algorithm. A more robust BCI system should be able to make classification decisions based on shorter segments of data in order to achieve both faster classification and more training examples for the learning algorithm. Some BCI systems can apply more sophisticated algorithms to raw neural signals and overcome some of these limitations by having the algorithm determine the most beneficial aspects of the signal. However, such systems may require large amounts of data from each participant for training the algorithm to pick up those beneficial aspects of the neural signal. In many cases, such large amounts of data may not be available. In addition, the approach of using raw data as input to the classification algorithm carries the risk of using the noise source for classification instead of the actual brain signal because the algorithm often does not know the distinction between the brain signal and the noise source. Therefore, a more robust BCI system will utilize neuroscience knowledge to extract beneficial features from neural signals before applying the classification algorithm.

[0007] Given that researchers are beginning to examine the importance of events such as transient changes or bursts that can be detected from ongoing or real-time neural recordings of subjects, the inability to pick up on finer transient changes in a subject's neural signals is a problem. For example, some researchers have discovered that spontaneous cortical beta rhythms appear as discontinuous beta events that depend not necessarily on rhythmic input but on the relative timing and strength of synchronous proximal and distal drives (see Sherman et al., "Neural mechanisms of transient neocortical beta rhythms: converging evidence from humans, computational modeling, monkeys, and mice." Proc Natl Acad Sci USA 113(2016):E4885-E4894). Other researchers have shown that differences in the incidence of beta events can be used to predict the detection of stimuli at perceptual thresholds, and that undetected trials are more likely to have beta events within about 200 milliseconds before the stimulus (see Shin et al., "The rate of transient beta frequency events predicts behavior across tasks and species." Elife 6(2017):e29086). Researchers have also used trial-by-trial analysis and discovered that short bursts of gamma-band activity are associated with the encoding and reactivation of sensory information at recording sites related to spikes that reflect items to be "remembered" (see Lundqvist et al., "Gamma and beta bursts underlie working memory." Neuron 90.1(2016):152-164).Furthermore, some researchers have pointed out that transient events may form more complex repetitive sequences of activation with millisecond accuracy, suggesting that the relevant information may be encoded by subtle time differences or cascades of transient events across the brain (see Felsenstein et al., "Decoding multimodal behavior using time differences of MEG events." arXiv preprint: 1901.08093 (2019), Tal, Idan, and Moshe Abeles. "Temporal accuracy of human cortico-cortical interactions." Journal of Neurophysiology 115.4 (2016), pp. 1810 - 1820, and Tal, I. and M. Abeles. "Imaging the spatiotemporal dynamics of cognitive processes at high temporal resolution." Neural Computation 30.3 (2018): 610 - 630).

Prior Art Documents

Patent Documents

[0008]

Patent Document 1

Patent Document 2

Patent Document 3

Patent Document 4

Patent Document 5

Patent Document 6

Patent Document 7

[0009] [Non-Patent Document 1] Sherman et al., "Neural mechanisms of transient neocortical beta rhythms: converging evidence from humans, computational modeling, monkeys, and mice." Proc Natl Acad Sci USA 113(2016):E4885-E4894 [Non-Patent Document 2] Shin et al., "The rate of transient beta frequency events predicts behavior across tasks and species." Elife 6(2017):e29086 [Non-Patent Document 3] Lundqvist et al., "Gamma and beta bursts underlie working memory." Neuron 90.1(2016):152-164 [Non-Patent Document 4] Felsenstein et al., "Decoding multimodal behavior using time differences of MEG events." arXiv preprint:1901.08093 (2019) [Non-Patent Document 5] Tal, Idan, and Moshe Abeles. "Temporal accuracy of human cortico - cortical interactions." Journal of Neurophysiology 115.4 (2016), pp. 1810 - 1820 [Non - Patent Document 6] Tal, I., and M. Abeles. "Imaging the spatiotemporal dynamics of cognitive processes at high temporal resolution." Neural Computation 30.3 (2018): 610 - 630 [Summary of the Invention] [Problems to be Solved by the Invention]

[0010] Therefore, there is a need for improvement in the field of BCI that utilizes these new findings regarding the importance of events such as transient bursts that can be detected from ongoing neural signals of a subject. Further, any such improved BCI system should also address the aforementioned drawbacks of conventional BCI systems. Such a system should enable patients with severe motor limitations to maintain or retain their independence even if such patients can only control their own thoughts or mental states. [Means for Solving the Problems]

[0011] A system and method for controlling a device based on the detection of transient oscillation bursts or pseudo - oscillation bursts are disclosed.

[0012] In some aspects, a method of controlling a device is disclosed. The method includes detecting, by a recording device, one or more transient oscillatory bursts or pseudo-oscillatory bursts from ongoing or real-time neural signal recordings of a subject, wherein at least a portion of the one or more transient oscillatory bursts or pseudo-oscillatory bursts are generated in response to thoughts generated or recalled by the subject or a change in a mental state induced by the subject; extracting, using one or more processors of a computing device communicatively coupled to the recording device, one or more burst features from the one or more transient oscillatory bursts or pseudo-oscillatory bursts detected during a detection period; predicting, using the one or more processors, a thought generated or recalled by the subject or a change in a mental state induced by the subject by applying at least one of a machine learning algorithm and a feature threshold to the one or more burst features extracted during the detection period; and transmitting, to the device, an input command associated with the prediction to control the device.

[0013] In some embodiments, ongoing or real-time neural signal recording of a subject may be performed by recording raw electrical signals from the subject's brain using a recording device, and the step of detecting one or more transient oscillation bursts or pseudo-oscillation bursts may include filtering the raw electrical signals in one or more desired frequency bands using one or more frequency decomposition methods such as at least one of a band-pass filter and a wavelet convolution; for each of the desired frequency bands, converting the voltage values of the filtered raw electrical signals into magnitude or power-related values; applying at least one of a power threshold for each of the magnitude or power-related values for each of the desired frequency bands and a duration threshold for each of the desired frequency bands; and for each of the desired frequency bands, identifying one of the transient oscillation bursts or pseudo-oscillation bursts in response to at least one of a magnitude or power-related value exceeding the power threshold and a duration of one of the raw electrical signals exceeding the duration threshold.

[0014] In some embodiments, at least one of the power threshold and the duration threshold may be selected based on at least one training session performed on the subject, the at least one training session instructing or prompting the subject to generate or recall thoughts or to induce a change in the subject's mental state; after prompting the subject to generate or recall thoughts, recording raw electrical signals from the subject's brain using a recording device; filtering the raw electrical signals in one or more desired frequency bands using one or more frequency decomposition methods such as at least one of a band-pass filter and a wavelet convolution; for each of the desired frequency bands, converting the voltage values of the filtered raw electrical signals into power values; and selecting at least one of a power threshold and a duration threshold for each of the desired frequency bands to distinguish transient oscillation bursts or pseudo-oscillation bursts from background noise.

[0015] In some aspects, a system for controlling a device is disclosed. The system includes a recording device configured to capture ongoing or real-time neural signal recordings of a subject, and a computing device having one or more processors communicatively coupled to the recording device. In these aspects, the one or more processors of the computing device are to detect one or more transient oscillation bursts or pseudo-oscillation bursts from the ongoing or real-time neural signal recordings of the subject, where the one or more transient oscillation bursts or pseudo-oscillation bursts are generated in response to thoughts generated or recalled by the subject, or changes in mental states induced by the subject; extract one or more burst features from the one or more transient oscillation bursts or pseudo-oscillation bursts detected during a detection period; predict a thought generated or recalled by the subject, or a change in a mental state induced by the subject, by applying at least one of a machine learning algorithm and a feature threshold to the one or more burst features extracted during the detection period; and transmit an input command associated with the prediction to the device to control the device.

[0016] In some embodiments, the recording device may be configured to capture real-time neural signal recordings of a subject by recording raw electrical signals from the subject's brain. In these embodiments, one or more processors of the computing device use one or more frequency decomposition methods to filter the raw electrical signals in one or more desired frequency bands, and for each of the desired frequency bands, convert the voltage values of the filtered raw electrical signals into magnitude or power-related values, apply at least one of a power threshold for the magnitude or power-related value for each of the desired frequency bands, or a duration threshold for each of the desired frequency bands, and for each of the desired frequency bands, identify one of a transient oscillation burst or a pseudo-oscillation burst in response to at least one of a magnitude or power-related value exceeding the power threshold and a filtered raw electrical signal exceeding the duration threshold.

[0017] In some embodiments, at least one of the power threshold and the duration threshold may be selected based on at least one training session performed on the subject using the computing device or another device and the recording device. The computing device or other device may be configured to instruct or prompt the subject to generate or recall thoughts or to induce a change in the subject's mental state. The recording device may be configured to record raw electrical signals from the subject's brain after the subject is prompted to generate or recall thoughts. One or more processors of the computing device use one or more frequency decomposition methods to filter the raw electrical signals in one or more desired frequency bands, and for each of the desired frequency bands, convert the voltage values of the filtered raw electrical signals into power values, To distinguish a transient oscillation burst or a pseudo-oscillation burst from background noise, it can be programmed to perform selecting at least one of a power threshold value and a duration threshold value to be applied for each of the desired frequency bands.

[0018] In some embodiments, the desired frequency band includes a frequency band between 0.1 Hz and 32 kHz (and, for example, between 4 Hz and 400 Hz).

[0019] In some embodiments, the desired frequency band includes at least one of a beta frequency band, a gamma frequency band, and a high gamma frequency band.

[0020] In some embodiments, one or more burst features include a burst rate, which can be calculated by dividing a burst count by the length of a detection period.

[0021] In some embodiments, the burst count can be calculated by summing all of the transient oscillation bursts or pseudo-oscillation bursts detected across all electrodes or a subset of electrodes of a recording device during a detection period.

[0022] In some embodiments, the feature threshold value can be a burst rate threshold value.

[0023] In some embodiments, the burst rate threshold value can be a median value of the burst rates calculated from previous detection periods.

[0024] In some embodiments, one or more burst features include at least one of a burst count, a burst rate, a burst band frequency or frequency distribution, an inter-burst interval length, a burst timing or timing pattern, an average burst duration, a burst waveform, and any variations thereof.

[0025] In some embodiments, the machine learning algorithm can be a neural network.

[0026] In some embodiments, the neural network can be a recurrent neural network.

[0027] In some embodiments, the recurrent neural network can be a long short-term memory (LSTM) neural network.

[0028] In some embodiments, the feature threshold can be a static threshold.

[0029] In some embodiments, the feature threshold can be a dynamic threshold that is adjusted over time by a computing device. In certain embodiments, the feature threshold may need to exceed the threshold for a particular duration.

[0030] In some embodiments, the detection period can be between 1 millisecond and 100 milliseconds. In other embodiments, the detection period can be between 10 milliseconds and 100 milliseconds.

[0031] In some embodiments, the device can be at least one of a personal computing device, an Internet of Things (IoT) device, and a moving vehicle.

[0032] In some embodiments, the device can be a personal computing device, and the input command can be a command to initiate a click of the cursor of the personal computing device.

[0033] In some embodiments, the subject's thought can be a thought generated or recalled by the subject to move one or more body parts of the subject.

[0034] In some embodiments, the thought can be generated or recalled by the subject without being prompted to do so, such that the control of the device is performed asynchronously.

[0035] In some embodiments, thoughts can be generated or recalled by a subject in response to being prompted to do so such that control of the device is performed synchronously.

[0036] In some embodiments, the recording device can be a non-invasive recording device. For example, the recording device can be an electroencephalogram (EEG) recording device.

[0037] In some embodiments, the recording device can be an invasive recording device.

[0038] In some embodiments, the recording device can be an intravascular recording device configured to be implanted within a vein or sinus of the subject's brain.

[0039] In some embodiments, the intravascular recording device can be an electrode array that includes a plurality of electrodes held by an expandable stent or scaffold.

[0040] In some embodiments, one or more transient oscillation bursts or pseudo-oscillation bursts can be detected using electrodes held by an expandable stent or scaffold, and the method further includes applying a weighting factor to one or more electrodes of the expandable stent or scaffold such that a transient oscillation burst or pseudo-oscillation burst detected at one or more electrodes is weighted more than a transient oscillation burst or pseudo-oscillation burst detected at another electrode of the expandable stent or scaffold.

[0041] In some embodiments, the recording device can be an implanted microelectrode array (MEA). For example, the recording device can be a Utah microelectrode array or a Michigan microelectrode array.

[0042] In some embodiments, the recording device can be an electrode array that can be implanted on the surface of the brain or cortex. For example, the recording device can be a electrocorticogram (eCoG) electrode array.

[0043] In some embodiments, the recording device can be a thin-film electrode array or thin-film microelectrodes.

[0044] In some embodiments, the machine learning algorithm can be trained using previous predictions made by the machine learning algorithm and burst features extracted from previous detection periods to enhance the predictions made by the machine learning algorithm.

Brief Description of the Drawings

[0045]

Figure 1A

Figure 1B

Figure 1C

Figure 1D

Figure 2A

Figure 2B

Figure 2C

Figure 2D

Figure 3A

Figure 3B

Figure 3C

Figure 3D

Figure 4

Figure 5A

Figure 5B

Figure 6

Figure 7A

Figure 7B

[0046] FIG. 1A shows one embodiment of a brain-computer interface (BCI) system 100 configured to control device 10 based on detection of transient oscillatory bursts or pseudo-oscillatory bursts 302 from ongoing or real-time neural signal recordings of a subject (see FIG. 3A). The BSI system 100 disclosed herein can be used by a subject (e.g., an individual with mobility problems) to control a device 10 such as a personal computing device, an Internet of Things (IoT) device, a moving vehicle (e.g., a wheelchair), or a combination thereof.

[0047] The BSI system 100 can transmit an input command 312 (see FIG. 3A) to the device 10 to control the device 10. For example, if the device 10 is a personal computing device, the input command 312 can be a command to initiate a click of the cursor of the personal computing device. In other embodiments, the input command 312 can be a command to steer or operate a moving vehicle (e.g., a wheelchair) carrying the subject.

[0048] The system 100 can include a recording device 102 (see FIG. 1B) and a computing device 104. The recording device 102 can be configured to record the brain activity of the subject. In some embodiments, the recording device 102 can be an invasive recording device 102 configured to be implanted within the cerebral blood vessels 106 of the subject.

[0049] For example, the recording device 102 can be a stent electrode array 108 configured to be implanted within the cerebral blood vessels 106 of the subject (see FIG. 1B). As a more specific example, the recording device 102 can be implanted within the cortex or cerebral veins or sinuses of the subject.

[0050] In some embodiments, the recording device 102 can be an implantable microelectrode array (MEA). For example, the recording device 102 can be a Utah microelectrode array or a Michigan microelectrode array.

[0051] In some embodiments, the recording device 102 can be a thin-film electrode array or can be composed of thin-film electrodes.

[0052] In some embodiments, the recording device 102 can be an electrode array that can be implanted on the surface of the brain or the cortex. For example, the recording device 102 can be an electrocorticogram (eCoG) electrode array (see, e.g., FIG. 2D).

[0053] In some embodiments, the recording device 102 can be a non-invasive recording device such as an electroencephalogram (EEG) recording device, a helmet, or other types of headgear.

[0054] FIG. 1B shows that the stent electrode array 108 can include a plurality of electrodes 110 attached, fixed, or otherwise coupled to an expandable stent 112 or an outer or radially outer portion of a scaffold that functions as an intravascular carrier for the electrode array. For example, the electrodes 110 can be disposed along the wall, rings, or filaments that make up the expandable stent 112.

[0055] In some embodiments, the recording device 102 can typically include from 8 to 24 electrodes. For example, the recording device 102 can include 16 electrodes. In other embodiments, the recording device 102 can include from 24 to 64 electrodes 110.

[0056] In some embodiments, some of the filaments of the expandable stent 112 can be made of a shape memory alloy. For example, some of the filaments of the expandable stent 112 can be made of nitinol or nitinol wire. The filaments of the expandable stent 112 can also be made, in part, of stainless steel, gold, platinum, nickel, titanium, tungsten, aluminum, nickel-chromium alloy, gold-palladium-rhodium alloy, chromium-nickel-molybdenum alloy, iridium, rhodium, or combinations thereof.

[0057] In an alternative embodiment, some of the filaments of the expandable stent 112 can also be made of a shape memory polymer.

[0058] The electrode 110 can be made, in part, of platinum, platinum black, gold, iridium, palladium, rhodium, or alloys or composites thereof (e.g., gold-palladium-rhodium alloy or composite). In certain embodiments, the electrode 110 can be made of a metal alloy or composite having a high charge injection capacity (e.g., platinum-iridium alloy or composite).

[0059] The electrode 110 can be formed as a circular disk having a disk diameter of from about 100 μm to 1.0 mm. In other embodiments, the electrode 110 can have a disk diameter between 1.0 mm and 1.5 mm. In other embodiments, the electrode 110 can be cylindrical, spherical, cuff-shaped, ring-shaped, partially ring-shaped (e.g., C-shaped), or semi-cylindrical.

[0060] In other embodiments, the stent electrode array 108 can be any of the stents, scaffolds, stent electrodes, or stent electrode arrays disclosed in U.S. Patent Publication No. 2021 / 0365117, U.S. Patent Publication No. 2021 / 0361950, U.S. Patent Publication No. 2020 / 0363869, U.S. Patent Publication No. 2020 / 0078195, U.S. Patent Publication No. 2020 / 0016396, U.S. Patent Publication No. 2019 / 0336748, U.S. Patent Publication No. 2014 / 0288667, U.S. Patent No. 10,575,783, U.S. Patent No. 10,485,968, U.S. Patent No. 10,729,530, and U.S. Patent No. 10,512,555, the entire contents of which are incorporated herein by reference.

[0061] When the recording device 102 (e.g., the stent electrode array 108) is implanted within the subject's cerebrovascular 106, each of the electrodes 110 of the recording device 102 can be configured to read or record the electrical activity of neurons within the vicinity of each electrode 110. The electrical activity of the neurons can be recorded as raw electrical signals. As discussed in more detail in a later section, the raw electrical signals can be filtered and processed to detect one or more transient oscillation bursts or pseudo-oscillation bursts 302.

[0062] The raw electrical signal can be band - divided according to its frequency. For example, the desired frequency band includes the frequency band between 0.1 Hz and 32 kHz. The transient oscillation bursts or pseudo - oscillation bursts 302 detected from these signals can also be associated with such frequency bands. For example, the transient oscillation bursts or pseudo - oscillation bursts 302 may be called beta bursts or beta - band bursts when these bursts are obtained from signals within the beta oscillation band (having frequencies of about 15 - 35 Hz). In addition, the transient oscillation bursts or pseudo - oscillation bursts 302 may be called gamma bursts or gamma - band bursts when these bursts are obtained from signals within the gamma oscillation band (having frequencies of about 45 - 100 Hz). Further, the transient oscillation bursts or pseudo - oscillation bursts 302 may be called alpha bursts or alpha - band bursts when these bursts are obtained from signals within the alpha oscillation band (having frequencies of about 7 Hz - 12 Hz). Further, the transient oscillation bursts or pseudo - oscillation bursts 302 may be called theta bursts or theta - band bursts when these bursts are obtained from signals within the theta oscillation band (having frequencies of about 4 Hz - 7 Hz).

[0063] In some embodiments, the recording device 102 can be implanted into the veins or sinuses of the subject's brain or cortex. For example, the recording device 102 can be implanted into the superior sagittal sinus, inferior sagittal sinus, Sigmoid sinus, transverse sinus, straight sinus, Labbe's vein, Trolard's vein, Sylvian vein, Rolandic vein and other superficial cerebral veins, Rosenthal's vein, Galen's vein and other deep cerebral veins, the superior thalamostriate vein, the inferior thalamostriate vein, or the internal cerebral vein, the central sulcus vein, the posterior central sulcus vein, or the anterior central sulcus vein. In certain embodiments, the recording device 102 can be implanted into a blood vessel extending through the subject's hippocampus or tonsil.

[0064] Alternatively or additionally, the recording device 102 can be an implantable microelectrode array (MEA). For example, the recording device 102 can be a Utah microelectrode array or a Michigan microelectrode array.

[0065] The recording device 102 can also be or be composed of a thin-film electrode array.

[0066] FIG. 1C shows that a communication conduit 114 (e.g., a lead wire) can connect an implantable recording device 102 (e.g., a stent electrode array 108) to a telemetry unit 116 communicatively coupled to a computing device 104. Alternatively, the communication conduit 114 can directly connect the implantable recording device 102 to the computing device 104.

[0067] The communication conduit 114 can be a biocompatible lead wire or cable. When the recording device 102 is a stent electrode array 108 deployed within a subject's cerebrovascular 106 (e.g., the superior sagittal sinus), the communication conduit 114 can pass through one or more cerebrovasculars and extend through the wall of a vein (e.g., the internal jugular vein) coupled to at least one of the subject's cerebrovasculars. The communication conduit 114 can then tunnel under the subject's skin to the area (e.g., under the pectoralis major muscle) where the telemetry unit 116 is implanted.

[0068] FIG. 1D shows an enlarged view of an embodiment of the telemetry unit 116. In some embodiments, the telemetry unit 116 can be configured to transmit signals received from the recording device 102 to the computing device 104 for processing and analysis. The telemetry unit 116 can also function as a communication hub between the recording device 102 and the computing device 104.

[0069] In certain embodiments, the telemetry unit 116 can be an internal telemetry unit 116 that can be implanted under the skin of a subject. For example, the telemetry unit 116 can be implanted within the chest region of the subject or within the infraclavicular space.

[0070] In other embodiments, the telemetry unit 116 can be an external telemetry unit 116 that is not implanted within the subject. In these embodiments, the communication conduit 114 can extend through the skin of the subject to connect to the telemetry unit 116. In additional embodiments, the telemetry unit 116 can comprise both an implantable portion and an external portion.

[0071] In some embodiments, the telemetry unit 116 can transmit data or signals to, or receive data or commands from, the computing device 104 via a wired connection. In other embodiments, the telemetry unit 116 can transmit data or signals to, or receive data or commands from, the computing device 104 via a wireless communication protocol such as Bluetooth™, Bluetooth Low Energy (BLE), ZigBee™, WiFi, or combinations thereof.

[0072] As will be discussed in more detail in a later section, one or more processors of computing device 104 may be programmed to detect one or more transient oscillation bursts or pseudo-oscillation bursts 302 from ongoing or real-time neural signal recordings of a subject captured by recording device 102. At least some of the one or more transient oscillation bursts or pseudo-oscillation bursts 302 may be generated in response to thoughts generated or recalled by the subject, or changes in mental states induced by the subject. Further, one or more processors of computing device 104 may also be programmed to extract one or more burst features 400 (see FIG. 4) from the one or more transient oscillation bursts or pseudo-oscillation bursts 302 detected within a detection period 404 or detection window. Further, one or more processors of computing device 104 may be programmed to predict thoughts generated or recalled by the subject, or changes in the mental state of the subject, by applying a feature threshold 310 and / or a machine learning algorithm 308 (see FIG. 3A) (e.g., a deep learning algorithm) to the one or more burst features 400 extracted during the detection period 404. Next, one or more processors of computing device 104 may be able to transmit an input command 312 associated with the prediction to device 10 in order to control device 10.

[0073] In some embodiments, one or more processors of the telemetry unit 116 may be programmed to detect one or more transient oscillations or spurious oscillations 302 from ongoing or real-time neural signal recordings of a subject captured by the recording device 102. At least one or more of the transient oscillations or spurious oscillations 302 may be generated in response to thoughts generated or recalled by the subject, or changes in mental states induced by the subject. Further, one or more processors of the telemetry unit 116 may also be programmed to extract one or more burst features 400 from one or more transient oscillation bursts or spurious oscillation bursts 302 detected within the detection period 404. Further, one or more processors of the telemetry unit 116 may be programmed to predict thoughts generated or recalled by the subject, or changes in mental states induced by the subject, by applying a feature threshold 310 and / or a machine learning algorithm 308 to one or more burst features 400 extracted within the detection period 404. Then, one or more processors of the telemetry unit 116 may be able to send an input command 312 associated with the prediction to the computing device 104 or directly to the device 10 to control the device 10.

[0074] Transient oscillation bursts or spurious oscillation bursts 302 may be generated in response to thoughts generated, recalled, or evoked by the subject, or changes in mental states induced by the subject.

[0075] In some embodiments, the thought can be a thought generated or recalled by a subject to instruct the device 10 (including its components or software applications running thereon). As a more specific example, the thought generated or recalled by the subject can be a thought to move a cursor displayed on the display of a personal computing device functioning as the device 10. As another example, the thought generated or recalled by the subject can be a thought to move a moving vehicle (e.g., a wheelchair carrying the subject) functioning as the device 10. In these examples, the thought may be referred to as a task-related thought.

[0076] As an additional example, the change in the mental state induced by the subject can include the subject concentrating on the cursor or concentrating on the moving vehicle. In these examples, the change in the mental state may be referred to as a task-related mental state change.

[0077] In other embodiments, the thought can be a thought generated or recalled by the subject that is unrelated to or detached from instructing the device 10 (including any of its components or software applications running thereon). As a more specific example, the thought generated or recalled by the subject can be a thought to move one or more body parts of the subject (e.g., hands, fingers, ankles, feet, toes, legs, arms, head, etc.). Also, for example, the thought generated or recalled by the subject can be a thought to tense or relax one or more body parts of the subject (e.g., to tense or relax one or more muscles or muscle groups of the subject). In these examples, the thought may be referred to as a task-unrelated thought.

[0078] Furthermore, the change in mental state induced by the subject can include the subject concentrating attention on one or more body parts of the subject, or concentrating attention on tasks other than instructing the device 10 (including any component thereof or the software application running thereon). In these examples, the change in mental state may be referred to as a task-irrelevant mental state change.

[0079] As additional examples, the change in mental state induced by the subject can include the subject concentrating attention on the cursor, or concentrating attention on a moving vehicle. In these examples, the change in mental state may be referred to as a task-related mental state change.

[0080] FIG. 2A shows another embodiment of the implantable recording device 102 as a coiled wire 200 with a plurality of electrodes 110. The coiled wire 200 can function as an intravascular carrier for the electrodes 110 and can be used in blood vessels that are too small to accommodate the stent electrode array 108.

[0081] The coiled wire 200 can be a biocompatible wire or micro-wire configured to wind itself into a coiled pattern or a substantially helical pattern. The electrodes 110 can be arranged such that the electrodes 110 are scattered along the length of the coiled wire 200. More specifically, the electrodes 110 can be attached, fixed, or otherwise coupled to discrete points along the length of the coiled wire 200.

[0082] The electrodes 110 can be separated from each other such that no two electrodes 110 are within a predetermined separation distance (e.g., at least 10 μm, at least 100 μm, or at least 1.0 mm) of each other. In some embodiments, the coiled wire 200 can hold from 8 to 24 electrodes. For example, the coiled wire 200 can hold 16 electrodes. In other embodiments, the coiled wire 200 can hold from 24 to 64 electrodes 110.

[0083] In some embodiments, the wire 200 can be configured to automatically wind itself into a coiled configuration (e.g., a helical pattern) when the wire 200 is deployed from the delivery catheter. For example, the coiled wire 200 can automatically achieve its coiled configuration via shape memory when the delivery catheter or sheath is retracted. The coiled configuration or shape can be a pre-set shape or shape memory shape of the wire 200 before the wire 200 is introduced into the delivery catheter. The pre-set shape or pre-trained shape can be made larger than the diameter of the expected deployed or implanted blood vessel in order to enable the radial force exerted by the coil to fix or position the coiled wire 200 at a predetermined position within the deployed or implanted blood vessel.

[0084] The wire 200 can be made in part of a shape memory alloy, a shape memory polymer, or a combination thereof. For example, the wire 200 can be made in part of nitinol (e.g., a nitinol wire). The wire 200 can also be made in part of stainless steel, gold, platinum, nickel, titanium, tungsten, aluminum, a nickel-chromium alloy, a gold-palladium-rhodium alloy, a chromium-nickel-molybdenum alloy, iridium, rhodium, or a combination thereof.

[0085] FIG. 2B shows yet another embodiment of the implantable recording device 102 as a fixed wire 202 with a plurality of electrodes 110. The fixed wire 202 can function as an intravascular carrier for the electrodes 110 and can be used in blood vessels that are too small to accommodate either the coiled wire 200 or the stent electrode array 108.

[0086] The fixation wire 202 can comprise a biocompatible wire or micro-wire attached or otherwise coupled to an anchor or another type of intravascular fixation mechanism. FIG. 2B shows that the fixation wire 202 can comprise a barbed anchor 204, a radially expandable anchor 206, or a combination thereof (both the barbed anchor 204 and the radially expandable anchor 206 are shown in FIG. 2B by dashed or dash-dot lines). In some embodiments, the barbed anchor 204 can be disposed at the distal end of the fixation wire 202. In other embodiments, the barbed anchor 204 can be disposed along one or more sides of the wire or micro-wire. The barbs of the barbed anchor 204 can fix or tether the fixation wire 202 to the implantation site within the subject. The radially expandable anchor 206 can be a segment of wire or micro-wire formed as a coil or loop. The coil or loop can be sized such that the coil or loop expands against the lumen wall such that the coil or loop conforms to the vascular lumen and fixes the fixation wire 202 to the intravascular implantation site. For example, the coil or loop can be sized to be larger than the diameter of the expected deployed or implanted blood vessel such that the radial force exerted by the coil or loop enables the fixation wire 202 to be fixed or positioned at a predetermined location within the deployed or implanted blood vessel.

[0087] The electrodes 110 of the fixation wire 202 can be dispersed along the length of the fixation wire 202. More specifically, the electrodes 110 can be attached, fixed, or otherwise coupled at discrete points along the length of the fixation wire 202. The electrodes 110 can be separated from each other such that two electrodes 110 are not within a predetermined separation distance (e.g., at least 10 μm, at least 100 μm, or at least 1.0 mm) of each other.

[0088] In some embodiments, the fixed wire 202 can hold from 8 to 24 electrodes. For example, the fixed wire 202 can hold 16 electrodes. In other embodiments, the fixed wire 202 can hold from 24 to 64 electrodes 110.

[0089] FIG. 2B shows a fixed wire 202 having only one barbed anchor 204 and one radially expandable anchor 206, but it is contemplated that according to the present disclosure, the fixed wire 202 can comprise multiple barbed anchors 204 and / or radially expandable anchors 206.

[0090] FIG. 2C shows that in another embodiment of the system 100, the recording device 102 can be a non-invasive device such as an electroencephalogram (EEG) recording device 208. The EEG device 208 can be a head-mounted EEG device. For example, the EEG device 208 can be an EEG cap or an EEG visor configured to be worn by a subject. The EEG device 228 can comprise a plurality of non-invasive electrodes 210 configured to contact the scalp of the subject.

[0091] The EEG device 208 can comprise from 8 to 24 electrodes 210. For example, the EEG device 208 can comprise 16 electrodes 210. In other embodiments, the EEG device 208 can comprise from 24 to 64 electrodes 210.

[0092] The EEG device 208 can be used as part of the system 100 to detect one or more transient oscillation bursts or pseudo-oscillation bursts 302 similar to those recorded by the implantable recording device 102. For example, the EEG device 208 can capture ongoing or real-time neural signal recordings of a subject by recording raw electrical signals from the subject's brain.

[0093] FIG. 2D shows that in yet another embodiment of system 100, recording device 102 can be an electrocorticogram (ECoG) device 212 (also referred to as an intracranial EEG device). The ECoG device 212 can be a flexible or stretchable electrode mesh or one or more electrode patches that are implanted or disposed on the surface of the subject's brain. The electrode mesh or electrode patches can each comprise a plurality of electrodes 214 disposed on the mesh or patch.

[0094] The ECoG device 212 can comprise several electrodes 214. For example, the ECoG device 212 can comprise from 8 to 24 electrodes 214. For example, the ECoG device 212 can comprise 16 electrodes 214. In other embodiments, the ECoG device 212 can comprise from 24 to 64 electrodes 214.

[0095] In some embodiments, the recording device 102 can be an implanted microelectrode array (MEA). For example, the recording device 102 can be a Utah microelectrode array or a Michigan microelectrode array.

[0096] In some embodiments, the recording device 102 can be a thin-film microelectrode array or can be composed of thin-film microelectrodes.

[0097] FIG. 3A shows an exemplary method of controlling a device using certain components of system 100 disclosed herein. The method can include recording raw electrical or neural signals 300 from the subject's brain using recording device 102.

[0098] For example, when the recording device 102 is an implantable recording device such as the stent electrode array 108 of FIG. 1B, the recording device 102 can typically include 16 to 64 electrodes 110. Each of the electrodes 110 can record neural signals 300 in one or more frequency bands. This combination of the electrodes 110 and the frequency band means that even neural signal recordings that last only 3 to 5 seconds can potentially include thousands or more transient oscillation bursts or pseudo-oscillation bursts 302.

[0099] The neural signals 300 captured by the electrodes 110 of the recording device 102 can be transmitted to the computing device 104 via the telemetry unit 116.

[0100] As shown in FIG. 3A, the computing device 104 can include a decoder module 304 and a classification layer 306 or module. The decoder module 304 can be configured to filter the raw electrical signals, convert the filtered raw electrical signals into magnitude or power-related values, and identify transient oscillation bursts or pseudo-oscillation bursts 302 each time one of the magnitude or power-related values exceeds a power threshold. The decoder module 304 can also be configured to extract one or more burst features 400 from one or more transient oscillation bursts or pseudo-oscillation bursts within a detection period 404 (see FIG. 4).

[0101] The classification layer 306 can be configured to predict thoughts generated or recalled by the subject, or changes in mental states induced by the subject, by applying at least one of a machine learning algorithm 308 (e.g., a deep learning algorithm) and a feature threshold 310 to one or more burst features 400 extracted within the detection period 404.

[0102] As will be discussed in more detail in connection with FIG. 3B, decoder module 304 can comprise several decoder submodules. The decoder module 304 can be configured to filter the raw electrical signal in one or more desired frequency bands (e.g., beta band, gamma band, alpha band, theta band, etc.) using one or more software filters or wavelet convolutions. For example, the decoder module 304 can comprise a decoder submodule or instruction for filtering the raw electrical signal in one or more desired frequency bands using one or more bandpass filters.

[0103] The decoder module 304 can also convert the voltage value of the filtered raw electrical signal into a power value for each of the desired frequency bands. Further, the decoder module 304 can apply a power threshold to the magnitude or power-related value for each of the desired frequency bands. Then, the decoder module 304 can identify one of the transient oscillation bursts or pseudo-oscillation bursts 302 each time one of the magnitude or power-related values exceeds the power threshold for each of the desired frequency bands.

[0104] The decoder module 304 can also comprise a submodule for extracting one or more burst features 400 (see FIG. 4) from one or more of the transient oscillation bursts or pseudo-oscillation bursts 302 within the detection period 404. Then, using the extracted burst features 400, the classification layer 306 can make a prediction regarding thoughts generated or recalled by the subject, or a change in the mental state of the subject, by applying at least one of the machine learning algorithm 308 and the feature threshold 310 to the burst features 400 extracted within the detection period 404.

[0105] In some embodiments, the machine learning algorithm 308 can be a neural network. For example, the machine learning algorithm 308 can be a recurrent neural network. As a more specific example, the classification layer 306 can use a recurrent neural network such as an LSTM network to make predictions regarding thoughts generated or recalled by the subject, or changes in mental states induced by the subject, by feeding the extracted burst features into the long short-term memory (LSTM) network.

[0106] In other embodiments, the classification layer 306 can make predictions regarding thoughts generated or recalled by the subject, or changes in mental states induced by the subject, by applying a feature threshold 310 (see also FIG. 5A) to the burst features 400 extracted during the detection period 404.

[0107] Next, the computing device 104 can send an input command 312 associated with the prediction made by the classification layer 306 to the device 10 to control the device 10.

[0108] In some embodiments, the device 10 can be a device separate from the computing device 104. In these and other embodiments, the input command 312 can be a command for controlling one or more peripheral components or hardware components of the device 10, a software application running on the device 10, or graphic elements (e.g., cursor, pointer, caret, etc.) displayed on the display of the computing device 10.

[0109] In other embodiments, the computing device 104 can generate input commands 312 associated with predictions to control one or more peripheral or hardware components of the device 104, software applications executed on the device 104, or graphic elements (e.g., cursor, pointer, caret, etc.) displayed on the display of the computing device 104.

[0110] In some embodiments, modules of the computing device 104 (e.g., decoder module 304, classification layer 306, etc.) can be described using the Java (trademark) programming language, Python (trademark) programming language, C / C++ programming language, JavaScript programming language, Ruby (trademark) programming language, Matlab programming language, C# programming language, or combinations thereof.

[0111] Figures 3B - 3D show exemplary sub - modules of the decoder module 304. Each sub - module (Figures 3B - 3D) can represent a different possible pipeline of the decoder module 304.

[0112] Figure 3B shows one embodiment of a sub - module that can extract vibration burst features from a single event or multiple events using wavelet transform and pass these features to a machine - learning algorithm for classification. Figure 3C shows another embodiment of a sub - module that can extract the vibration burst count by applying several finite impulse response (FIR) filters to the signal and detect power values exceeding a predefined threshold for vibration burst detection. For classification, another threshold can be applied to the detected bursts. The latter threshold can be dynamic and can require the signal to exceed the threshold for a specific duration. Figure 3C shows yet another embodiment of a sub - module that uses both threshold classification and a machine - learning algorithm.

[0113] FIG. 4 shows an exemplary burst feature 400 that can be extracted from a 3 - second neural signal recording of a subject, where the power values detected in various frequency bands are converted into a spectrogram depicted using a color gradient. Depicted below the spectrogram is a burst timing chart showing the timing of some transient oscillation bursts or pseudo - oscillation bursts 302 detected at one of the electrodes 110 of the recording device 102 or at one burst frequency during this period, and the exemplary waveforms of five such bursts 302.

[0114] As discussed above, the recording device 102 can be used to capture raw electrical signals or raw neural signals 300 from a subject's brain. Next, the decoder module 304 of the computing device 104 can filter the raw electrical signals in one or more desired frequency bands using one or more band - pass filters, wavelet convolutions, or combinations thereof. Then, the decoder module 304 can convert the voltage values of the filtered raw electrical signals into power values (expressed as V 2 / Hz or microV 2 / Hz, dB / Hz, S 2 (where S represents the unit of the signal), etc.), or into normalized power values (expressed as z - scores, ratios, differences, percentage changes). As a more specific example, the normalized power units in FIG. 4 can be normalized based on the median value of the signal.

[0115] Next, the decoder module 304 can apply a power threshold 402 to the magnitude or power - related values for each of the desired frequency bands. The decoder module 304 can identify or detect one of the transient oscillation bursts or pseudo - oscillation bursts 302 in response to one of the magnitudes or power - related values that exceed the power threshold 402 for each of the desired frequency bands. The power threshold 402 can be selected or optimized for each user or patient.

[0116] In some embodiments, decoder module 304 can apply a duration threshold 403 for each of the desired frequency bands. Decoder module 304 can identify or detect one of the transient oscillation bursts or pseudo-oscillation bursts 302 in response to the duration of one of the raw electrical signals that exceeds the duration threshold 403 for each of the desired frequency bands. The duration threshold 403 can be selected or optimized for each user or patient.

[0117] Transient oscillation bursts or pseudo-oscillation bursts 302 may in this field be referred to as "transients", "oscillation events", "[band] bursts (e.g., beta bursts)", "[band] events (e.g., gamma events)", "miniature evoked responses", or "oscillation bursts". Oscillation bursts or pseudo-oscillation bursts 302 can be characterized by being transient, which means that each burst persists for only a very short duration and that each burst is a high-energy burst, which means that the power of each burst exceeds a determined threshold power level (power threshold 402) relative to the baseline level of neural activity and / or background noise.

[0118] In some embodiments, the duration of a typical transient oscillation burst or pseudo-oscillation burst 302 can last from 1 millisecond to 100 milliseconds. In other embodiments, the duration of a typical transient oscillation burst or pseudo-oscillation burst 302 can last between 10 milliseconds and 100 milliseconds. The duration of the transient oscillation burst or pseudo-oscillation burst 302 can depend on factors such as the measured frequency band. For example, the transient oscillation burst or pseudo-oscillation burst 302 can last from 1 millisecond to 10 milliseconds if the measured frequency band is relatively high (e.g., gamma band), or can last longer than 10 milliseconds if the measured frequency band is lower (e.g., alpha band).

[0119] Typical transient oscillation bursts or pseudo-oscillation bursts 302 can result in thousands of oscillation bursts or pseudo-oscillation bursts 302 across various electrodes 110 of the recording device 102 and across various desired frequency bands, even for neural signal recordings that last only for a few seconds (e.g., 3 to 5 seconds) because they are very brief or of limited duration.

[0120] In some embodiments, the desired frequency band can be between 0.1 Hz and 32 kHz. The desired frequency band can also be between 4 Hz and 400 Hz. In certain embodiments, the desired frequency band can be between 20 Hz and 200 Hz. In other embodiments, the desired frequency band can be between 35 Hz and 150 Hz.

[0121] Each of the electrodes 110 can record neural signals 300 in one or more frequency bands. This combination of electrodes 110 and frequency bands means that even for neural signal recordings that last only 3 to 5 seconds, it can include thousands or more transient oscillation bursts or pseudo-oscillation bursts 302. For example, each of the electrodes 110 can record in multiple frequency bands (e.g., the full band of neural signals), and the recordings can be decomposed into specific frequency components.

[0122] Transient oscillation bursts or pseudo-oscillation bursts 302 can be distinguished from a sustained increase in power in a specific frequency band.

[0123] These transient oscillation bursts or pseudo-oscillation bursts 302 can also be distinguished from action potential spike events. Action potential spikes are ubiquitous when sampled. That is, they have characteristic response functions regardless of the thoughts generated or recalled by the subject or the changes in mental states induced by the subject. On the other hand, the transient oscillation bursts or pseudo-oscillation bursts 302 exhibit specific burst characteristics or features that vary depending on the mental state or in response to the thoughts generated or recalled by the subject. For example, even when recorded from the same location in the subject's brain, the burst features detected during right hand versus left hand movement or during movement trials are different, including different burst rates or other burst features.

[0124] The transient oscillation bursts or pseudo-oscillation bursts 302 can be detected without the need to (or since there is no need to) average the results over multiple trials. That is, the transient oscillation bursts or pseudo-oscillation bursts 302 can be detected from ongoing or real-time neural signal recordings of the subject. The transient oscillation bursts or pseudo-oscillation bursts 302 may exhibit shorter power increases compared to the average data.

[0125] In some embodiments, the transient oscillation bursts or pseudo-oscillation bursts 302 can be detected using a power threshold 402, a duration threshold 403, or a combination thereof. In other embodiments, the transient pseudo-oscillation bursts 302 can be detected using a Hidden Markov Model (HMM) or template matching.

[0126] In certain embodiments, at least one of the power threshold 402 and the duration threshold 403 used to detect a transient oscillation burst or a pseudo-oscillation burst 302 can be selected or optimized based on at least one training session performed on the subject. For example, the training session can include instructing or prompting the subject to generate or recall thoughts (e.g., task-related thoughts or task-unrelated thoughts) or to induce a change in the mental state of the subject. Next, the decoder module 304 of the computing device 104 can use the recording device 102 to record the raw neural signals 300 from the subject's brain after prompting the subject. Next, the decoder module 304 can filter the raw electrical signals in a desired frequency band using one or more frequency decomposition methods such as one or more bandpass filters, wavelet convolutions, or combinations thereof. Next, the decoder module 304 can convert the voltage values of the filtered raw electrical signals into power values for each of the desired frequency bands.

[0127] Next, the computing device 104 can determine or otherwise set the optimal power threshold 402 to be applied to the power values for each of the desired frequency bands to distinguish the transient oscillation burst or the pseudo-oscillation burst 302 from the baseline level of neural activity and / or background noise.

[0128] Alternatively or additionally, the computing device 104 can determine or otherwise set the optimal duration threshold 403 to be applied to the raw electrical signals for each of the desired frequency bands to distinguish the transient oscillation burst or the pseudo-oscillation burst 302 from the baseline level of neural activity and / or background noise.

[0129] In some embodiments, at least one of the power threshold 402 and the duration threshold 403 can be determined based on a statistic or statistical variance, such as the number of standard deviations exceeding the median or average level of neural activity. As a more specific example, at least one of the power threshold 402 and the duration threshold 403 can be 1 to 6 standard deviations (SD) above the average level of neural activity.

[0130] The decoder module 304 can also extract one or more burst features 400 from the detected transient oscillation bursts or pseudo-oscillation bursts 302 within a predetermined or pre-set detection period 404. The decoder module 304 can detect hundreds or more transient oscillation bursts or pseudo-oscillation bursts 302 within each detection period 404 across various electrodes 110 of the recording device 102 and across various frequency bands (e.g., 0.1 Hz to 32 kHz).

[0131] In some embodiments, the detection period 404 can be between 10 milliseconds (ms) and 100 ms. More specifically, the detection period 404 can be between 50 ms and 100 ms. For example, the detection period 404 can be about 100 ms.

[0132] The decoder module 304 can extract one or more burst features 400 by counting or summing the number of detected transient oscillation bursts or pseudo-oscillation bursts 302 and determining the timing of such bursts 302. The decoder module 304 can also determine the frequency, power value, and duration of each burst 302.

[0133] The burst feature 400 can include a burst count 406, a burst rate 408, a burst band frequency 410 or frequency distribution, a burst interval length 412 (across a single channel and multiple channels), a burst timing 414 or timing pattern, an average burst duration 416, a burst waveform 418 (e.g., the time-domain waveform of the burst), or any modification or combination thereof. The burst feature 400 can also include the average power across bursts 302 within a time window, the maximum power of the burst 302, the number of cycles, the peak frequency of the burst 302, the minimum frequency of the burst 302, the maximum frequency of the burst, the frequency range (expressed in octaves), the average power immediately before / after the burst 302, the low-frequency instantaneous phase during a high-frequency burst, the alpha and beta power during a high-frequency burst, a vibration score (i.e., the correlation between the filtered signal and the raw signal during the burst 302). The burst feature can also include burst synchronization or distance (i.e., a measure of the correlation between bursts in different channels when treated as a point process), the left and / or right slope of a transient burst (i.e., how fast the amplitude rises or falls), and a repeating sequence in time of transient bursts (e.g., the type of thinking or movement of a particular user may generate a series of bursts that appear at a particular electrode at a particular time interval).

[0134] Any or all such burst features 400 can be extracted from the detected transient oscillatory bursts or pseudo-oscillatory bursts detected within each of the detection periods 404 (e.g., between 1 ms and 100 ms).

[0135] In some embodiments, the burst rate 408 can be calculated by dividing the burst count 406 by the length of the detection period 404. The burst count 406 can be calculated by summing all of the transient oscillation bursts or pseudo-oscillation bursts detected across all electrodes 110 or a subset of electrodes 110 of the recording device 102 during the detection period 404. For example, the burst rate 408 for an individual channel and frequency band can range between 1 and 200 bursts per second, depending on the selected frequency band (lower frequency bands typically exhibit lower burst rates compared to higher frequency bands).

[0136] One unexpected discovery made by the present applicants is that the burst rate 408 is a reliable and effective marker for predicting thoughts generated or recalled by a subject, or changes in mental states induced by a subject, when transient oscillation bursts or pseudo-oscillation bursts 302 are measured.

[0137] In certain alternative embodiments, the burst count 406 can be calculated by summing transient oscillation bursts or pseudo-oscillation bursts 302 detected across a portion, but not all, of the electrodes 110 of the recording device 102 during the detection period 404. For example, if the recording device 102 is an implantable recording device 102 such as a stent electrode array 108, the burst count 406 can be calculated by summing transient oscillation bursts or pseudo-oscillation bursts 302 detected across a portion, but not all, of the electrodes 110 of the stent electrode array 108.

[0138] In other embodiments, the computing device 104 may apply a weighting factor to such an electrode 110 (or channel) such that a transient oscillation burst or spurious oscillation burst 302 detected at one or more electrodes 110 (also referred to as “channels” of the recording device 102) of the recording device 102 is weighted more (e.g., 1.5 times, 2 times, 3 times, etc.) than a transient oscillation burst or spurious oscillation burst 302 detected at another electrode 110 (or another channel) of the recording device 102. For example, the recording device 102 can be a stent electrode array 108 that includes a plurality of electrodes 110 coupled to an expandable stent 112 or a scaffold. In these embodiments, the computing device 104 may apply a weighting factor to such an electrode 110 or channel such that a transient oscillation burst or spurious oscillation burst 302 detected at one or more electrodes 110 or channels of the stent electrode array 108 is weighted more (e.g., 1.5 times, 2 times, 3 times, etc.) than a transient oscillation burst or spurious oscillation burst 302 detected at another electrode 110 or another channel of the stent electrode array 108.

[0139] In some embodiments, the extracted burst feature 400 can be a change in one or more of the other burst features 400, such as a burst count 406, burst rate 408, burst band frequency 410, inter-burst interval length 412, burst timing 414, burst duration 416, or change in burst waveform 418, over a plurality of detection periods 404. In these embodiments, the change in one or more burst features 400 can be provided as an input to the classification layer 306 to enable the classification layer 306 to make a prediction regarding a thought generated or recalled by the subject, or a change in a mental state induced by the subject.

[0140] As discussed above, the classification layer 306 of the computing device 104 can predict thoughts generated or recalled by a subject, or changes in mental states induced by the subject, by applying a machine learning algorithm 308 (e.g., a deep learning algorithm) to the extracted one or more burst features 400. For example, the classification layer 306 of the computing device 104 can predict thoughts generated or recalled by a subject, or changes in mental states induced by the subject, by applying the machine learning algorithm 308 to the one or more burst features 400 extracted within each detection period 404.

[0141] Alternatively or additionally, the classification layer 306 of the computing device 104 can predict thoughts generated or recalled by a subject, or changes in mental states induced by the subject, by applying a feature threshold 310 to the extracted one or more burst features 400. For example, the classification layer 306 of the computing device 104 can predict thoughts generated or recalled by a subject, or changes in mental states induced by the subject, by applying the feature threshold 310 to the one or more burst features 400 extracted within each detection period 404.

[0142] In some embodiments, the classification layer 306 can make a prediction at the end of each detection period 404. For example, a 3 - second recording can include 30 detection periods 404 and 30 predictions regarding thoughts generated or recalled by a subject, or changes in mental states induced by the subject.

[0143] In other embodiments, the classification layer 306 can make a prediction only when a plurality of detection periods 404 have elapsed. In these embodiments, the classification layer 306 can depend on data from previous detection periods 404 to make a prediction.

[0144] If the extracted burst feature 400 is the burst rate 408 or includes the burst rate 408, the feature threshold 310 can be a burst rate threshold or can include a burst rate threshold. For example, the classification layer 306 of the computing device 104 can predict that the subject is thinking about moving a body part of the subject if the burst rate 408 calculated from the transient oscillation burst or pseudo-oscillation burst 302 detected within the detection period 404 exceeds the burst rate threshold during that particular detection period 404.

[0145] In other embodiments, if the extracted burst feature 400 is the burst count 406, the feature threshold can be a burst count threshold. Further, if the extracted burst feature 400 is the burst interval length 412 (the average time length between consecutive bursts 302), the feature threshold can be an interval length threshold. Further, if the extracted burst feature 400 is the average burst duration 416, the feature threshold can be a burst duration threshold.

[0146] In some embodiments, the feature threshold can be a static threshold. For example, the feature threshold can be set or determined after a short training period or training session before the subject uses the system 100.

[0147] In other embodiments, the feature threshold can be a dynamic threshold. In these embodiments, the feature threshold can be adjusted over time by the computing device 102. For example, the feature threshold can be adjusted over time by the classification layer 306 of the computing device 104. In certain embodiments, the feature threshold may need to exceed the threshold for a particular duration.

[0148] In certain embodiments, the feature threshold can be the median or average threshold amount calculated from the previous detection period 404. For example, the burst rate threshold can be the median of the burst rates calculated from the burst rates 408 determined from the previous detection period 404. Also, for example, the burst count threshold can be the median of the burst counts calculated from the burst counts 406 determined from the previous detection period 404.

[0149] As discussed above, the classification layer 306 can also make predictions by passing one or more burst features 400 to a machine learning algorithm 308 executed on the computing device 104. In some embodiments, the machine learning algorithm 308 can be a deep learning network such as a neural network. For example, the machine learning algorithm 308 can be a recurrent neural network. As a more specific example, the classification layer 306 can use a recurrent neural network such as an LSTM network to make predictions regarding thoughts generated or recalled by a subject, or changes in mental states induced in the subject, by feeding the extracted burst features to the long short-term memory (LSTM) network.

[0150] In some embodiments, the input to a recurrent neural network (RNN) (e.g., an LSTM or another RNN) can be features extracted from a transient oscillation burst or a pseudo-oscillation burst 302 (e.g., a burst count 406 for each electrode 110 and frequency band, the amplitude of the burst, the specific frequency at which the burst occurs, the waveform, or any combination thereof). The RNN can be trained using a single session or multiple sessions (each session can include hundreds of examples), and the optimal model can be selected based on the accuracy of the results obtained during the training session. The final output of the RNN can be a prediction of the cluster type (e.g., a movement type or a resting state, etc.) or the probability of each class (e.g., if three classes are selected, the output can be a 60% probability that this is a specific movement type, a 20% probability that this is a different movement type, and a 20% probability that this is no movement or the subject is in a resting state). These probabilities can be further processed by the model to make a final prediction.

[0151] The machine learning algorithm 308 can be trained using one or more burst features 400 extracted from a previous detection period 404, along with past predictions made by the machine learning algorithm 308. For example, the machine learning algorithm 308 can be trained using the burst count 406 and / or the burst rate 408 using past predictions made by the machine learning algorithm 308. The machine learning algorithm can be trained to enhance future predictions made by the machine learning algorithm 308.

[0152] The computing device 104 can map or associate a particular prediction made by the classification layer 306 to a particular input command 312. For example, the computing device 104 can associate a prediction made by the classification layer 306 regarding thoughts generated or recalled by a subject, or a change in a mental state induced by a subject, with a particular input command 312. As a more specific example, the computing device 104 can associate a prediction made by the classification layer 306 regarding a thought generated or recalled by a subject to move a body part of the subject (e.g., move the subject's ankle) with an input command 312 that initiates a click of the cursor of a personal computing device functioning as device 10.

[0153] In some embodiments, the computing device 104 can associate a first prediction regarding a first thought (e.g., a thought to move the subject's ankle) generated or recalled by a subject with a first input command (e.g., a cursor click), and can associate a second prediction regarding a second thought (e.g., a thought to move the subject's arm) generated or recalled by a subject with a second input command (e.g., a command to close a software application).

[0154] As discussed above, thoughts generated or recalled by a subject can be related to input command 312 or a device 10 intended to be controlled by the subject. For example, thoughts generated or recalled by a subject can be thoughts of moving a cursor displayed on a display of a personal computing device functioning as device 10, or thoughts of moving a moving vehicle (e.g., a wheelchair carrying the subject) functioning as device 10. In these examples, the thoughts are considered task-related thoughts and may be referred to as task-related thoughts. Further, changes in the mental state induced by the subject can include the subject focusing attention on the cursor or on the moving vehicle. In these examples, the changes in the mental state may be referred to as task-related mental state changes.

[0155] In other embodiments, thoughts generated or recalled by a subject can be thoughts that are unrelated or disengaged from instructing input command 312 or device 10 (including any component thereof or software application executed thereon). As a more specific example, thoughts generated or recalled by a subject can be thoughts for moving one or more body parts of the subject (e.g., hands, fingers, ankles, feet, toes, legs, arms, head, etc.). Also, for example, thoughts generated or recalled by a subject can be thoughts for tensing or relaxing one or more body parts of the subject (e.g., tensing or relaxing one or more muscles or muscle groups of the subject). In these examples, the thoughts may be referred to as task-unrelated thoughts. Further, changes in the mental state induced by the subject can include the subject focusing attention on one or more body parts or on a task other than instructing device 10 (including any component thereof or software application executed thereon). In these examples, the changes in the mental state may be referred to as task-unrelated mental state changes.

[0156] In some embodiments, the control of device 10 using system 100 can be performed asynchronously such that the subject can control device 10 by generating or recalling appropriate thoughts or inducing appropriate mental state changes without being prompted to do so beforehand. In fact, one of the technical problems faced by the applicant was how to enable the subject to effectively control the device without being prompted to do so. One technical solution discovered by the applicant is the system 100 and method disclosed herein, where transient oscillation bursts or pseudo-oscillation bursts 302 generated in response to thoughts generated or recalled by the subject, or mental state changes induced by the subject, are detected from the ongoing or real-time neural signal recording 300 captured by recording device 102. Then, burst features 400 are extracted from the transient oscillation bursts or pseudo-oscillation bursts 302 detected during detection period 404, and based on the extracted burst features 400, a prediction is made regarding the thoughts generated or recalled by the subject, or the mental state changes induced by the subject. In these embodiments, recording device 102 continuously captures raw electrical or raw neural signals 300 from the subject's brain, such signals are continuously filtered, converted to power values, and thresholds are continuously applied to identify transient oscillation bursts or pseudo-oscillation bursts 302 within each detection period 404. Then, at the end of each detection period 404, a prediction is made by classification layer 306 of computing device 104 regarding whether the subject intended to send an input command to device 10.

[0157] Alternatively, control of the device 10 using the system 100 can be performed synchronously in response to a subject being prompted or instructed to do so by a computing device 104, a telemetry unit 116, or another device, by generating or recalling appropriate thoughts, or by inducing a change in mental state so as to control the device 10. In these embodiments, one or more detection periods 404 can be initiated when the subject is prompted or instructed to generate or recall appropriate thoughts or to induce an appropriate change in mental state. One unexpected discovery made by the present applicants is that when control of the device 10 is performed synchronously, certain burst characteristics 400, such as inter-burst interval length 412, burst timing 414, or timing pattern, can be used to predict thoughts generated or recalled by the subject, or changes in mental state induced by the subject.

[0158] FIG. 5A is a graph showing the total burst count over all channels (or electrodes 110) of the recording device 102 and all burst frequencies as a function of the predictions made. FIG. 5B is a step chart showing the results and accuracy of the predictions.

[0159] As shown in FIG. 5A, predictions are made at the end of each 100 millisecond detection period 404, and a total of 3000 predictions can be made during this 5-minute recording. A burst count threshold can be set as a feature threshold 310. For example, the burst count threshold can be approximately 220 bursts over all channels and burst frequencies.

[0160] In this example, the recording device 102 can be an implantable recording device such as the stent electrode array 108. The total number of transient oscillation bursts or pseudo-oscillation bursts 302 detected across all electrodes 110 or a subset of electrodes 110 of the stent electrode array 108, and across all frequencies (e.g., from 0.1 Hz to 32 kHz), can be counted at the end of each detection period 404. When the total burst count exceeds the burst count threshold, the classification layer 306 of the computing device 104 can predict that the subject attempted to move the subject's ankle or recalled thoughts of moving it. If the total burst count does not exceed the burst count threshold during the detection period 404, the classification layer 306 of the computing device 104 predicts that the mental state of the subject was a resting mental state.

[0161] Figures 5A and 5B also show that the accuracy of the prediction can be determined by periodically providing cues or prompts to the subject to either move the subject's ankle or maintain / achieve a resting state. The predictions made after the cue can be used to determine the accuracy of the prediction. For example, Figure 5B shows that the computing device 104 had high accuracy in predicting a subject in a resting state, but was slightly less accurate when predicting that the subject attempted to move the subject's ankle or generated / recalled thoughts of moving the subject's ankle. This result is acceptable considering that a slightly lower accuracy prediction of a resting or inactive state only results in some instances of the input command 312 not being sent by the subject when desired during one or more detection periods 404. Since the detection period 404 is only between about 1 millisecond and 100 milliseconds, missing or insufficient input commands can be easily corrected by subsequent or next input command transmissions (i.e., if the subject intends to click the cursor and the subject's intent is missing during one detection period, the click can be detected in the next detection period, and as a result, the subject may not even perceive a delay or may think it was caused by a delay in the machine). The opposite result (i.e., the computing device 104 being unable to accurately detect a resting state) is unacceptable because it causes the system 100 to send input commands 312 (e.g., cursor clicks or user input) when not desired by the subject.

[0162] Figures 5A and 5B show that the systems 100 and methods disclosed herein can be used to accurately transmit the input command 312 to control the device 10 based on the detection of transient oscillation bursts or pseudo-oscillation bursts 302 from ongoing or real-time neural signal recordings.

[0163] FIG. 6 shows a confusion matrix indicating the classification accuracy of the predictions made by system 100 regarding thoughts generated or recalled by a subject, or changes in mental states induced by the subject. For example, as shown in FIG. 6, thoughts generated or recalled by a subject, or changes in mental states induced by the subject, are related to the movements of both ankles of the subject. When the subject is not generating or recalling such thoughts, or inducing such changes in their mental state, the subject has achieved or maintained a resting state.

[0164] To generate this confusion matrix, a total of 15 burst features 400 were extracted from a number of transient oscillation bursts or pseudo-oscillation bursts 302 detected from 10 minutes of neural signal recordings of the subject. For example, the 15 burst features 400 are from the following list of burst features 400, namely, burst count 406, burst rate 408, burst band frequency 410 or frequency distribution, inter-burst interval length 412 (across a single channel and multiple channels), burst timing 414 or timing pattern, average burst duration 416, burst waveform 418 (e.g., the time-domain waveform of the burst), average power across bursts 302 within a time window, maximum power of burst 302, number of cycles, peak frequency of burst 302, minimum frequency of burst 302, maximum frequency of the burst, frequency range (expressed in octaves), average power immediately before / after burst 302, low-frequency instantaneous phase during high-frequency bursts, alpha and beta power during high-frequency bursts, oscillation score (i.e., the correlation between the filtered signal and the raw signal during burst 302), burst synchronization or distance (i.e., a measure of the correlation between bursts in different channels when treated as a point process), left and / or right slope of the transient burst (i.e., how fast the amplitude rises or falls), and repeating sequences in time of the transient burst (e.g., a particular type of a user's thought or movement may generate a series of bursts that appear at a particular electrode at a particular time interval) can be selected.

[0165] As shown in FIG. 6, the system 100 was most accurate (having an f1 score of about 0.70) in predicting that the subject was in such a state when the subject was actually at rest. This prediction can be regarded as a true negative prediction.

[0166] The system 100 was slightly less accurate (having an f1 score of about 0.65) in predicting that the subject was generating / recalling thoughts related to moving both ankles of the subject or inducing a change in the mental state of the subject related to moving both ankles of the subject when the subject was actually doing so. This prediction can be regarded as a true positive prediction.

[0167] The system 100 was not very likely to predict that the subject was at rest when the subject was actually generating / recalling thoughts related to moving both ankles of the subject or inducing a change in the mental state of the subject related to moving both ankles of the subject. This prediction can be regarded as a false negative prediction.

[0168] The system 100 was least likely to predict that the subject was generating / recalling thoughts related to moving both ankles of the subject or inducing a change in the mental state of the subject related to moving both ankles of the subject when the subject was actually at rest. This prediction can be regarded as a false positive prediction. The goal of any effective BCI system 100 is to minimize false positive predictions. This is because false positive predictions most interfere with the use of the system 100 by the subject, as the BCI system 100 will erroneously send the input command 312 to the device 10 (resulting in an incorrect cursor click or user input) when the subject does not have such a desire to do so.

[0169] FIG. 7A is a graph showing the cumulative burst count across all channels of recording device 102 over a 5 - second period. The subject was instructed to either generate / recall thoughts of moving both of the subject's ankles or maintain a resting state for 5 seconds. During this 5 - second period, all transient oscillatory bursts or pseudo - oscillatory bursts 302 detected across all channels or electrodes 110 of recording device 102 were counted. FIG. 7A shows that, with rare exceptions, a large number of transient oscillatory bursts or pseudo - oscillatory bursts 302 were detected in response to the subject generating or recalling thoughts of moving both of the subject's ankles, and that the amount of such bursts 302 decreased significantly when the subject remained in a resting state.

[0170] FIG. 7B is a classification report showing the accuracy, recall, and f1 - score of the predictions made by system 100 using transient oscillatory bursts or pseudo - oscillatory bursts 302 detected from a 5 - minute neural signal recording of a subject. The f1 - score is an error metric that measures the performance or accuracy of a classification model or classification layer 306. The f1 - score provides insight into the ability (recall) of a classification model or classification layer 306 to capture positive cases and the accuracy (precision) of the cases it captures. The relatively high f1 - score shown in FIG. 7B indicates that device 10 can be reliably controlled with high accuracy at a click rate of over 3 clicks per second.

[0171] The present disclosure also covers the following examples.

[0172] Example 1. A method of controlling a device, comprising: detecting one or more transient oscillation bursts or pseudo-oscillation bursts from ongoing neural signal recordings of a subject captured by a recording device, wherein the one or more transient oscillation bursts or pseudo-oscillation bursts are generated in response to thoughts generated by the subject or changes in mental states induced by the subject; extracting one or more burst features from the one or more transient oscillation bursts or pseudo-oscillation bursts detected during a detection period using one or more processors of a computing device communicatively coupled to the recording device; predicting thoughts generated by the subject or changes in mental states induced by the subject by applying at least one of a machine learning algorithm and a feature threshold to the one or more burst features extracted during the detection period using the one or more processors; and transmitting an input command associated with the prediction to the device to control the device.

[0173] Example 2. The method of Example 1, wherein the ongoing neural signal recordings of the subject are performed by recording raw electrical signals from the subject's brain using a recording device, and the step of detecting one or more transient oscillation bursts or pseudo-oscillation bursts comprises: filtering the raw electrical signals in one or more desired frequency bands using one or more frequency decomposition methods; for each of the desired frequency bands, converting the voltage values of the filtered raw electrical signals into magnitude or power-related values; applying at least one of a power threshold for the magnitude or power-related values for each of the desired frequency bands and a duration threshold for each of the desired frequency bands; and for each of the desired frequency bands, identifying one of the transient oscillation bursts or pseudo-oscillation bursts in response to at least one of the magnitude or power-related values exceeding the power threshold and the filtered raw electrical signals exceeding the duration threshold.

[0174] Example 3. At least one of the power threshold and the duration threshold is selected based on at least one training session performed on a subject, the at least one training session instructing or prompting the subject to generate thoughts or induce a change in the mental state of the subject, and after prompting the subject to generate thoughts, using a recording device to record raw electrical signals from the subject's brain, filtering the raw electrical signals in one or more desired frequency bands using one or more frequency decomposition methods, for each of the desired frequency bands, converting the voltage value of the filtered raw electrical signals into a power value, and selecting at least one of the power threshold and the duration threshold to be applied for each of the desired frequency bands to distinguish transient oscillation bursts or pseudo-oscillation bursts from background noise, the method of Example 2.

[0175] Example 4. The method of Example 2, wherein the desired frequency band includes a frequency band between 0.1 Hz and 32 kHz.

[0176] Example 5. The method of Example 2, wherein the desired frequency band includes at least one of the beta frequency band, the gamma frequency band, and the high gamma frequency band.

[0177] Example 6. The method of Example 1, wherein one or more burst features include a burst rate, and the burst rate is calculated by dividing the burst count by the length of the detection period.

[0178] Example 7. The method of Example 6, wherein the burst count is calculated by summing all of the transient oscillation bursts or pseudo-oscillation bursts detected across all electrodes or a subset of electrodes of the recording device during the detection period.

[0179] Example 8. The method of Example 1, wherein the feature threshold is a burst rate threshold.

[0180] Example 9. The method of Example 8, wherein the burst rate threshold is a median value of the burst rates calculated from previous detection periods.

[0181] Example 10. The method of Example 1, wherein one or more burst features include at least one of burst count, burst rate, burst band frequency or frequency distribution, burst interval length, burst timing or timing pattern, average burst duration, burst waveform, and any variations thereof.

[0182] Example 11. The method of Example 1, wherein the machine learning algorithm is a neural network.

[0183] Example 12. The method of Example 11, wherein the neural network is a recurrent neural network.

[0184] Example 13. The method of Example 12, wherein the recurrent neural network is a long short-term memory (LSTM) neural network.

[0185] Example 14. The method of Example 1, wherein the feature threshold is a static threshold.

[0186] Example 15. The method of Example 1, wherein the feature threshold is a dynamic threshold adjusted over time by a computing device.

[0187] Example 16. The method of Example 1, wherein the detection period is between 1 millisecond and 100 milliseconds.

[0188] Example 17. The method of Example 1, wherein the device is at least one of a personal computing device, an Internet of Things (IoT) device, and a moving vehicle.

[0189] Example 18. The method of Example 17, wherein the device is a personal computing device and the input command is a command to initiate a click of the cursor of the personal computing device.

[0190] Example 19. The method of Example 1, wherein the subject's thought is a thought generated by the subject to move one or more body parts of the subject.

[0191] Example 20. The method of Example 1, wherein the thinking is generated by the subject without being prompted to do so, such that the control of the device is performed asynchronously.

[0192] Example 21. The method of Example 1, wherein the thinking is generated by the subject in response to being prompted to do so, such that the control of the device is performed synchronously.

[0193] Example 22. The method of Example 1, wherein the recording device is a non-invasive recording device.

[0194] Example 23. The method of Example 1, wherein the recording device is an invasive recording device.

[0195] Example 24. The method of Example 23, wherein the recording device is an intravascular recording device comprising a plurality of electrodes held by an intravascular carrier configured to be implanted within a vein or sinus of the subject's brain.

[0196] Example 25. The method of Example 24, wherein one or more transient oscillation bursts or pseudo-oscillation bursts are detected using electrodes held by an expandable stent or scaffold, and the method further comprises applying a weighting factor to one or more electrodes of the expandable stent or scaffold such that the transient oscillation bursts or pseudo-oscillation bursts detected at one or more electrodes are weighted more than the transient oscillation bursts or pseudo-oscillation bursts detected at another electrode of the expandable stent or scaffold.

[0197] Example 26. The method of Example 23, wherein the recording device is an implanted microelectrode array.

[0198] Example 27. The method of Example 26, wherein the recording device is a Utah microelectrode array.

[0199] Example 28. The method of Example 23, wherein the recording device is a thin-film electrode array.

[0200] Example 29. The method of Example 23, wherein the recording device is an electrode array configured to be implanted on the brain surface.

[0201] Example 30. The method of Example 1, further comprising the step of training a machine learning algorithm using previous predictions made by the machine learning algorithm and burst features extracted from a previous detection period to enhance the predictions made by the machine learning algorithm.

[0202] Example 31. A system for controlling a device, the system comprising a recording device configured to capture ongoing neural signal recordings of a subject, and a computing device having one or more processors communicatively coupled to the recording device, the one or more processors being configured to detect one or more transient oscillatory bursts or pseudo-oscillatory bursts from the ongoing neural signal recordings of the subject, the one or more transient oscillatory bursts or pseudo-oscillatory bursts being generated in response to thoughts generated by the subject or changes in mental states induced by the subject, extracting one or more burst features from the one or more transient oscillatory bursts or pseudo-oscillatory bursts detected during a detection period, predicting thoughts generated by the subject or changes in mental states induced by the subject by applying at least one of a machine learning algorithm and a feature threshold to the one or more burst features extracted during the detection period, and transmitting an input command associated with the prediction to the device to control the device.

[0203] Example 32. The recording device is configured to capture ongoing neural signal recordings of a subject by recording raw electrical signals from the subject's brain, and one or more processors of the computing device use one or more frequency decomposition methods to filter the raw electrical signals in one or more desired frequency bands, and for each of the desired frequency bands, convert the voltage values of the filtered raw electrical signals into magnitude or power-related values, and apply at least one of a power threshold for the magnitude or power-related values for each of the desired frequency bands, or a duration threshold for each of the desired frequency bands, and for each of the desired frequency bands, identify one of a transient oscillation burst or a pseudo-oscillation burst in response to at least one of a magnitude or power-related value exceeding the power threshold and a filtered raw electrical signal exceeding the duration threshold. The system of Example 31 is programmed to perform the above operations.

[0204] Example 33. At least one of the power threshold and the duration threshold is selected based on at least one training session performed on the subject using the computing device or another device and the recording device, the computing device or another device is configured to instruct or prompt the subject to generate thoughts or induce a change in the mental state of the subject, the recording device is configured to record raw electrical signals from the subject's brain after the subject is prompted to generate thoughts, and one or more processors of the computing device use one or more frequency decomposition methods to filter the raw electrical signals in one or more desired frequency bands, and for each of the desired frequency bands, convert the voltage values of the filtered raw electrical signals into power values, and select at least one of the power threshold and the duration threshold to be applied for each of the desired frequency bands to distinguish a transient oscillation burst or a pseudo-oscillation burst from background noise. The system of Example 32 is programmed to perform the above operations.

[0205] Example 34. The system of Example 32, wherein the desired frequency band includes a frequency band between 0.1 Hz and 32 kHz.

[0206] Example 35. The system of Example 32, wherein the desired frequency band includes at least one of a beta frequency band, a gamma frequency band, and a high gamma frequency band.

[0207] Example 36. The system of Example 31, wherein one or more burst features include a burst rate, and the burst rate is calculated by dividing a burst count by the length of a detection period.

[0208] Example 37. The system of Example 36, wherein the burst count is calculated by summing all of the transient oscillation bursts or pseudo-oscillation bursts detected across all electrodes or a subset of electrodes of a recording device during a detection period.

[0209] Example 38. The system of Example 31, wherein the feature threshold is a burst rate threshold.

[0210] Example 39. The system of Example 38, wherein the burst rate threshold is a median value of burst rates calculated from previous detection periods.

[0211] Example 40. The system of Example 31, wherein one or more burst features include at least one of a burst count, a burst rate, a burst band frequency or frequency distribution, an inter-burst interval length, a burst timing or timing pattern, an average burst duration, a burst waveform, and any variations thereof.

[0212] Example 41. The system of Example 31, wherein the machine learning algorithm is a neural network.

[0213] Example 42. The system of Example 41, wherein the neural network is a recurrent neural network.

[0214] Example 43. The system of Example 42, wherein the recurrent neural network is a long short-term memory (LSTM) neural network.

[0215] Example 44. The system of Example 31, wherein the feature threshold is a static threshold.

[0216] Example 45. The system of Example 31, wherein the feature threshold is a dynamic threshold that is adjusted over time by a computing device.

[0217] Example 46. The system of Example 31, wherein the detection period is between 1 millisecond and 100 milliseconds.

[0218] Example 47. The system of Example 31, wherein the device is at least one of a personal computing device, an Internet of Things (IoT) device, and a moving vehicle.

[0219] Example 48. The system of Example 47, wherein the device is a personal computing device and the input command is a command to initiate a click of the cursor of the personal computing device.

[0220] Example 49. The system of Example 31, wherein the subject's thought is a thought generated by the subject to move one or more body parts of the subject.

[0221] Example 50. The system of Example 31, wherein the thought is generated by the subject without being prompted to do so, such that the control of the device is performed asynchronously.

[0222] Example 51. The system of Example 31, wherein the thought is generated by the subject in response to being prompted to do so, such that the control of the device is performed synchronously.

[0223] Example 52. The system of Example 31, wherein the recording device is a non-invasive recording device.

[0224] Example 53. The system of Example 31, wherein the recording device is an invasive recording device.

[0225] Example 54. The system of Example 53, wherein the recording device is an intravascular recording device comprising a plurality of electrodes held by an intravascular carrier configured to be implanted within a vein or sinus of a subject's brain.

[0226] Example 55. The system of Example 54, wherein one or more transient oscillation bursts or pseudo-oscillation bursts are detected using electrodes held by an expandable stent or scaffold, and a weighting factor is configured to be applied to one or more electrodes of the expandable stent or scaffold such that the transient oscillation bursts or pseudo-oscillation bursts detected at one or more electrodes are weighted more than the transient oscillation bursts or pseudo-oscillation bursts detected at another electrode of the expandable stent or scaffold.

[0227] Example 56. The system of Example 53, wherein the recording device is an implanted microelectrode array.

[0228] Example 57. The system of Example 56, wherein the recording device is a Utah microelectrode array.

[0229] Example 58. The system of Example 53, wherein the recording device is a thin-film electrode array.

[0230] Example 59. The system of Example 53, wherein the recording device is an electrode array configured to be implanted on the surface of the brain.

[0231] Example 60. The system of Example 31, wherein the machine learning algorithm is configured to be trained using previous predictions made by the machine learning algorithm and burst features extracted from a previous detection period to enhance the predictions made by the machine learning algorithm.

[0232] Example 61. When executed by one or more processors, detecting one or more transient oscillatory bursts or pseudo-oscillatory bursts from ongoing neural signal recordings of a subject, wherein the one or more transient oscillatory bursts or pseudo-oscillatory bursts are generated in response to thoughts generated by the subject or changes in mental states induced by the subject; extracting one or more burst features from the one or more transient oscillatory bursts or pseudo-oscillatory bursts detected during a detection period; predicting thoughts generated by the subject or changes in mental states induced by the subject by applying at least one of a machine learning algorithm and a feature threshold to the one or more burst features extracted during the detection period; and transmitting an input command associated with the prediction to a device to control the device. One or more non-transitory computer-readable media storing instructions for performing the steps.

[0233] Example 62. The ongoing neural signal recordings of the subject are made by recording raw electrical signals from the subject's brain using a recording device. The step of detecting one or more transient oscillatory bursts or pseudo-oscillatory bursts includes filtering the raw electrical signals in one or more desired frequency bands using one or more frequency decomposition methods; for each of the desired frequency bands, converting voltage values of the filtered raw electrical signals into magnitude or power-related values; applying at least one of a power threshold for the magnitude or power-related values for each of the desired frequency bands and a duration threshold for each of the desired frequency bands; and for each of the desired frequency bands, identifying one of the transient oscillatory bursts or pseudo-oscillatory bursts in response to at least one of a magnitude or power-related value exceeding the power threshold and a filtered raw electrical signal exceeding the duration threshold. One or more non-transitory computer-readable media of Example 61 further comprising the steps.

[0234] Example 63. At least one of a power threshold and a duration threshold is selected based on at least one training session performed on a subject, the at least one training session including steps of instructing or prompting the subject to generate thoughts or induce a change in the mental state of the subject, after prompting the subject to generate thoughts, receiving raw electrical signals of the subject's brain from a recording device, filtering the raw electrical signals in one or more desired frequency bands using one or more frequency decomposition methods, for each of the desired frequency bands, converting the voltage value of the filtered raw electrical signals into a power value, and selecting at least one of a power threshold and a duration threshold to be applied for each of the desired frequency bands to distinguish a transient oscillation burst or a pseudo-oscillation burst from background noise, the one or more non-transitory computer-readable media of Example 62.

[0235] Example 64. The one or more non-transitory computer-readable media of Example 62, wherein the desired frequency band includes a frequency band between 0.1 Hz and 32 kHz.

[0236] Example 65. The one or more non-transitory computer-readable media of Example 62, wherein the desired frequency band includes at least one of a beta frequency band, a gamma frequency band, and a high gamma frequency band.

[0237] Example 66. The one or more non-transitory computer-readable media of Example 61, wherein one or more burst characteristics include a burst rate, and the burst rate is calculated by dividing a burst count by the length of a detection period.

[0238] Example 67. The one or more non-transitory computer-readable media of Example 66, wherein the burst count is calculated by summing all of the transient oscillation bursts or pseudo-oscillation bursts detected across all electrodes or a subset of electrodes of the recording device during the detection period.

[0239] Example 68. One or more non-transitory computer-readable media of Example 61, wherein the feature threshold is a burst rate threshold.

[0240] Example 69. One or more non-transitory computer-readable media of Example 68, wherein the burst rate threshold is a median value of burst rates calculated from a previous detection period.

[0241] Example 70. One or more non-transitory computer-readable media of Example 61, wherein one or more burst features include at least one of a burst count, a burst rate, a burst bandwidth frequency or frequency distribution, an inter-burst interval length, a burst timing or timing pattern, an average burst duration, a burst waveform, and any variations thereof.

[0242] Example 71. One or more non-transitory computer-readable media of Example 61, wherein the machine learning algorithm is a neural network.

[0243] Example 72. One or more non-transitory computer-readable media of Example 71, wherein the neural network is a recurrent neural network.

[0244] Example 73. One or more non-transitory computer-readable media of Example 72, wherein the recurrent neural network is a long short-term memory (LSTM) neural network.

[0245] Example 74. One or more non-transitory computer-readable media of Example 61, wherein the feature threshold is a static threshold.

[0246] Example 75. One or more non-transitory computer-readable media of Example 61, wherein the feature threshold is a dynamic threshold that is adjusted over time.

[0247] Example 76. One or more non-transitory computer-readable media of Example 61, wherein the detection period is between 1 millisecond and 100 milliseconds.

[0248] Example 77. One or more non-transitory computer-readable media of Example 61, wherein the device is at least one of a personal computing device, an Internet of Things (IoT) device, and a moving vehicle.

[0249] Example 78. One or more non-transitory computer-readable media of Example 77, wherein the device is a personal computing device and the input command is a command to initiate a click of a cursor of the personal computing device.

[0250] Example 79. One or more non-transitory computer-readable media of Example 61, wherein the subject's thought is a thought generated by the subject to move one or more body parts of the subject.

[0251] Example 80. One or more non-transitory computer-readable media of Example 61, wherein the thought is generated by the subject without being prompted to do so, such that the control of the device is performed asynchronously.

[0252] Example 81. One or more non-transitory computer-readable media of Example 61, wherein the thought is generated by the subject in response to being prompted to do so, such that the control of the device is performed synchronously.

[0253] Example 82. One or more non-transitory computer-readable media of Example 61, wherein the recording device is a non-invasive recording device.

[0254] Example 83. One or more non-transitory computer-readable media of Example 61, wherein the recording device is an invasive recording device.

[0255] Example 84. One or more non-transitory computer-readable media of Example 83, wherein the recording device is an intravascular recording device comprising a plurality of electrodes held by an intravascular carrier configured to be implanted within a vein or a cavity of the subject's brain.

[0256] Example 85. One or more transient oscillation bursts or pseudo-oscillation bursts are detected using electrodes held by an expandable stent or scaffold, and the step further includes applying a weighting factor to one or more electrodes of the expandable stent or scaffold such that the transient oscillation bursts or pseudo-oscillation bursts detected at one or more electrodes are weighted more than the transient oscillation bursts or pseudo-oscillation bursts detected at another electrode of the expandable stent or scaffold. One or more non-transitory computer-readable media of Example 84.

[0257] Example 86. One or more non-transitory computer-readable media of Example 83, wherein the recording device is an implantable microelectrode array.

[0258] Example 87. One or more non-transitory computer-readable media of Example 86, wherein the recording device is a Utah microelectrode array.

[0259] Example 88. One or more non-transitory computer-readable media of Example 83, wherein the recording device is a thin-film electrode array.

[0260] Example 89. One or more non-transitory computer-readable media of Example 83, wherein the recording device is an electrode array configured to be implanted on the brain surface.

[0261] Example 90. The step further includes training a machine learning algorithm using a previous prediction made by the machine learning algorithm and burst features extracted from a previous detection period to enhance a prediction made by the machine learning algorithm. One or more non-transitory computer-readable media of Example 61.

[0262] Example 91. Filtering a raw electrical signal in one or more desired frequency bands using one or more frequency decomposition methods, converting, for each of the desired frequency bands, the voltage value of the filtered raw electrical signal to a magnitude or power-related value, applying at least one of a power threshold for the magnitude or power-related value for each of the desired frequency bands, or a duration threshold for each of the desired frequency bands, and identifying one of a transient burst or a pseudo-burst in response to at least one of a magnitude or power-related value exceeding the power threshold and a filtered raw electrical signal exceeding the duration threshold for each of the desired frequency bands. A method for detecting one or more transient bursts or pseudo-bursts.

[0263] Some implementations have been described. However, it will be understood by those skilled in the art that various changes and modifications can be made to the present disclosure without departing from the spirit and scope of the embodiments. The elements of the systems, devices, apparatuses, and methods shown in any of the embodiments are illustrative of specific embodiments and can be combined with or used in other ways in other embodiments within the present disclosure. For example, the steps of any of the methods shown in the figures or described in the present disclosure do not require a specific or sequential order as illustrated or described to achieve the desired result. In addition, other step operations can be provided to achieve the desired result, or steps or operations can be excluded or omitted from the methods or processes described. Further, any component or part of any device or system described or shown in the present disclosure can be removed, excluded, or omitted to achieve the desired result. In addition, specific components or parts of the systems, devices, or apparatuses illustrated or described herein have been omitted for brevity and clarity.

[0264] Accordingly, other embodiments are within the scope of the following claims, and the present specification and / or drawings may be regarded in an illustrative rather than a limiting sense.

[0265] Each individual variation or embodiment described and illustrated herein has separate components and features that can be readily separated from, or readily combined with, the features of any other variation or embodiment. Modifications may be made to adapt a particular situation, material, composition of matter, process, process act, or step to the purposes, spirit, or scope of the invention.

[0266] The methods recited herein may be performed in any order of the recited events, as well as in any logically possible order of the recited events. Additionally, additional steps or acts may be provided, or steps or acts may be eliminated, to achieve the desired result.

[0267] Further, when a range of values is provided, all intervening values between the upper and lower limits of that range, as well as any other recited values or intervening values within the recited range, are encompassed within the invention. Also, any optional features of the described variations of the invention may be described and claimed independently, or in combination with any one or more of the features described herein. For example, a description of a range from 1 to 5 should be considered to disclose sub-ranges such as from 1 to 3, from 1 to 4, from 2 to 4, from 2 to 5, from 3 to 5, etc., as well as individual numerical values within that range, such as 1.5, 2.5, etc., and any whole or partial increments therebetween.

[0268] All existing subject matter (e.g., publications, patents, patent applications) referred to in this specification is incorporated herein by reference in its entirety, except where such subject matter may conflict with the subject matter of the present invention (in which case, what is present in this specification shall prevail). The items being referred to are provided for their disclosure prior to the filing date of the present application. Nothing in this specification should be construed as an admission that the present invention has no right to antedate such material by virtue of prior invention.

[0269] References to items in the singular include the possibility that there may be a plurality of the same items. More specifically, the singular forms "a", "an", "the" and "said" used in this specification and the appended claims include the plural referents unless the context clearly dictates otherwise. It should further be noted that the claims may be drafted to exclude any optional elements. Accordingly, this description is intended to serve as a basis for the use of exclusive terms such as "alone", "only" etc. in relation to claim elements, or the use of "negative" limitations. Unless defined otherwise, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention pertains.

[0270] References to the phrase "at least one" mean any one or any combination of one or more of the items or components, where such phrase modifies a plurality of items or components (or a recited list of items or components). For example, the phrase "at least one of A, B, and C" means (i) A, (ii) B, (iii) C, (iv) A, B, and C, (v) A and B, (vi) B and C, or (vii) A and C.

[0271] In understanding the scope of the present disclosure, the term "comprising" and its derivatives used herein are intended to be open-ended terms that specify the presence of the recited features, elements, components, groups, integers, and / or steps, but do not preclude the presence of other unrecited features, elements, components, groups, integers, and / or steps. The above also applies to words having similar meanings such as the terms "including", "having", and their derivatives. Also, the terms "portion", "section", "part", "member", "element", or "component" may have a dual meaning of a single part or multiple parts when used in the singular form. The following directional terms used herein, "front, rear, upper, lower, vertical, horizontal, below, lateral, horizontal direction, and vertical", and any other similar directional terms refer to those positions of the device or apparatus, or those directions of the device or apparatus being translated or moved.

[0272] Finally, terms of degree such as "substantially", "about", and "approximately" used herein mean the specified value, or a reasonable amount of deviation from the specified value (e.g., a deviation of up to ±0.1%, ±1%, ±5%, or ±10% such that such variation is appropriate) so that the final result is not significantly or materially changed. For example, "about 1.0 cm" can be interpreted to mean "1.0 cm", or between "0.9 cm and 1.1 cm". When a term of degree such as "about" or "approximately" is used to refer to a numerical or value that is part of a range, the term can be used to modify both the minimum numerical or value and the maximum numerical or value.

[0273] As used herein, the terms "engine" or "module" may refer to software, firmware, hardware, or combinations thereof. For example, in a software implementation, these may represent program code that executes tasks designated to be executed by a processor (e.g., a CPU, GPU, or a processor core therein). The program code may be stored in one or more computer-readable memories or storage devices. Any reference to a function, task, or operation performed by an "engine" or "module" may also refer to one or more processors of a device or server programmed to execute such program code to perform the function, task, or operation.

[0274] It will be understood by those skilled in the art that the various methods disclosed herein may be embodied in a non-transitory readable medium, a machine-readable medium, and / or a machine-accessible medium that includes instructions compatible, readable, and / or executable by a processor of a machine, device, or computing device or a server processor. The structures and modules in the figures are shown as separate and may communicate with only a few specific structures and may not communicate with other structures. Structures may be combined with each other, may perform overlapping functions, and may communicate with other structures not shown as connected in the figures. Thus, the specification and / or drawings may be considered in an illustrative rather than a limiting sense.

[0275] The present disclosure is not intended to be limited to the specific forms described, but is intended to cover alternatives, modifications, and equivalents of the variations or embodiments described herein. Further, the scope of the present disclosure fully encompasses other variations or embodiments that may become apparent to those skilled in the art from the perspective of the present disclosure.

Description of the Reference Numerals

[0276] 10 Device 100 Brain-computer interface (BCI) system, BCI system 102 Recording device, invasive recording device, implantable recording device 104 Computing device 106 Cerebral blood vessel 108 Stent electrode array 110 Electrode 112 Expandable stent 114 Communication conduit 116 Telemetry unit, internal telemetry unit, external telemetry unit 200 Coiled wire, wire 202 Fixed wire 204 Barbed anchor 206 Radially expandable anchor 208 Electroencephalogram (EEG) recording device, EEG device 210 Non-invasive electrode, electrode 212 Electrocorticogram (ECoG) recording device, ECoG device 214 Electrode 300 Raw electrical signal or nerve signal, nerve signal, raw nerve signal, nerve signal recording 302 Transient oscillation burst or pseudo-oscillation burst, transient oscillation or pseudo-oscillation, transient pseudo-oscillation burst, burst 304 Decoder module 306 Classification layer 308 Machine learning algorithm 310 Feature threshold 312 Input command 400 Burst feature 402 Power threshold 403 Duration threshold 404 Detection period 406 Burst count 408 Burst rate 410 Burst band frequency 412 Inter-burst interval length 414 Burst timing 416 Average burst duration 418 Burst waveform

Claims

1. A method for controlling a device, comprising: detecting, from ongoing neural signal recordings of a subject captured by a recording device, one or more transient oscillatory bursts or pseudo-oscillatory bursts, wherein the one or more transient oscillatory bursts or pseudo-oscillatory bursts are generated in response to thoughts generated by the subject or changes in mental states induced by the subject; extracting, using one or more processors of a computing device communicatively coupled to the recording device, one or more burst features from the one or more transient oscillatory bursts or pseudo-oscillatory bursts detected during a detection period; predicting, using the one or more processors, the thoughts generated by the subject or the changes in the mental states induced by the subject by applying at least one of a machine learning algorithm and a feature threshold to the one or more burst features extracted during the detection period; transmitting, to the device, input commands associated with the prediction to control the device A method as described above.

2. The ongoing neural signal recordings of the subject are performed by recording raw electrical signals from the subject's brain using the recording device, and the step of detecting the one or more transient oscillatory bursts or pseudo-oscillatory bursts comprises: filtering the raw electrical signals in one or more desired frequency bands using one or more frequency decomposition methods; for each of the desired frequency bands, converting voltage values of the filtered raw electrical signals into magnitude or power-related values; applying at least one of a power threshold for the magnitude or power-related values for each of the desired frequency bands or a duration threshold for each of the desired frequency bands; for each of the desired frequency bands, identifying one of the transient oscillatory bursts or pseudo-oscillatory bursts in response to at least one of the magnitude or power-related values exceeding the power threshold and the filtered raw electrical signals exceeding the duration threshold The method according to claim 1, further comprising the above steps.

3. At least one of the power threshold and the duration threshold is selected based on at least one training session performed on the subject, and the at least one training session includes instructing or prompting the subject to generate the thought or induce the change in the mental state of the subject; after prompting the subject to generate the thought, using the recording device to record the raw electrical signals from the brain of the subject; filtering the raw electrical signals in the one or more desired frequency bands using one or more frequency decomposition methods; for each of the desired frequency bands, converting the voltage value of the filtered raw electrical signals into a power value; selecting at least one of the power threshold and the duration threshold to be applied for each of the desired frequency bands to distinguish the transient oscillation burst or the pseudo-oscillation burst from background noise The method according to claim 2, comprising.

4. The method according to claim 2, wherein the desired frequency band includes a frequency band between 0.1 Hz and 32 kHz.

5. The method according to claim 2, wherein the desired frequency band includes at least one of a beta frequency band, a gamma frequency band, and a high gamma frequency band.

6. The method according to claim 1, wherein the one or more burst characteristics include a burst rate, and the burst rate is calculated by dividing the burst count by the length of the detection period.

7. The method according to claim 6, wherein the burst count is calculated by summing all of the transient oscillation bursts or pseudo-oscillation bursts detected across all electrodes or a subset of electrodes of the recording device during the detection period.

8. The method according to claim 1, wherein the feature threshold is a burst rate threshold.

9. The method according to claim 8, wherein the burst rate threshold is a median value of the burst rates calculated from previous detection periods.

10. The method according to claim 1, wherein the one or more burst features include at least one of a burst count, a burst rate, a burst band frequency or frequency distribution, an inter-burst interval length, a burst timing or timing pattern, an average burst duration, a burst waveform, and any variations thereof.

11. The method according to claim 1, wherein the machine learning algorithm is a neural network.

12. The method according to claim 11, wherein the neural network is a recurrent neural network.

13. The method according to claim 12, wherein the recurrent neural network is a long short-term memory (LSTM) neural network.

14. The method according to claim 1, wherein the feature threshold is a static threshold.

15. The method according to claim 1, wherein the feature threshold is a dynamic threshold adjusted over time by the computing device.

16. The method according to claim 1, wherein the detection period is between 1 millisecond and 100 milliseconds.

17. The method according to claim 1, wherein the device is at least one of a personal computing device, an Internet of Things (IoT) device, and a moving vehicle.

18. The method according to claim 17, wherein the device is the personal computing device and the input command is a command to initiate a click of the cursor of the personal computing device.

19. The method according to claim 1, wherein the thought of the subject is a thought generated by the subject to move one or more body parts of the subject.

20. The method according to claim 1, wherein the thought is generated by the subject without being prompted to do so such that the control of the device is performed asynchronously.

21. The method according to claim 1, wherein the thought is generated by the subject in response to being prompted to do so such that the control of the device is performed synchronously.

22. The method according to claim 1, wherein the recording device is a non-invasive recording device.

23. The method according to claim 1, wherein the recording device is an invasive recording device.

24. The method according to claim 23, wherein the recording device is an intravascular recording device comprising a plurality of electrodes held by an intravascular carrier configured to be implanted into a vein or a cave of the brain of the subject.

25. The method according to claim 24, wherein the one or more transient oscillation bursts or pseudo-oscillation bursts are detected using electrodes held by an expandable stent or scaffold, and the method comprises applying a weighting factor to the one or more electrodes of the expandable stent or scaffold such that the transient oscillation bursts or pseudo-oscillation bursts detected at the one or more electrodes are weighted more than the transient oscillation bursts or pseudo-oscillation bursts detected at another electrode of the expandable stent or scaffold.

26. The method according to claim 23, wherein the recording device is an implanted microelectrode array.

27. The method according to claim 26, wherein the recording device is a Utah microelectrode array.

28. The method according to claim 23, wherein the recording device is a thin-film electrode array.

29. The method according to claim 23, wherein the recording device is an electrode array configured to be implanted on the brain surface.

30. The method according to claim 1, further comprising training the machine learning algorithm using previous predictions made by the machine learning algorithm and the burst features extracted from a previous detection period to enhance the predictions made by the machine learning algorithm.

31. A system for controlling a device, the system comprising a recording device configured to capture ongoing neural signal recordings of a subject, a computing device having one or more processors communicatively coupled to the recording device wherein the one or more processors are detecting one or more transient oscillation bursts or pseudo-oscillation bursts from the ongoing neural signal recordings of the subject, wherein the one or more transient oscillation bursts or pseudo-oscillation bursts are generated in response to thoughts generated by the subject or changes in mental states induced by the subject. Extracting one or more burst features from the one or more transient oscillation bursts or pseudo-oscillation bursts detected during the detection period; Predicting the thought generated by the subject or the change in the mental state induced by the subject by applying at least one of a machine learning algorithm and a feature threshold to the one or more burst features extracted during the detection period; Transmitting an input command associated with the prediction to the device to control the device A system programmed to perform.

32. When executed by one or more processors, Detecting one or more transient oscillation bursts or pseudo-oscillation bursts from ongoing neural signal recordings of a subject, wherein the one or more transient oscillation bursts or pseudo-oscillation bursts are generated in response to a thought generated by the subject or a change in the mental state induced by the subject; Extracting one or more burst features from the one or more transient oscillation bursts or pseudo-oscillation bursts detected during the detection period; Predicting the thought generated by the subject or the change in the mental state induced by the subject by applying at least one of a machine learning algorithm and a feature threshold to the one or more burst features extracted during the detection period; Transmitting an input command associated with the prediction to the device to control the device One or more non-transitory computer-readable recording media comprising stored instructions for performing steps including.

33. A method for detecting one or more transient oscillation bursts or pseudo-oscillation bursts, comprising: Filtering raw electrical signals in one or more desired frequency bands using one or more frequency decomposition methods; For each of the desired frequency bands, converting the voltage value of the filtered raw electrical signal into a magnitude or power-related value; Applying at least one of a power threshold for the magnitude or power-related value for each of the desired frequency bands or a duration threshold for each of the desired frequency bands; For each of the desired frequency bands, in response to at least one of the magnitude or power-related value exceeding the power threshold and the filtered raw electrical signal exceeding the duration threshold, identifying one of the transient oscillation bursts or pseudo-oscillation bursts A method comprising.

Images