System and method for seizure detection based on change in electroencephalogram (EEG) non-linearity

Abstract

To solve the problem that seizure detection is sometimes difficult because of the absence of visible indication that a patient is experiencing a seizure.SOLUTION: A seizure detection system includes one or more circuits configured to receive an electroencephalogram (EEG) signal generated based on electrical brain activity of a patient. The one or more circuits are configured to determine metrics based on the EEG signal, the metrics indicating non-linear features of the EEG signal, determine that the EEG signal indicates a candidate seizure by determining, based at least in part on the metrics, a change in the non-linear features of the EEG signal over time, and generate a seizure alert indicating that the EEG signal indicates the candidate seizure. The change in the non-linear features indicates a physiological force that gives rise to the candidate seizure.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] This disclosure relates generally to electroencephalogram (EEG) analysis and, more particularly, to electroencephalogram analysis for seizure detection in a patient. [Background technology]

[0002] CROSS-REFERENCE TO RELATED PATENT APPLICATIONS This application claims the benefit of and priority to U.S. Provisional Patent Application No. 62 / 890,497, filed Aug. 22, 2019, which is incorporated by reference in its entirety.

[0003] Seizures commonly occur in patients with a wide range of medical problems. Over 50 million people worldwide suffer from seizures. In some cases, seizures may be benign, but in extreme cases, seizures can be life-threatening. It is therefore important to detect and address seizures. The earlier a seizure is detected and treated, the better the patient's outcome. However, detecting a seizure can be difficult, as there may be no visible signs that the patient is experiencing a seizure. In particular, it can be difficult to visually detect that patients in intensive care units and young patients (children or infants) are experiencing a seizure.

[0004] Thus, often times, recordings of such patients' EEG data may be collected for analysis by an epileptologist, and in some cases, up to 24 hours of continuous EEG data recording may be required for manual analysis by an epileptologist. Manual analysis of such large amounts of data may be cumbersome, time-consuming, and costly. Although several EEG analysis algorithms exist for seizure detection, these algorithms perform poorly in young children. For example, some seizure detection algorithms may achieve 80% detection rates in adults but only 50-60% detection rates in young children. Such algorithms may also experience a large number of false positive rates, and in some cases, more than 100 false positives per day for a single patient when the patient is a young child. This number of false positives necessitates manual review of all recordings for children, as analysis algorithms do not adequately reduce the amount of manual EEG data review required. The failure of such EEG detection algorithms in children is due in part to the high variability in the nature of the abnormal EEG waveforms recorded for children. Summary of the Invention [Means for solving the problem]

[0005] One implementation of the present disclosure is a seizure detection system including one or more circuits. The one or more circuits are configured to receive an electroencephalogram (EEG) signal generated based on electrical brain activity of a patient. The one or more circuits are configured to identify a metric based on the EEG signal. The metric is indicative of a nonlinear characteristic of the EEG signal. The one or more circuits are configured to identify the EEG signal as indicative of a candidate seizure by identifying a change over time in the nonlinear characteristic of the EEG signal based at least in part on the metric, and generate a seizure alert indicating that the EEG signal may be indicative of a candidate seizure. As used herein, a "candidate seizure" may refer to any seizure, epileptic discharge, subclinical event, potential seizure for technician review, and the like. The change in the nonlinear characteristic is indicative of physiological forces causing a candidate seizure. The time from the occurrence of the change in the nonlinear characteristic to the occurrence of the candidate seizure may vary.

[0006] In some embodiments, the processing circuitry is configured to identify the EEG signal as indicative of a candidate seizure based on at least one of default parameter values ​​or user-defined parameter values.

[0007] In some embodiments, the seizure detection system is a cloud-based system, and the one or more circuits are configured to receive EEG signals from a local EEG collection system over a network and provide resultant data to the local EEG collection system over the network.

[0008] In some embodiments the seizure detection system is a local system, hi some embodiments the local system is integrated with a local EEG system or locally connected to an EEG acquisition system.

[0009] In some embodiments, identifying a change in the nonlinear feature of the EEG signal based at least in part on the metric includes identifying an increase in the nonlinear feature over time. In some embodiments, the method includes identifying an overall change in the nonlinear feature (e.g., a value indicative of how much the nonlinear feature has changed). In some embodiments, the method includes identifying an overall increase or decrease in the nonlinear feature (e.g., for each of the nonlinear features) over a period of time. In some embodiments, the method includes identifying whether the trajectory of the nonlinear feature is increasing or decreasing over the period of time.

[0010] In some embodiments, the metric includes at least one of the following: dimension, synchrony, Lyapunov exponent, entropy, global nonlinearity, distance difference between recursive trajectories, self-similarity, or eigenvalue.

[0011] In some embodiments, the one or more circuits are configured to identify that the EEG signal indicates a seizure alert by performing a preliminary analysis using one of the metrics to indicate that the EEG signal indicates a candidate seizure or that the EEG signal includes noise, and performing a secondary analysis using one or more of the metrics to identify whether the EEG signal indicates a candidate seizure or includes noise and / or other artifacts, and by identifying changes over time in non-linear characteristics of the EEG signal based at least in part on the metrics.

[0012] In some embodiments, the one or more circuits are configured to determine a probability of a trajectory of each of the plurality of metrics at the plurality of time points and determine whether the trajectory of each of the plurality of metrics is significant based on the probability. In some embodiments, determining the change in the nonlinear feature of the EEG signal based at least in part on the plurality of metrics includes mapping a significant metric of the plurality of metrics to a category, the category resulting in a seizure alert. In some embodiments, determining the change in the nonlinear feature may include determining a trajectory of the nonlinear feature. The trajectory may be a pattern of change of the metric over time. The trajectory of the metric may be determined by plotting (or recording) values ​​of the metric. In some embodiments, determining the trajectory includes plotting values ​​of the metric in a phase space and determining the trajectory of the metric from the plot of the phase space.

[0013] In some embodiments, the one or more circuits are configured to identify a dimensionality of the EEG signal by performing a phase space analysis by increasing a value of the dimension until the number of false neighbors reaches zero, where an initial value of the dimension is based on the age of the patient.

[0014] In some embodiments, the one or more circuits are configured to determine whether one or more of the metrics exhibit a statistically significant change over time and generate a user interface. In some embodiments, the user interface includes a real-time trend of the EEG signal and one or more of the metrics. In some embodiments, the one or more circuits are configured to cause a user interface device to display the user interface.

[0015] In some embodiments, one of the metrics is an eigenvalue and the user interface further includes a trend of the eigenvalue, hi some embodiments, the eigenvalue may be plotted along with the EEG signal such that the user interface can provide the operator with a view of the EEG signal and the trend of the eigenvalue.

[0016] In some embodiments, the user interface further comprises a history window of EEG signals associated with the candidate seizure.

[0017] In some embodiments, the one or more circuits are configured to identify a moving window of eigenvalues ​​in the EEG signal, identify that the eigenvalue is decreasing, and in response to identifying that the eigenvalue is decreasing, identify that the EEG signal is consistent with a candidate seizure based at least in part on the metric.

[0018] In some embodiments, the one or more circuits are configured to determine a Renyi permutation entropy value (and / or any other type of entropy measurement) based on the EEG signal, determine that the Renyi permutation entropy value is decreasing, and determine that the EEG signal is indicative of a candidate seizure in response to a determination that the eigenvalues ​​are decreasing and a second determination that the Renyi permutation entropy value is decreasing.

[0019] In some embodiments, the one or more circuits are configured to determine a Renyi permutation entropy value based on the EEG signal, determine that the Renyi permutation entropy value is increasing, determine a sample entropy value based on the EEG signal in response to a first determination that the Renyi permutation entropy value is increasing, determine that the EEG signal is indicative of a candidate seizure in response to a second determination that the sample entropy value is decreasing (e.g., negative), and determine that the EEG signal does not indicative of a candidate seizure in response to a third determination that the sample entropy value is increasing (e.g., positive).

[0020] Another implementation of the present disclosure is a method of seizure detection, the method including: receiving, by a processing circuit, an electroencephalogram (EEG) signal generated based on electrical brain activity of a patient; determining, by the processing circuit based on the EEG signal, a metric indicative of a nonlinear characteristic of the EEG signal; determining, by at least in part based on the metric, that the EEG signal is indicative of a candidate seizure by determining a change over time in the nonlinear characteristic (e.g., each of the nonlinear characteristics); and generating, by the processing circuit, a seizure alert indicating that the EEG signal is indicative of a candidate seizure. The change in the nonlinear characteristic reflects a physiological force causing the candidate seizure.

[0021] In some embodiments, the processing circuit's identification of the EEG signal as indicative of a candidate seizure is based on at least one of default parameter values ​​or user-defined parameter values.

[0022] In some embodiments, the metric includes at least one of the following: dimension, synchrony, Lyapunov exponent, entropy, global nonlinearity, distance difference between recursive trajectories, self-similarity, or eigenvalue.

[0023] In some embodiments, the processing circuitry identifies the EEG signal as indicative of a candidate seizure by identifying changes over time in a non-linear characteristic of the EEG signal based at least in part on the metrics by performing a preliminary analysis using one of the metrics to indicate that the EEG signal is indicative of a candidate seizure or that the EEG signal is noisy, and performing a secondary analysis using one or more of the metrics to identify whether the EEG signal is indicative of a candidate seizure or is noisy.

[0024] In some embodiments, the method further includes the processing circuit determining a probability of a trajectory of each of the plurality of metrics at the plurality of time points, and the processing circuit determining whether the trajectory of each of the plurality of metrics is significant based on the probability. In some embodiments, the processing circuit determining a change in a non-linear feature of the EEG signal (e.g., a change in the trajectory of the non-linear feature) based at least in part on the plurality of metrics includes mapping a significant metric of the plurality of metrics to a category, the category being a seizure category.

[0025] In some embodiments, the method includes the processing circuitry identifying a dimensionality of the EEG signal by performing a phase space analysis by increasing a value of the dimension until a number of false neighbors reaches zero, where an initial value of the dimension is based on an age of the patient.

[0026] Another implementation of the present disclosure is a seizure detection system including one or more electrodes connected to a patient, the electrodes configured to generate an electroencephalogram (EEG) signal based on the patient's electrical brain activity. The system further includes a processing circuit configured to receive the electroencephalogram (EEG) signal generated based on the patient's electrical brain activity, determine a metric indicative of a nonlinear feature of the EEG signal based on the EEG signal, determine that the EEG signal is indicative of a candidate seizure by determining a change in the nonlinear feature (e.g., each of the nonlinear features) over time based at least in part on the metric, and generate a seizure alert indicating that the EEG signal is indicative of a candidate seizure. The change in the nonlinear feature is indicative of physiological forces causing the candidate seizure.

[0027] In some embodiments, the processing circuitry is configured to identify the EEG signal as indicative of a candidate seizure based on at least one of default parameter values ​​or user-defined parameter values.

[0028] In some embodiments, the processing circuitry is configured to identify the EEG signal indicative of a candidate seizure by: identifying a trajectory of one of the nonlinear features by identifying a value of the nonlinear feature over time; determining that a probability value of the trajectory is less than a probability level indicating that the occurrence of the trajectory is statistically significant (e.g., a predefined critical probability level set by a classification criterion); and, in response to identifying the probability value of the trajectory being less than the probability level, identifying that the EEG signal indicative of a candidate seizure. In some embodiments, the probability levels of each of the trajectories of the multiple nonlinear features are compared to the probability levels to identify whether the trajectory is each statistically significant and / or whether the signal indicative of a candidate seizure.

[0029] In some embodiments, each value of the non-linear feature decreases relative to a previous value of the non-linear feature value (i.e., each value of the pattern of non-linear feature values ​​decreases relative to a previous value). In some embodiments, the processing circuitry is configured to determine a probability value of the trajectory based on the number of values ​​of the non-linear feature that each decrease relative to a previous value of the non-linear feature value (i.e., the number of successively decreasing values).

[0030] In some embodiments, the processing circuitry is configured to receive user input through a user interface and determine the probability level by setting the probability level to a value selected by the user via the user input.

[0031] In some embodiments, the processing circuitry is configured to obtain a default value from a memory device of the seizure detection system and determine the probability level by setting the probability level to the default value obtained from the memory device.

[0032] In some embodiments, the processing circuitry is configured to identify that the EEG signal is indicative of a candidate seizure by: identifying a respective trajectory of the non-linear features by identifying values ​​of the respective non-linear features over time; determining that a probability value of the respective trajectory of the non-linear features is less than a probability level indicating an occurrence of the respective trajectory of the non-linear features is statistically significant; and in response to determining that the probability value of the respective trajectory of the non-linear features is less than the probability level, identifying that the EEG signal is indicative of a candidate seizure. [Brief description of the drawings]

[0033] Various objects, aspects, features, and advantages of the present disclosure will become more apparent and be better understood by reference to the detailed description in conjunction with the accompanying drawings, in which like reference numbers identify corresponding elements throughout and generally indicate identical, functionally similar, and / or structurally similar elements.

[0034] [Figure 1] FIG. 1 is a block diagram of a local EEG system including a seizure detector for candidate seizure detection based on trends in non-linear features in EEG signals, according to an exemplary embodiment.

[0035] [Diagram 2] 1 is a block diagram of an EEG acquisition system for collecting EEG data and an analysis system including a seizure detector for analyzing the EEG data to detect candidate seizures, in accordance with an illustrative embodiment.

[0036] [Diagram 3] FIG. 1 is a block diagram of a remote cloud-based system including a seizure detector for candidate seizure detection, according to an exemplary embodiment.

[0037] [Figure 4] FIG. 4 is a block diagram of the seizure detector of FIGS. 1-3 shown in greater detail, according to an exemplary embodiment.

[0038] [Diagram 5] 5 is a flow diagram of a process for detecting candidate seizures by identifying changes in non-linear characteristics of an EEG signal that may be performed by the seizure detector of FIG. 4 in accordance with an exemplary embodiment.

[0039] [Figure 6] 5 is a flow diagram of a process for detecting candidate seizures by identifying changes in non-linear features of an EEG signal using eigenvalues, Renyi permutation entropy, and sample entropy that can be performed by the seizure detector of FIG. 4 in accordance with an exemplary embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0040] overview Referring generally to the figures, a system and method for seizure detection based on changes in EEG nonlinearity is shown, according to various exemplary embodiments. The seizure detector is configured in some embodiments to detect whether a candidate seizure is present in the EEG data by analyzing changes in nonlinearity over time, i.e., by detecting changes (e.g., increases) in the nonlinearity of the EEG data over time. The seizure detector is configured in some embodiments to analyze the EEG signal to identify whether there are epochs or events in the EEG signal that can be presented to an epileptologist or other clinical technician for manual review to ultimately identify whether a candidate seizure has occurred. The seizure detector provides significant time savings in the evaluation of patients undergoing EEG monitoring for seizure detection.

[0041] The seizure detector can detect physiological forces that cause abnormal waveforms, such as a set of diverse morphologies of spikes and sharp waves in the EEG data, rather than trying to detect these abnormal waveforms. In some embodiments, a seizure detector configured to detect candidate seizures by nonlinearity performs an algorithm that focuses on direct detection of abnormal waveform patterns, especially when the patient is a child or infant. In children and infants, abnormal waveforms can be very different and diverse, and it can be difficult for a seizure detector to remember the symptoms of each abnormal waveform. Rather than detecting specific abnormal waveforms, the seizure detector can detect physiological forces that cause candidate seizures by monitoring trends in nonlinearity of the EEG signal.

[0042] Seizure detectors can utilize physiological or mathematical models of the seizure process. Seizures result from abnormal interactions (i.e., nonlinear behavior) between neuronal populations. These interactions are in contrast to the seemingly additive or linear interactions present in many non-seizure conditions. These interactions are also dynamic in the sense that they develop during a seizure. Thus, seizure occurrence reflects or is driven by abnormal nonlinear transformations between neuronal populations that develop along with the development of the seizure. The branch of mathematics that deals with this type of phenomenon is called nonlinear dynamical systems analysis.

[0043] In some embodiments, the seizure detection performed by the seizure detector can be used (or integrated with) another system to activate a dedicated closed-loop stimulator to stop the patient's seizures. In some embodiments, the seizure detector can also be used to screen a long-term historical record of the patient's EEG data, which can provide immediate feedback to the patient or caregiver and lead to early intervention in seizures. The seizure detector can also be used when trained epileptologists are not available for manual review of EEG signal data. The shortage of neurologists to review EEG signals is high in the United States and even higher outside the United States. It is estimated that approximately 200 pediatric intensive care units (ICUs) and 800 neonatal ICUs in the United States do not have properly trained personnel to read EEG data.

[0044] Reliance on visual cues that patients pick up on (e.g., twitches, changes in muscle tone, etc.) is problematic. Observers (e.g., family members, nurses, technicians, etc.) often fail to detect a patient's seizures by visual cues, and as many as 30 to 40% of seizures are often missed by observers. In part, this occurs because many seizures are non-convulsive and therefore more difficult to detect visually. Detecting seizures by visual cues is particularly difficult in settings such as intensive care units, where patients may have impaired ability to communicate non-convulsive events (e.g., confusion, aphasia, visual impairment, or other sensory impairments) or may be too young to communicate these events or may be paralyzed during the course of their illness. It is estimated that approximately 30 to 80% of seizures in the ICU may fall into one of these non-convulsive categories.

[0045] Nearly 85% of adult seizures originate from the temporal lobe and generally affect the hippocampus. However, infant and child seizures occur more widely from an anatomical point of view and their temporal morphology is much more diverse compared to adults. Therefore, methodologies that are applied to adults, i.e., morphology-based analysis, are not necessarily applicable to children. An example of morphology-based analysis may include analysis of periodicity present in the EEG data. Often, as a seizure progresses, the amplitude of the EEG increases, so the analysis system may analyze the overall amplitude. Analysis of the periodicity or amplitude indicates the build-up of the seizure. Other analysis algorithms may be linear time-domain analysis based on machine learning of specific seizure patterns in patients. A partial list of these methods includes independent component analysis, morphological analysis, template matching, mimicking methods, parametric approaches, clustering techniques, and knowledge-based rules. However, such analysis may be less successful in children. Instead, the nonlinear trend analysis of the seizure detector described herein may be applied to children. Such nonlinear trend analysis may also be applied to adults.

[0046] There is little uniformity in the clinical presentation of seizures in children. Some have no visible correlates, some are dramatic in appearance (e.g., drop attacks), while others include multiple behavioral components (e.g., Lennox-Gastaut syndrome). Similarly, there are a variety of electrographic patterns or features seen in pediatric seizures. There are the typical rhythmic build-up patterns seen in adults, periods of voltage suppression (e.g., infantile spasms), "stop-start" patterns, high amplitude rhythmic decelerations, polyspike bursts, and more. Multiple patterns may be seen in a single patient. Efforts to model or template match all these patterns and pattern transitions have not been successful. Instead of using waveform morphology as a determinant of the mathematical treatment of detection, the seizure detector described herein is, in some embodiments, configured to detect the morphology of the underlying physiological forces that produce these diverse changes.

[0047] Also, the same "driving function" may give different results depending on the background EEG (reflecting the dependence on the initial state). The moment-to-moment components of the EEG vary so widely that the action of the driving function on a variable input results in a possibly variable output. The system behavior is a combined function of the shape of the background activity, the shape of the physiological forces that cause state transitions, and any external variables (e.g., elevated temperature, the presence of drugs, metabolic parameters, etc.). Obtaining the shape of the driving function is easy during short stationary periods, but during transitional periods it is more difficult to specify its shape.

[0048] It should be noted that artifacts such as electromyogram (EMG), movement, patting, electrode "popping", etc. should not have a preceding pattern of trajectory change as seen in seizures. Although the observed artifacts or other spurious waveform morphology may show visual similarities to seizures, these artifacts and other spurious events should not show the same patterns of physiological trajectory development seen in epilepsy. The seizure detector is configured in some embodiments to separate spurious events from seizure events to minimize false alarms.

[0049] Seizure detection 1, a system 100 is shown that includes an EEG system 104 that includes a seizure detector 108 for candidate seizure detection based on non-linear feature trends, according to an exemplary embodiment. In the system 100, signal processing firmware and / or software is integrated into an EEG data acquisition system with or without additional signal processing boards to form the EEG system 104. The EEG system 104, in some embodiments, is configured to collect EEG data from a patient 102 and further detect candidate seizures based on the EEG data using the seizure detector 108 (e.g., detect possible candidate seizures for review by a user). The seizure detector 108, in some embodiments, is a fully integrated parallel processor.

[0050] 1 shows a patient 102 with multiple electrodes attached to the head of the patient 102. The electrodes sense electrical brain activity of the patient 102. The patient 102 may be a human, such as an adult, a teenager, a child, an infant, etc. The patient 102 may also be an animal, such as a cat, a dog, a horse, a cow, etc. The number of electrodes attached to the patient 102 for collection of EEG data for analysis by the seizure detector 108 depends on the desired localization accuracy (localization accuracy is less important if the focus is on detection), the dimensionality of the driving function to be identified by the seizure detector 108, the physical limitations of the skull size, the spatial distribution of the electrodes, the spatial extent of the source, and the correlation structure between the electrodes, etc.

[0051] The electrodes, in some embodiments, are connected to an electrode interface 106 included in the EEG system 104. The electrode interface 106 may include one or more auxiliary hardware circuits for generating EEG data for analysis by the seizure detector 108. The hardware circuits may include amplifier circuits (e.g., differential amplifier circuits), filters (e.g., high pass, low pass, band pass), analog to digital converters (ADCs), etc.

[0052] In some embodiments, the seizure detector 108 may be configured to analyze signals generated by the full set of electrodes attached to the patient 102 and / or to analyze a subset of the electrodes attached to the patient 102. In some embodiments, because the dimension of a seizure generally does not exceed a value of four, about 10 or fewer electrodes may be analyzed by the seizure detector 108 to detect a seizure. In this regard, even if a technician attaches a full set of electrodes, the seizure detector 108 may select the appropriate number of electrodes needed (e.g., select 10 electrodes).

[0053] In some embodiments, the seizure detector 108 is configured to identify a Lyapunov spectrum that typically varies from about 2 to 9, with most seizures showing a decrease in dimensionality with the onset of the seizure. It is rare to see more than 3 dimensions during a seizure. Using multi-channel EEG techniques, a trajectory can be characterized with 2d+1 electrodes, where d is the estimated dimensionality of the underlying function using the Lyapunov spectrum. During operation, the seizure detector 108 can identify the dimensionality of the underlying function and cause the user interface device 110 to recommend a specific number of electrodes to the patient 102. In this regard, the patient 102 can start with a predefined number of electrodes, but following the analysis of the seizure detector 108, a technician can add additional electrodes to the patient 102 based on the identified dimensionality.

[0054] The EEG data may represent one or more EEG signals for brain activity of the patient 102. The seizure detector 108 may receive the EEG data and perform a nonlinear analysis of the EEG data to detect whether the EEG data is indicative of a candidate seizure that has occurred, is occurring, or is occurring in the patient 102. The detection of a candidate seizure detected by the seizure detector 108 may be provided to the user interface device 110 for visual and / or audio notification to a user, e.g., a doctor, a nurse, a family member of the patient 102, an epileptologist, a technician, etc. The user may also provide configuration data through the user interface device 110. The configuration data may indicate an age of the patient 102, a weight of the patient 102, past EEG data of the patient 102, a medical condition of the patient, etc. The nonlinear analysis that the seizure detector 108 is configured to perform may be based at least in part on the configuration data.

[0055] The user interface device 110 may be a system or device configured to receive input from a user and / or provide output to a user. The user interface device 110 may be a monitor, e.g., a display screen. The display screen may be a light emitting diode (LED) screen, a cathode ray tube display (CRT), a liquid crystal display (LCD), and / or any other type of display screen. The user interface device 110 may further include an input device, a mouse, a keyboard, a touch screen, etc. The user interface device 110 may also include a speaker for audio output, a microphone for audio input, etc. In some embodiments, the user interface device 110 is a computer, a smartphone, a tablet, etc. in communication with the EEG system 104 and / or the seizure detector 108.

[0056] In principle, if a particular sequence of events leading to the occurrence of a seizure is known, in some embodiments the system is configured to search for that particular sequence of events. For example, much is known about the abnormal electrical behavior of single neurons in anatomical regions that cause seizures. For example, temporal lobe seizures are especially recognized in adults. However, seizures occur due to the dysfunction of networks or ensembles of brain cells. Thus, instead of searching for known morphological patterns (e.g., activity in specific regions of the brain, sharp spikes of activity, etc.), the seizure detector 108 can analyze populations of behaviors in search of the driving forces behind the onset of a seizure in the patient 102. The forces behind the physiology of a seizure are not random. In fact, the forces are deterministic and can be detected by the seizure detector 108 by activating nonlinear dynamical systems tools.

[0057] The seizure detector 108 is configured to apply seizure detection to any range of ages and, in some embodiments, can be performed in real time, and the accuracy of the seizure detector 108 can be greater than 90% and less than two orders of magnitude false positives in EEG data collected on the patient 102 over a 24-hour period. The seizure detector 108 can be implemented locally (as shown in FIGS. 1 and 2) and / or remotely (as shown in FIG. 3).

[0058] In some embodiments, the seizure detector 108 is configured to select parameter values ​​for seizure detection and / or event classification by the seizure detector 108 based on user input (instead of or in addition to default values ​​programmed into the seizure detector 108). In some embodiments, the parameter values ​​may be manually selected by a user, who provides user input through a user interface 110 associated with the seizure detector 108. The parameter values ​​may be a statistical significance level of the trajectory and / or a metric parameter value between component metrics when multiple metrics are simultaneously applied to a data set. By selecting the parameter values ​​based on user input, false alarms generated by the seizure detector 108 may be reduced or a hit rate by the seizure detector 108 may be increased. Also, by allowing the user to select the parameter values, an appropriate tradeoff between false positives and true positives may be made by the seizure detector 108.

[0059] The user input can indicate a balance level (e.g., weight) between reducing false positives and increasing hit rate. This can be achieved by using suggested default values ​​or adjusting the values, which can be based on personal preferences or to best suit a particular patient's situation. The balance level can be a range of values ​​and can correspond to lower or higher statistical significance levels (e.g., a balance level that favors reducing false positives can be associated with a lower probability value of the trajectory of the metric changing in a particular direction (increasing or decreasing), whereas a balance level that favors increasing hit rate can be associated with a higher probability value of the trajectory of the metric change).

[0060] The seizure detector 108, in some embodiments, is configured to detect shifting patterns of forces that result in state transitions from non-seizure to seizure without attempting to detect a target waveform morphology. These abnormal physiological forces generate waveform trajectories that the seizure detector 108, in some embodiments, is configured to quantify. The seizure detector 108 can utilize the trajectories to detect multiple state changes, including seizure state changes, i.e., changes from a normal state of the patient 102 to a seizure state. More specifically, the seizure detector 108, in some embodiments, is configured to identify one or more nonlinear metrics based on the EEG data that reflect the occurrence of these trajectories. The seizure detector 108, in some embodiments, is configured to activate nonlinear dynamical system tools to detect the occurrence of these abnormal trajectories.

[0061] The seizure detector 108, in some embodiments, is configured to search for seizures in the EEG time series by searching for specific categories of state changes. More specifically, the seizure detector 108, in some embodiments, is configured to search for changes, i.e., structural nonlinearity changes, in the EEG data. The seizure detector 108 is configured to utilize nonlinear methods to detect state changes in the mathematical state space, i.e., starting points for the reconstruction of the system dynamics (the dynamics of the brain activity of the patient 102).

[0062] The seizure detector 108, in some embodiments, is configured to detect candidate seizures when the patient 102 has a gradual and / or sudden transition into a seizure state. More specifically, the seizure detector 108 may utilize dynamic systems analysis to detect both sudden changes (e.g., bifurcations) as well as many forms of gradual changes. The search for seizures is not a search for a specific isolated event or a search for a specific single value of a feature; instead, the seizure detector 108 may identify multiple nonlinear metrics and track the nonlinear metrics over time to detect diagnostic shifts and patterns of change in the nonlinear features of the EEG data. For example, the seizure detector 108 may identify whether a statistically significant increasing or decreasing trajectory of the nonlinear feature is occurring. For example, for a metric that indicates nonlinearity, seven consecutive increasing values ​​of the metric may be statistically significant to indicate an increasing trajectory. However, five consecutive increases in the value of the metric may not be statistically significant to indicate an increasing trajectory. Similarly, the number of consecutive decreasing values ​​may be related to the probability of occurrence; for example, five consecutive decreasing values ​​of a metric may not be significant, but seven consecutive decreasing values ​​of a metric may be statistically significant.

[0063] For example, the probability of five consecutive increasing values ​​of the metric may be 0.032 (which may be determined by the seizure detector 108 from the five consecutive increasing values ​​and / or historical trajectory data). The probability of seven consecutive increasing values ​​of the metric may be 0.01. Because the probability of seven consecutive increasing values ​​is less than the probability of five consecutive increasing values, seven consecutive increasing values ​​may be more statistically significant (lower probability level) than five consecutive increasing values. To determine whether an increase or decrease in the metric is statistically significant, e.g., whether the probability of occurrence is less than a probability threshold, the seizure detector 108 may utilize a probability threshold. Also, the seizure detector 108 may utilize a change count threshold, i.e., the number of consecutive increasing values ​​or consecutive decreasing values ​​equal to or greater than the change count threshold, i.e., if the threshold is 7, then seven consecutive increasing values ​​of the metric are statistically significant, but five consecutive increasing values ​​of the metric are not statistically significant.

[0064] The threshold for determining statistical significance can define the amount of false positives and missed seizure detections. For example, a threshold that requires many consecutive increasing values ​​or consecutive decreasing values ​​may result in fewer false positives but miss many seizures. However, a lower threshold may result in more false positives but fewer missed seizures. To appropriately set the threshold for determining statistical significance, the seizure detector 108 can perform optimization. The optimization attempts to minimize missed seizures and minimize false positives. The optimization can be based on user input, for example, user feedback that identifies certain periods of past EEG signals as corresponding to seizures or identifies other periods of EEG signals as associated with false positives.

[0065] The seizure detector 108 is configured in some embodiments to detect changes in the pattern of nonlinear dynamics of the EEG data, since the pattern of changes is an invariant feature of seizure state transitions. Often, the state change of the nonlinearity precedes in time the appearance of spikes, sharp waves, or other visual signs in the electrographic or clinical EEG data of a seizure. Thus, the seizure detector 108 is configured in some embodiments to first identify nonlinearity changes from the EEG, i.e., nonspecific detectors of eigenvalues. If the nonspecific detectors indicate a candidate seizure, the seizure detector 108 can utilize subsequent metric calculations and / or analyses. This allows the seizure detector 108 to conserve computational resources by utilizing computations with low computational requirements, e.g., eigenvalues, followed by computations with high computational requirements, e.g., dimensions.

[0066] Due to potential instability of multiple non-linear measurements with small sample sizes, the seizure detector 108 is configured in some embodiments to utilize a moving window for the calculation of the metric. The specific values ​​of the moving window duration and overlap ratio within the window can be predefined based on the specific metric, i.e., each metric can be associated with its own window duration and overlap ratio. To ensure the stability of the estimation, the more dependent a particular metric is on the sample size, the seizure detector 108 is configured in some embodiments to identify the metric with a longer window duration.

[0067] The seizure detector 108, in some embodiments, is configured to analyze changes in eigenvalues ​​to detect seizures. However, changes in eigenvalues ​​may result from either a quantitative change in the ratio of linear to nonlinear activity in the EEG data or the presence of noise in the EEG data. Thus, the seizure detector 108 may identify and analyze multiple nonlinear metrics together to detect candidate seizures. For example, the seizure detector 108, in some embodiments, is configured to identify entropy along with eigenvalues ​​to help distinguish between seizures and noise. If the noise is not rhythmic, it often increases entropy, while most seizures decrease entropy.

[0068] The seizure detector 108, in some embodiments, is configured to identify and analyze many other non-linear metrics. The metrics that the seizure detector 108 is configured to analyze can be based on configuration data, i.e., the clinical profile or clinical syndrome of the patient 102 (e.g., drop attacks, infantile spasms, Lennox-Gastaut syndrome, post-hypoxic encephalopathy, age, weight, etc.), and baseline electroencephalogram patterns associated with the patient 102. The seizure detector 108, in some embodiments, is configured to analyze the particular configuration data and identify and / or analyze a metric appropriate for the patient 102.

[0069] Clinical syndromes and baseline EEG patterns (e.g., EEG patterns of normal brain activity, seizure patterns, etc.), patient age, patient weight, etc. may be included in the configuration data and utilized by the seizure detector 108 in selecting the configuration of the mixture of nonlinear metrics (or mixture weightings) in the second stage (and / or preliminary stage). Candidate metrics include, but are not limited to, dimensionality, synchrony, Lyapunov exponents, various forms of entropy, global nonlinearity (by proxy testing), distance difference between recursive trajectories in phase space, self-similarity, etc.

[0070] The output of the analysis performed by the seizure detector 108 can be a panel of non-linear values ​​that change over time. Some of these patterns may be indicative of candidate seizures, while other patterns reflect sleep onset and other artifacts. Thus, the seizure detector 108 can map the panel of non-linear values ​​to specific categories, e.g., seizure, noise, sleep, etc. The number of metrics in the panel can be set by the seizure detector 108 based on the signal processing capabilities of the EEG system's hardware and / or firmware architecture. The selection of metrics can be changed based on whether the seizure detector 108 is operating in a real-time mode, analyzing EEG data in real-time, or in a historical analysis mode, analyzing previously recorded EEG data.

[0071] 2, a system 200 is shown that includes an EEG acquisition system 206 for collecting EEG data and an analysis system 204 that includes a seizure detector 108 for analyzing the EEG data to detect candidate seizures, according to an exemplary embodiment. In system 200, the signal processing hardware, firmware and / or software of seizure detector 108 is separate from EEG acquisition system 206, i.e., fully integrated into a standalone local computer within analysis system 204.

[0072] Analysis system 204 is configured to operate with EEG collection system 206 using the output of the EEG collection system, i.e., the EEG data collected by EEG collection system 206, in some embodiments. System 200 can be implemented in multiple embodiments, for example, analysis system 204 can be a screening device with a simplified headbox for EEG collection system 206 and limited signal processing capabilities. The headbox can be configured to be located on top of the enclosure of EEG collection system 206. System 200 may be suitable for alerting and / or screening in a hospital or in the patient's home. In some embodiments, analysis system 204 is a plug-in card (e.g., a circuit board configured with a connection port that can be connected to a connection port of EEG collection system 206). A user can insert the plug-in card into EEG collection system 206 to give EEG collection system 206 all the operating power of analysis system 204. For example, the plug-in card can include graphics or digital signal processing circuitry and a memory that includes instructions for performing the operations described herein.

[0073] EEG collection system 206 may include collection manager 202. Collection manager 202 is configured to collect EEG data and maintain a historical record of the EEG data. EEG collection manager 202 may also provide the EEG data to analysis system 204 for analysis and seizure detection. Upon receiving a request from analysis system 204, collection manager 202 may provide the requested historical EEG data stored by collection manager 202 to analysis system 204.

[0074] 3, there is shown a system 300, a cloud-based implementation of the seizure detector 108, according to an exemplary embodiment. In the system 300, seizure detection and related signal processing is performed at a remote site, i.e., by a cloud platform 306. The cloud platform 306 can be one or more remote and / or local servers in the hospital, and can be a cloud analytics system such as MICROSOFT AZURE, AMAZON WEB SERVICES, etc.

[0075] EEG system 100 includes a network interface 302 that communicates EEG data and / or configuration data to cloud platform 306 for analysis by seizure detector 108 via network 304. Network 304 may act as a pipeline between EEG system 100 and cloud platform 306 where seizure detector 108 performs feature extraction and / or analysis. Results of the analysis performed by analysis system 204 may be sent back to EEG system 100 for display by user interface device 110 and decision-making by the user.

[0076] The network 304 may include one or more different wired and / or wireless networks. The network may be a local area network (LAN) or a wide area network (WAN). The network may be wired, including Ethernet wires, cables, and / or fiber optic connections, and / or may be wireless, including Wi-Fi and / or cellular-based networks. The network interface 302, in some embodiments, may include one or more receivers, transmitters, transceivers, radios, signal processing circuitry, etc., configured to operate the network interface 302 to communicate over the network 304.

[0077] 4, a system 400 is shown including a seizure detector 108, according to an exemplary embodiment. The seizure detector 108 is shown to receive configuration data and EEG data. The seizure detector 108 is also shown to output a user interface and cause a user interface device 110 to display the user interface. The user interface may include an indication of the presence of a candidate seizure and / or a display of calculated metrics that the seizure detector 108 determines from the EEG data.

[0078] The seizure detector 108 includes an analysis circuit 428. The analysis circuit 428 may include one or more processing circuits for digital signal processing. The analysis circuit 428 may include a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), one or more central processing units (CPUs), one or more digital signal processing (DSP) units, one or more graphic processing units (GPUs), etc. The seizure detector 108 may have high processing demands, and the seizure detector 108 may utilize shared computing across multiple processing units (e.g., separate processing cards, graphics cards, remote servers, cloud-based systems, etc.).

[0079] The analysis circuit 428 may also include one or more memory devices. The memory devices may store instructions and / or computational data for execution by one or more processors. The memory devices may include random access memory (RAM), solid state drives (SSD), hard disk drives (HDD), flash memory, electrically erasable programmable ROM (EEPROM), and / or any other type of memory, temporary or non-temporary.

[0080] The seizure detector 108, in some embodiments, is configured to detect candidate seizures by analysis of historical data and / or with real-time analysis. The seizure detector 108, in some embodiments, is configured to detect candidate seizures with less than a 15 second delay between seizure onset and detection. The seizure detector 108, in some embodiments, is configured to detect broad electrographic patterns (i.e., a mapping between seizure types and the optimal detection algorithm that the seizure detector 108 performs for that type of seizure). The seizure detector 108, in some embodiments, is also configured to distinguish seizure state transitions from artifacts and noise. The seizure detector 108, in some embodiments, is also configured to detect several seizure types in the same patient, originating from several locations in the same patient (multifocality). The seizure detector 108 may have a true positive rate of over 90% and fewer than 8 false positives per day for a single patient.

[0081] A constant quantitative feature of the transition from a non-ictal to a seizure state is a change in the factor of nonlinearity relative to the energy level of the signal. In many, and most, adults, the transition to a seizure state results in increased rhythmicity or increased synchronization between cell populations. This is reflected in a decrease in eigenvalues ​​(a decrease in the factor of linearity), a decrease in entropy, a decrease in dimensionality, and an increase in overall nonlinearity as revealed by surrogate tests. However, this pattern is not universal. Exceptions to this pattern are particularly evident in children or infants, where existing algorithms fail. For example, in many patients with drop attacks, the electrographic correlation is an initial brief period of high energy slowing followed by low voltage asynchronous activity. The temporal pattern of the quantitative metrics is more complex, showing periods of increased entropy, increased eigenvalues, and decreased overall nonlinearity. For this reason, the seizure detector 108 focuses on the changes in metrics rather than absolute values ​​and utilizes multiple forms of change to detect candidate seizures. This captures a wide range of seizure electrographic morphologies.

[0082] A challenge arises when baseline EEG activity is poorly organized, has excessive slow wave activity, and is interrupted by high voltage sharp waves or spikes (e.g., Lennox-Gastaut syndrome). In this case, the baseline eigenvalues ​​that the seizure detector 108 is configured to detect may be so low that the appearance of a seizure is not reflected in a decrease in the eigenvalues ​​(e.g., floor effect). In this case, the analysis with multiple metrics performed by the seizure detector 108 increases the likelihood of avoiding floor and / or ceiling effects. All these changes can be distinguished from the intrusion of increased noise or artifacts (noise generally reduces eigenvalues ​​and increases entropy).

[0083] The seizure detector 108, in some embodiments, is configured to apply quantitative time-change analysis to multiple metrics to identify patterns of change across multiple metrics, which classifies events as candidate seizures (e.g., review alert), signal noise (no alert), or uncertain classification (probable seizure alert occurs). The seizure detector 108 can be tuned to change the tradeoff between true detection, false alarms, and missed detections by adjusting the significance level of the probability that a significant change needs to be recognized. The detection of the seizure detector 108 can be performed without the use of machine learning, which requires a data collection period with delays in treatment that may be damaging to the patient. In some embodiments, the temporal patterns of the metric trajectories can be post-processed (e.g., smoothed to remove transients) to reduce variability in the application of statistical criteria.

[0084] The analysis circuit 428 may employ a pipeline of analysis stages and may include components configured to employ each stage. The components may be software modules, circuits, etc. The analysis circuit 428 includes a channel selector 402, a filtering stage 404, a preliminary analyzer 406, a secondary analyzer 408, and an interface generator 410. EEG data received by the seizure detector 108 may first pass through the channel selector 402. The channel selector 402 may control which channel of EEG data the seizure detector 108 performs analysis on. For example, if multiple electrodes are present, one or more sets of electrodes may be suitable for analysis by the analysis circuit 428. Thus, the channel selector 402 may select an appropriate EEG signal channel and provide the EEG signal of the selected channel to the filtering stage 404.

[0085] The filtering stage 404 can filter the EEG data using one or more low-pass, high-pass, and / or band-pass filters. The filters can be digital and / or hardware filters, such as infinite impulse response (IIR) and / or finite impulse response (FIR) filters. The bandwidth appropriate for the signal analyzed by the analysis circuit 428 can be specific to the age of the patient 102. Thus, the filtering stage 404, in some embodiments, can receive configuration data indicative of a characteristic (e.g., age) of the patient and is configured to perform filtering based on the configuration data.

[0086] The bandwidth passed by the filtering stage 404 may depend on the patient's age as well as the metrics identified by the seizure detector 108 for detecting candidate seizures. In some embodiments, the filtering stage 404 passes frequencies between 100 Hz and 200 Hz. In some embodiments, the frequency band passed by the filtering stage 404 can range from 2 Hz to 400 Hz.

[0087] The analysis circuit 428, in some embodiments, is configured to determine multiple nonlinear metrics and combine patterns of evolution of the multiple nonlinear metrics to detect candidate seizures by the preliminary analyzer 406 and the secondary analyzer 408, which in some embodiments are configured to concatenate the use of several nonlinear metric algorithms in sequence and / or in parallel. The seizure detector 108 can detect the presence of a candidate seizure by first screening for nonspecific global nonlinear changes, i.e., a metric 430, which may be an eigenvalue. The secondary analyzer 408, in some embodiments, is configured to process more computationally intensive metrics focused on more specific types of nonlinearity by the secondary analyzer 408.

[0088] The preliminary analyzer 406, in some embodiments, is configured to perform a screening step by determining an implementation of a moving window of eigenvalues, e.g., metric 430. The eigenvalues ​​decrease with the appearance of non-linear interactions (e.g., seizures) or with the appearance of noise. The eigenvalues ​​increase when the EEG data is no longer rhythmic or periodic. The preliminary analyzer 406, in some embodiments, is also configured to identify whether a change in the eigenvalue is statistically significant (e.g., has a significance value greater than a predefined amount or a probability of error less than a predefined amount) by determining a statistical significance 432 of a trend in the metric 430 such that only statistically significant changes in the eigenvalues ​​are analyzed by the preliminary analyzer 406 to identify candidate seizures.

[0089] In some embodiments, the preliminary analyzer 406 identifies statistical significance using a moving window, i.e., identifies trends in the metric 430 using a moving window similar to or the same as the moving window based trend analysis described with reference to the secondary analyzer 408. A particular window of samples of the metric 430 allows the preliminary analyzer 406 to identify whether the metric 430 is increasing or decreasing. Using multiple windows, the preliminary analyzer 406 can identify trends in the metric 430 and identify a probability level (statistical significance 432) of an increasing or decreasing trend over a future window based on past windows.

[0090] In response to the preliminary analyzer 406 identifying a decrease in the metric 430 of a statistically significant amount (e.g., the probability of an increase or decrease is less than a specified probability), the secondary analyzer 408 can calculate and analyze other metrics, i.e., metrics 416-420. For example, the metrics 416-420 can include Renyi permutation entropy. The Renyi permutation can be identified by the secondary analyzer 408 for only those samples of EEG data for which the preliminary analyzer 406 detects a statistically significant decrease in eigenvalues. The permutation entropy can be computationally simplest and robust to noise and artifacts. The secondary analyzer 408 is configured in some embodiments to further identify the statistical significance of the metrics 416-420, i.e., identify the statistical significance 422-426 of the metrics 416-420.

[0091] More specifically, each of the metrics 416-420, Mi, calculated by the secondary analyzer 408 may be a time series of data. Based on the time series of the metrics 416-420, the secondary analyzer 408, in some embodiments, is configured to identify statistical significances 422-426, P(Mi), which indicate the probability of a shifting pattern of the metric's trajectory under the null hypothesis. Similarly, the statistical significances 422-426 may be a time series. The trend analyzer 414, in some embodiments, is configured to analyze the pattern of significant and non-significant values ​​of the metrics 416-420 based on the statistical significances 422-426 across time. The current set of significant metrics may be analyzed by the trend analyzer 414 as a group or panel of results. Each panel may be mapped to a particular category, for example, a clinical category such as candidate seizure event, no seizure, indeterminate condition, etc. The panels may also be mapped to other types of pseudo-events (non-seizures).

[0092] The metrics 416-420 can be numerous and varied, for example, there may be more than a dozen nonlinear metric types described with many variations of each of these metric types. For example, there are at least 14 different forms or calculation methods of entropy. The metrics 416-420 may include a complexity loss metric. Each entropy metric may have performance advantages and disadvantages in certain settings (e.g., sample entropy performs better than most in detecting voltage inhibition, Kolmogorov entropy is weaker than several forms of permutation entropy that also have low computational complexity, etc.). Fuzzy entropy is attractive in that class membership is categorized to allow the user better control over class boundaries. The frequency of the target event (seizure) may be included in the parameter values ​​for some entropy forms, for example Tsarris entropy. Renyi entropy may be a better choice when state changes are frequent or severe (e.g., paresthesia). Information regarding frequency of seizures, whether paresthesia is present, etc. can be included in the configuration data so that the secondary analyzer 408 can identify and analyze an appropriate mixture of non-linear metrics. An example of a methodology for calculating entropy can be found in Liang, Zhenhu et al., "EEG Entropy Measures in Anesthesia," Frontiers in Computational Neuroscience, vol. 9, 2015, doi:10.3389 / fncom.2015.00016, which is incorporated by reference herein in its entirety.

[0093] As described, the metrics 416-420 are based on, and thus may be derived from, the EEG signal. One important aspect of the metrics 416-420 may be the trajectory of each of the metrics 416-420 over time. The absolute values ​​of the metrics 416-420 may vary greatly as a function of the patient's age, condition, syndrome, concomitant medications, etc. Thus, the trend analyzer 414 is configured to analyze the trajectory of the metrics 416-420, and not necessarily the absolute values ​​of the metrics 416-420, to detect and / or classify candidate seizures. The direction of change over time (increase vs. decrease) of the metrics 416-420 caused by a candidate seizure may vary based on the patient's age and / or the type of candidate seizure. For this reason, the secondary analyzer 408 is configured to identify the trajectory of the metrics 416-420 such that the trend analyzer 414 may determine, based on the trajectory, whether any segment of the EEG signal is indicative of a candidate seizure and / or should be presented for visual evaluation by an electroencephalographer.

[0094] The metrics 416-420 themselves also vary in terms of stability and reliability depending on the sample size. The sample size can be increased by lengthening the sample period. However, extending the sample period runs the risk of missing a seizure event if the seizure event is shorter than the required sample period. In some embodiments, the secondary analyzer 408 is configured to identify the direction of change of the metrics 416-420 by using a moving window.

[0095] For example, at a sampling rate of 400 Hz, a 5 second window that the secondary analyzer 408 may be configured to utilize contains 2,000 samples. The step size and overlap of each window applied to the metrics 416-420 by the secondary analyzer 408 may be user defined and / or predefined through the user interface device 110. A typical value is a 1 second step size with 4 out of 5 samples overlapping between windows (i.e., 4 out of 5 samples are the same between two positions of the window as it moves).

[0096] Each window, when analyzed by the secondary analyzer 408, may indicate an increase or decrease in the value of one of the metrics 416-420, constituting the trajectory of the metric over time. The trend analyzer 414 may have statistical criteria for reviewing and / or analyzing segments defined by one of the window positions of one of the metrics 416-420. For example, assuming each sample is independent and behaves randomly, the probability of n consecutive changes in the same direction is 1 / 2 to the power of n. In some embodiments, the secondary analyzer 408 is configured to identify probabilities 422-426 for a pattern (increase or decrease) of the metrics 416-420. The probabilities may be the probability that a predefined amount of change will occur in one of the metrics 416-420 in a particular direction (e.g., a predefined amount of future windows will indicate an increase or decrease value of the metric based on the trajectory of past windows). The trend analyzer 414 may utilize thresholds that, when above or below the threshold, indicate that a particular one of the metrics 416-420 is increasing or decreasing at a statistically significant level. The trend analyzer 414 may utilize one or more user-defined and / or pre-defined thresholds to identify the statistically significant metrics 416-420 and / or to map the statistically significant metrics 416-420 to categories, e.g., seizures, noise, etc.

[0097] The choice of a particular metric calculation method performed by the secondary analyzer 408 may depend on the frequency of events, their spatial extent, sample size, dimensionality, the condition of the patient 102, the seizure syndrome of the patient 102, and the signal-to-noise ratio of the time epoch, all of which may be indicated by the configuration data or extracted from the EEG signal by the secondary analyzer 408 (e.g., signal-to-noise ratio). The calculation and mapping of the metric performed by the secondary analyzer 408 may take into account signal and subject factors, as well as the inherent computational complexity of determining which features should be prioritized. This same process applies to the calculation of dimensionality, complexity (or complexity loss), Lyapunov exponents, etc.

[0098] The metrics 416-420 and their statistical significance 422-426 may be passed to a trend analyzer 414 that may detect which trends in the statistically significant metrics are indicative of candidate seizures, noise, etc. For example, if the trend analyzer 414 detects an increase in the Renyi permutation entropy identified by the secondary analyzer 408 along with a decrease in the eigenvalues, the EEG data may indicate a noise or seizure burst suppression pattern, in which case additional metrics need to be analyzed. For example, one of the metrics 416-420 may be sample entropy that the secondary analyzer 408 identifies in phase space by the phase space analyzer 412. The sample entropy may be calculated by the secondary analyzer 408 after the calculation and analysis of the Renyi permutation entropy and / or in parallel with the eigenvalues ​​and / or the Renyi permutation entropy. The calculation of the sample entropy may be delayed by less than one second.

[0099] Sample entropy may be more sensitive to burst suppression than permutation entropy. The trend analyzer 414 may identify whether the sample entropy is positive or negative, and if the sample entropy is positive, classify the EEG data associated with the decreasing eigenvalue as noise. These results of the metrics 416-420 may be integrated by the trend analyzer 414 to classify the event. If both the Renyi permutation entropy and the eigenvalue decrease, the trend analyzer 414 may identify that the EEG data indicates a candidate seizure, and the secondary analyzer 408 may identify no sample entropy.

[0100] The phase space analyzer 412 is configured in some embodiments to perform a phase space analysis to identify metrics such as dimension. The phase space analyzer 412 is configured in some embodiments to generate a phase space plot of the EEG signal. A dynamical system can be represented by a set of differential equations for which a solution may not exist in a closed form. However, the phase space analyzer 412 can identify candidate seizure behavior by generating trajectories in phase space. At each instance of a time series of EEG signals, the phase space analyzer 412 is configured in some embodiments to generate a single point in phase space, and the sequence of these points forms a trajectory whose pattern provides insight into the nature of the driving function, i.e., the presence or absence of a seizure. The trajectories can occupy the entirety of the phase space or can converge to a low-dimensional region called an attractor. Noisy phase space trajectories never converge. If adjacent points start to get close to each other and then diverge, a strange attractor is said to exist, suggesting the presence of chaotic behavior.

[0101] The phase space analyzer 412 is configured in some embodiments to perform a time-shifted Tarkens method to generate a phase space plot based on empirical data of EEG time series. The Tarkens method is described in detail in Basar, Erol et al., "Strange Attractor EEG as Sign of Cognitive Function," Machinery of the Mind, 1990, pp 91-114, doi:10.1007 / 978-1-4757-1083-0_5. The EEG signal is a time series,

number

[0102] In some embodiments, from this time series, the phase space analyzer 412 is configured to identify a phase space representation of the EEG signal using the time delay td and the embedding dimension m.

number

[0103] The shape of the trajectory in phase space can be strongly influenced by the choice of time lag that the phase space analyzer 412 utilizes to generate the phase space plot. In some embodiments, the time lag is the first zero in the autocorrelation function, which is identified by the phase space analyzer 412 and then used to embed the signal in phase space. The phase space analyzer 412 is configured to utilize one or more different methods for estimating the time lag. In some embodiments, the phase space analyzer 412 can identify the time lag based on the nonlinear metric that the phase space analyzer 412 is attempting to identify. In some embodiments, the estimator that the phase space analyzer 412 uses to identify the time lag is linear and / or nonlinear.

[0104] The second value selected by the phase space analyzer 412 is the embedding dimension: if the dimension of the attractor is k, then Whitney's embedding theorem states that the embedding dimension must be 2k + 1. Thus, the phase space analyzer 412 is configured in some embodiments to select the embedding dimension based on the known or specified dimension of the attractor.

[0105] The phase space analyzer 412, in some embodiments, is configured to estimate the dimension of the phase space using a Kao method. The dimension that the phase space analyzer 412 estimates may be the dimension of an attractor in the phase space. The phase space analyzer 412, in some embodiments, is configured to start with a low dimension and continuously increase the dimension until the number of false neighbors reaches zero. The dimension that the phase space analyzer 412 reaches may be related to the presence or absence of a candidate seizure. For example, the trend analyzer 414 may identify whether or not there is a candidate seizure based on the metrics 416-420 and / or the dimension that the phase space analyzer 412 identifies.

[0106] In some cases, the dimension value may be as small as the dimension during seizures. Also, the dimension is typically 8 or less during interictal periods. From a practical standpoint, this ascent technique performed by the phase space analyzer 412, i.e., starting from low dimensions and increasing the dimension value, may result in real-time degradation of performance based on the computational load of the ascent technique. To overcome this computational load, the phase space analyzer 412 may receive configuration data indicative of the age of the patient. The phase space analyzer 412 is configured in some embodiments to select an initial dimension value based on the age of the patient to reduce the number of steps in which the phase space analyzer 412 increments the dimension value and identifies when the number of false neighbors reaches zero.

[0107] The initial dimensionality values ​​utilized by the phase space analyzer 412 in the ascent method may be low for young children and high for older children. This may be because younger ages have lower dimensionality, whether ictal or interictal. The selection of the initial dimensionality values ​​may be applied only to young children, for example, only if the configuration data indicates that the patient 102 is under the age of 10. There may not be a clear dimensionality difference between wakefulness and sleep in newborns, and age adjustment of the dimensionality may not be significant in older children and adults. The trend analyzer 414 may analyze the trend of the dimensionality, not necessarily the absolute value of the dimensionality. For example, if the dimensionality decreases over time, the trend analyzer 414 may classify the EEG signal as indicative of a candidate seizure. In this regard, the impact of small errors in the estimation of the absolute value of the dimensionality is partially reduced because the classification of the events is based on the change in the metric, not the absolute value.

[0108] The interface generator 410, in some embodiments, is configured to generate an interface for display on the user interface device 110 based on metrics identified by the preliminary analyzer 406 and / or the secondary analyzer 408. The interface generator 410, in some embodiments, is also configured to generate an interface based on the presence of candidate seizures identified by the trend analyzer 414. The user interface generated by the interface generator 410 can also be based on user input, such as a request to display a particular metric, a request to display historical EEG data, etc.

[0109] In some embodiments, the interface includes real-time EEG data trends. In some embodiments, the EEG data trends are always displayed. The interface generated by the interface generator 410 may also include a graph overlaying the trends of the eigenvalues ​​identified by the preliminary analyzer 406 on top of the EEG trends. There may be a 750 millisecond delay between the eigenvalues ​​and the EEG waveform. In some embodiments, every 750 milliseconds, the secondary analyzer 408 is configured to identify a new value for each of the metrics 416-420. Together, these values ​​form a trajectory for each of the metrics 416-420. Assuming that each value can only rise or fall relative to the previous value, the secondary analyzer 408 is configured in some embodiments to calculate the probability of n consecutive changes in the same direction. If the secondary analyzer 408 detects eight consecutive changes in the same direction, this may indicate a sufficient probability of a change in a particular direction. Since each metric is determined over a period of 750 milliseconds, and a total of eight values ​​can be determined to detect an increase or decrease, the secondary analyzer 408 can use a period of six seconds to determine the probability of an increase or decrease in metrics 416-420. The pre-analyzer 406 can be configured to perform the same process for metric 430.

[0110] When the change in eigenvalues ​​becomes significant, the interface generator 410, in some embodiments, is configured to change the color of the overlaid eigenvalue waveform. The interface generator 410, in some embodiments, is configured to store the time of the EEG trend related to the eigenvalue trend. This allows a user to request a particular portion of past EEG data through the user interface device 110. In response to the request, the interface generator 410 can cause the interface to display the requested portion of the EEG data and the eigenvalue trend corresponding to the requested portion. In some embodiments, any section of the EEG data that is classified as a candidate seizure and the corresponding metrics identified for this section of the EEG data are highlighted in the user interface generated by the interface generator 410. This allows a trained clinician to review a particular section of the EEG data that may be a candidate seizure and make a final determination as to whether this section of data is indicative of a seizure.

[0111] In some embodiments, the interface generator 410 includes in the interface a panel of statistically significant non-linear metrics identified by the secondary analyzer 408. The interface generator 410 can receive a user-specified significance level through the user interface device 110 and include a particular non-linear metric in the interface in response to the statistical significance level of the non-linear metric being greater than the user-specified significance level for that metric.

[0112] In some embodiments, when a metric transitions from non-significant to significant based on a threshold significance level and the statistical significance of each metric, the interface generator 410 changes the metric displayed in the interface from a first color to a second color, e.g., from black and white to yellow. If the pattern of the metric changes at a higher significance level (which may be defined based on user settings or predefined parameters), the interface generator 410 may make the metric a third color, e.g., blue. In some embodiments, the interface generator 410 displays a split screen of the EEG data such that the EEG data is shown in real time in a first window and periods of past EEG data that were classified as candidate seizures are also displayed.

[0113] If there are no significant changes in the EEG trajectory, there may be no significant EEG data to review, and the interface generator 410 may include an indication of the absence of a seizure in the interface. In some embodiments, the patterns and significance levels of the metrics may be user-defined. In some embodiments, the patterns and / or significance levels may be pre-defined.

[0114] 5, a process 500 for detecting candidate seizures by identifying changes in non-linear characteristics of an EEG signal is shown, according to an example embodiment. Seizure detector 108, in some embodiments, is configured to perform process 500. In particular, channel selector 402, filtering stage 404, pre-analyzer 406, and / or secondary analyzer 408 of seizure detector 108, in some embodiments, are configured to perform some and / or all of process 500. Additionally, any computing system or device described herein may be configured to perform process 500.

[0115] In step 502, the channel selector 402 receives EEG data from an array of EEG electrodes configured to sense brain activity of the patient 102. In some embodiments, the channel selector 402 receives the EEG data in real time, i.e., directly from the electrodes as the data is collected. In some embodiments, the channel selector 402 receives the data after the data has been collected, i.e., from a database or other memory device that stores the EEG data.

[0116] In step 504, the channel selector 402 may select an EEG signal from a particular channel or from multiple channels of the EEG data received in step 502. The filtering stage 404 may also perform filtering on the EEG signal. In some embodiments, the channel selection includes selecting an EEG signal from a particular electrode or group of electrodes from other EEG signals from other electrodes. In some embodiments, the selection made by the channel selector 402 is predefined, i.e., the same channel is always selected. In some embodiments, the channel selection is based on configuration data, i.e., data indicative of characteristics of the patient 102, such as age, height, medical syndrome, etc. The filtering may allow a particular range of frequencies to pass. In some embodiments, the range of frequencies to pass is predefined. In some embodiments, the range of frequencies to pass is also based on the configuration data.

[0117] In step 506, the preliminary analyzer 406 performs a preliminary analysis using a generalized metric suitable for detecting changes in the ratio of nonlinear factors to linear factors using a moving window. The preliminary analyzer 406 may detect trends in the metric. For example, the preliminary analyzer 406 may use a moving window to identify eigenvalues ​​and identify whether the trajectory of the eigenvalues ​​is increasing or decreasing. The preliminary analyzer 406 may calculate a probability value for the trajectory based on the value of the eigenvalues. If the probability value becomes small, i.e., less than a predefined amount, the statistical significance that the eigenvalue is increasing or decreasing may be identified by the preliminary analyzer 406 (e.g., the increase or decrease is statistically significant because the probability of the increase or decrease occurring is low).

[0118] A decrease in the eigenvalue may indicate a change in the ratio of nonlinear to linear factors, i.e., an increase in the nonlinear characteristics of the EEG signal. In response to detecting an increase in the nonlinear characteristics of the EEG signal in step 506, step 508 of process 500 may be performed. If the nonlinear characteristics of the EEG signal have not increased, step 508 may be skipped so that computational resources are not inefficiently utilized.

[0119] In step 508, the secondary analyzer 408 performs a second analysis, which includes identifying metrics that more accurately classify the form of nonlinear changes (e.g., changes in dimension, entropy, degree of separation of recursive loops in phase space, Lyapunov exponents, etc.). Because the metrics analyzed in the preliminary analysis (step 506) may be computationally efficient, while the metrics analyzed in the second stage (step 508) may require more computational resources, processing the metrics in separate stages allows for the use of computational resources only when necessary, i.e., after the preliminary analysis indicates a possible seizure. In step 508, the secondary analyzer 408 can identify trends in the second metric and, based on a particular pattern of changes in the second metric, identify whether the EEG data is indicative of a candidate seizure or not.

[0120] 6, a process 600 is shown for detecting candidate seizures by identifying changes in nonlinear features of an EEG signal using eigenvalues, Renyi permutation entropy, and sample entropy, according to an exemplary embodiment. The process 600 provides an exemplary metric analysis that the seizure detector 108 is configured to perform in some embodiments. The identification of the process 600 is exemplary, and there may be many combinations of metrics and / or analysis rules that may be utilized by the seizure detector 108 to detect candidate seizures. The seizure detector 108 is configured in some embodiments to analyze a variety of different patterns using a variety of different nonlinear metrics in addition to or in place of the eigenvalues, Renyi permutation entropy, and / or sample entropy. The seizure detector 108 is configured in some embodiments to perform the process 600. In particular, the preliminary analyzer 406 and / or the secondary analyzer 408 of the seizure detector 108 are configured in some embodiments to perform some and / or all of the process 600. Additionally, any computing system or device described herein may be configured to perform the process 600.

[0121] In step 602, the pre-analyzer 406 may receive an EEG signal. The EEG signal may be a signal generated based on electrical brain activity of the patient 102. The EEG signal may also be processed by a channel selector 402 and / or a filtering stage 404 before being received by the pre-analyzer 406. The EEG signal may be a time series of data samples.

[0122] In step 604, the preliminary analyzer 406 can identify eigenvalues ​​based on the EEG signal. In some embodiments, the preliminary analyzer 406 identifies eigenvalues ​​using a moving window of eigenvalues. For example, the preliminary analyzer 460 can utilize a window having a predefined length and a predefined overlap with a previous position of the window to generate eigenvalues ​​based on samples of the EEG signal that fall within the window.

[0123] In step 606, the preliminary analyzer 406 may analyze the trends of the eigenvalues ​​identified in step 604 to identify whether the eigenvalues ​​are increasing or decreasing over time. In some embodiments, the preliminary analyzer 406 identifies the probability of the trajectory, i.e., an overall increase or decrease in the eigenvalues. If the probability of the trajectory increasing is greater than a predefined amount, the preliminary analyzer 406 performs step 608. Similarly, if the probability of the trajectory decreasing is greater than a predefined amount, the preliminary analyzer 406 performs step 610. If the eigenvalues ​​do not show a significant increase or decrease, the preliminary analyzer 406 may classify the EEG data as not significant.

[0124] If the eigenvalue is increasing, the preliminary analyzer 406 may classify the EEG data (particular samples of the EEG signal) showing the increase as non-significant, step 608. However, if the preliminary analyzer 406 determines that the eigenvalue is decreasing, then the data in the EEG signal may be classified by the preliminary analyzer 406 as significant and likely indicative of a candidate seizure, step 610.

[0125] In step 612, the secondary analyzer 408 may determine a Renyi permutation entropy value based on the EEG signal. In some embodiments, the secondary analyzer 408 determines the Renyi permutation entropy value for only those segments of the EEG data that the pre-analyzer 406 classified as significant. This allows the secondary analyzer 408 to only perform calculations and efficiently utilizes computing resources.

[0126] In step 614, the secondary analyzer 408 may analyze the trend of the Renyi permutation entropy identified in step 612 to identify whether the Renyi permutation entropy is increasing or decreasing over time. In some embodiments, the secondary analyzer 408 identifies an overall increase or decrease in the probability of the trajectory, i.e., the Renyi permutation entropy. If the probability of the trajectory increasing is greater than a predefined amount, the secondary analyzer 408 performs step 618. Similarly, if the probability of the trajectory decreasing is greater than a predefined amount, the secondary analyzer 408 performs step 616. If the Renyi permutation entropy does not show a significant increase or decrease, the secondary analyzer 408 may perform step 618.

[0127] In step 618, the secondary analyzer 408 may determine a sample entropy value based on the EEG signal. In some embodiments, the secondary analyzer 408 determines a sample permutation entropy value for only those segments of the EEG data that the pre-analyzer 406 classified as significant. This allows the secondary analyzer 408 to only perform calculations and utilize computational resources as needed. Sample permutation entropy may be determined by the secondary analyzer 408 using a moving window in some embodiments.

[0128] In step 620, the secondary analyzer 408 determines whether the sample entropy is positive or negative. Based on the polarity of the sample entropy, the secondary analyzer 408 classifies the data as a candidate seizure in step 624 if the sample entropy is negative, or as noise in step 622 if the sample entropy is positive.

[0129] Configuration of an Exemplary Embodiment The structure and arrangement of the systems and methods shown in the various exemplary embodiments are merely illustrative. Although only a few embodiments are described in detail in this disclosure, many modifications are possible (e.g., changes in size, dimensions, structure, shape and proportions of various elements, parameter values, mounting methods, use of materials, colors, orientations, etc.). For example, the positions of elements can be reversed or otherwise changed, and the nature or number of separate elements or positions can be changed or varied. Accordingly, all such modifications are intended to be included within the scope of this disclosure. The order or sequence of any process or method steps can be changed or reordered according to alternative embodiments. Other substitutions, changes, and omissions can be made in the design, operating conditions, and arrangements of the exemplary embodiments without departing from the scope of this disclosure.

[0130] The present disclosure contemplates any method, system, and program product of a machine-readable medium for accomplishing various operations. The embodiments of the present disclosure can be implemented using existing computer processors, or by dedicated computer processors for suitable systems incorporated for this or another purpose, or by wired systems. Embodiments within the scope of the present disclosure include program products including machine-readable media or data structures stored on a machine-readable medium for holding or having machine-executable instructions. Such machine-readable media can be any available medium that can be accessed by a general purpose computer or a special purpose computer or other machine with a processor. By way of example, such machine-readable media can include RAM, ROM, EPROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to hold or store desired program code in the form of machine-executable instructions or data structures and that can be accessed by a general purpose computer or a special purpose computer or other machine with a processor. Combinations of the above are also included within the scope of machine-readable media. Machine-executable instructions include, for example, instructions and data that cause a general purpose computer, a special purpose computer, or a special purpose processing machine to perform a certain function or group of functions.

[0131] Although the figures show method steps in a particular order, the order of the steps may differ from that shown. Also, two or more steps may occur simultaneously or with partial concurrence. Such variations depend on the software and hardware systems selected, as well as designer choice. All such variations are within the scope of this disclosure. Similarly, the implementation of the software may be accomplished using standard programming techniques with rule-based logic and other logic to accomplish the various connection, processing, comparison, and decision steps.

Claims

1. 1. A seizure detection system comprising one or more circuits, The one or more circuits include: configured to receive electroencephalogram (EEG) signals generated based on the patient's electrical brain activity; configured to determine a plurality of metrics indicative of nonlinear characteristics of the EEG signal based on the EEG signal; configured to identify the EEG signal as indicative of a candidate seizure by identifying changes over time in a non-linear characteristic of the EEG signal based at least in part on the plurality of metrics; configured to generate a seizure alert indicating that the EEG signal is indicative of the candidate seizure; A seizure detection system, wherein the change in the non-linear characteristic is indicative of physiological forces causing the candidate seizure.

2. 10. The seizure detection system of claim 1, wherein the processing circuitry is configured to identify the EEG signal as indicative of the candidate seizure based on at least one of default parameter values ​​or user-defined parameter values.

3. 10. The seizure detection system of claim 1, wherein the seizure detection system is a cloud-based system, and the one or more circuits are configured to receive the EEG signals from a local EEG collection system over a network and provide resultant data to the local EEG collection system over the network.

4. the seizure detection system is a local system; The seizure detection system of claim 1 , wherein the local system is integrated with a local EEG system or locally connected to an EEG acquisition system.

5. 2. The seizure detection system of claim 1, wherein identifying a change in a non-linear characteristic of the EEG signal based at least in part on the plurality of metrics comprises identifying an increase in the non-linear characteristic over time.

6. 2. The seizure detection system of claim 1, wherein the plurality of metrics comprises at least one of dimensionality, synchrony, Lyapunov exponent, entropy, global nonlinearity, distance difference between recursive trajectories, or self-similarity.

7. The one or more circuits include: performing a preliminary analysis using one of the plurality of metrics to indicate that the EEG signal is indicative of the candidate seizure or that the EEG signal is noisy; performing a secondary analysis using one or more of the plurality of metrics to identify whether the EEG signal is indicative of the candidate seizure or includes noise; 2. The seizure detection system of claim 1, configured to identify that the EEG signal is indicative of the candidate seizure by identifying changes over time in non-linear characteristics of the EEG signal based at least in part on the plurality of metrics.

8. The one or more circuits include: configured to determine a probability of a trajectory of each of the plurality of metrics at a plurality of time points; configured to determine whether a trajectory of each of the plurality of metrics is significant based on the probability; 2. The seizure detection system of claim 1, wherein identifying changes in non-linear characteristics of the EEG signal based at least in part on the plurality of metrics comprises mapping significant metrics of the plurality of metrics to categories, the categories being seizure categories.

9. 2. The seizure detection system of claim 1, wherein the one or more circuits are configured to identify a dimensionality of the EEG signal by performing a phase space analysis by increasing a value of a dimension until a number of false neighbors reaches zero, and an initial value of the dimension is based on an age of the patient.

10. The one or more circuits include: configured to determine whether one or more of the plurality of metrics exhibits a statistically significant change over time; configured to generate a user interface including a real-time trend of the EEG signal and one or more of the plurality of metrics; The seizure detection system of claim 1 , configured to cause a user interface device to display the user interface.

11. The seizure detection system of claim 10 , wherein one of the plurality of metrics is an eigenvalue, and the user interface further comprises a trend of the eigenvalue.

12. The seizure detection system of claim 10 , wherein the user interface further comprises a history window of the EEG signal associated with the candidate seizure.

13. The one or more circuits include: configured to identify a moving window of eigenvalues ​​of the EEG signal; configured to identify that the eigenvalue is decreasing; 2. The seizure detection system of claim 1, configured to, in response to determining that the eigenvalue is decreased, determine that the EEG signal is indicative of the candidate seizure based at least in part on the plurality of metrics.

14. The one or more circuits include: configured to determine a Renyi permutation entropy value based on the EEG signal; configured to determine that the Renyi permutation entropy value is decreasing; 14. The seizure detection system of claim 13, configured to determine that the EEG signal is indicative of the candidate seizure in response to a determination that the eigenvalue is decreasing and a second determination that the Renyi permutation entropy value is decreasing.

15. The one or more circuits include: configured to determine a Renyi permutation entropy value based on the EEG signal; configured to determine that the Renyi permutation entropy value is increasing; configured to determine a sample entropy value based on the EEG signal in response to a first determination that the Renyi permutation entropy value is increasing; configured to determine that the EEG signal is indicative of the candidate seizure in response to a second determination that the sample entropy value is negative; 14. The seizure detection system of claim 13, configured to determine that the EEG signal is not indicative of the candidate seizure in response to a third determination that the sample entropy value is positive.

16. 1. A method for seizure detection, comprising: receiving, by a processing circuit, electroencephalogram (EEG) signals generated based on the patient's electrical brain activity; determining, by the processing circuitry, a plurality of metrics indicative of non-linear characteristics of the EEG signal based on the EEG signal; determining, by the processing circuitry, that the EEG signal is indicative of a candidate seizure by identifying a change over time in a non-linear characteristic of the EEG signal based at least in part on the plurality of metrics; generating a seizure alert indicating that the EEG signal is indicative of the candidate seizure; Including, A method, wherein a change in the non-linear characteristic is indicative of physiological forces causing the candidate seizure.

17. 17. The method of claim 16, wherein the processing circuitry's identification of the EEG signal as indicative of the candidate seizure is based on at least one of default parameter values ​​or user-defined parameter values.

18. 17. The method of claim 16, wherein the plurality of metrics comprises at least one of: dimensionality, synchrony, Lyapunov exponent, entropy, global nonlinearity, distance difference between recursive trajectories, or self-similarity.

19. determining that the EEG signal is indicative of the candidate seizure by identifying a change in a non-linear characteristic of the EEG signal over time based at least in part on the plurality of metrics; performing a preliminary analysis using one of the plurality of metrics to indicate that the EEG signal is indicative of the candidate seizure or that the EEG signal is noisy; performing a secondary analysis using one or more of the plurality of metrics to identify whether the EEG signal is indicative of the candidate seizure or contains noise; The method according to claim 16,

20. determining a probability of a trajectory of each of the plurality of metrics at a plurality of time points; determining whether a trajectory of each of the plurality of metrics is significant based on the probability; and Further comprising:

17. The method of claim 16, wherein the processing circuitry identifying changes in non-linear characteristics of the EEG signal based at least in part on the plurality of metrics comprises mapping significant metrics of the plurality of metrics to a category, the category being a seizure category.

21. 17. The method of claim 16, further comprising the processing circuitry identifying a dimensionality of the EEG signal by performing a phase space analysis by increasing a value of the dimensionality until a number of false neighbors reaches zero, and an initial value of the dimensionality is based on an age of the patient.

22. 1. A seizure detection system comprising: one or more electrodes coupled to the patient, the electrodes configured to generate electroencephalogram (EEG) signals based on the patient's electrical brain activity; A processing circuit, configured to receive electroencephalogram (EEG) signals generated based on the patient's electrical brain activity; configured to determine a plurality of metrics indicative of nonlinear characteristics of the EEG signal based on the EEG signal; configured to identify the EEG signal as indicative of a candidate seizure by identifying changes over time in a non-linear characteristic of the EEG signal based at least in part on the plurality of metrics; a processing circuit configured to generate a seizure alert indicating that the EEG signal is indicative of the candidate seizure; Including, A seizure detection system, wherein the change in the non-linear characteristic is indicative of physiological forces causing the candidate seizure.

23. 23. The seizure detection system of claim 22, wherein the processing circuitry is configured to identify the EEG signal as indicative of the candidate seizure based on at least one of default parameter values ​​or user-defined parameter values.

24. The processing circuitry includes: identifying a trajectory of each of the non-linear features by identifying a plurality of values ​​over time of each of the non-linear features; determining that a probability value of the respective trajectory of the non-linear feature is less than a probability level indicating that an occurrence of the respective trajectory of the non-linear feature is statistically significant; determining that the EEG signal is indicative of the candidate seizure in response to determining that a probability value of each trajectory of the non-linear features is less than the probability level; 23. The seizure detection system of claim 22, configured to identify the EEG signal as indicative of the candidate seizure by:

25. The processing circuitry includes: identifying a trajectory of one of the nonlinear features by identifying a plurality of values ​​of the nonlinear feature over time; determining that the probability value of the trajectory is less than a probability level indicating that the occurrence of the trajectory is statistically significant; determining that the EEG signal is indicative of the candidate seizure in response to determining that the probability value of the trajectory is less than the probability level; 23. The seizure detection system of claim 22, configured to identify the EEG signal as indicative of the candidate seizure by:

26. the value of each of the plurality of values ​​of the non-linear characteristic is decreased relative to a previous value of the plurality of values; 26. The seizure detection system of claim 25, wherein the processing circuitry is configured to determine a probability value for the trajectory based on the number of values ​​that each value decreases relative to a previous value of the plurality of values.

27. The processing circuitry includes: configured to receive user input through a user interface; 26. The seizure detection system of claim 25, configured to determine the probability level by setting the probability level to a value selected by the user via the user input.

28. The processing circuitry includes: configured to obtain a default value from a memory device of the seizure detection system; 26. The seizure detection system of claim 25, configured to determine the probability level by setting the probability level to the default value obtained from the memory device.

Images