Seizure cluster forecasting

Abstract

A method (100) for seizure clustering forecasting is provided. The method (100) comprises steps of: continually monitoring (S110) an EEG of a user; and detecting (S12) when and what type of cluster seizures occur; if the seizure detected was a start-cluster seizure (51) or an intra-cluster seizure (52), the method further comprises a step of notifying (S130) the user and / or a third party a forecast that additional seizures are more likely to occur within the next 24 hours; and if the seizure detected was a terminal cluster seizure (53) or an isolated seizure (60), the method further comprises a step of notifying (S130) the user and / or the third party a forecast that additional seizures are less likely to occur within the next 24 hours: A method (200) for seizure forecasting, an EEG sensor device (10), and an EEG sensor system (30) are further provided.

Description

[0001] SEIZURE CLUSTER FORECASTING

[0002] Technical Field

[0003] The present inventive concept relates to a method for forecasting seizures. In particular, the present inventive concept relates to a method for forecasting seizure clusters.

[0004] Background

[0005] The inherent unpredictability of seizures poses significant challenges for individuals with epilepsy, impacting their safety and well-being. Despite optimized treatments, many epilepsy patients still experience seizures, making the ability to forecast seizures crucial for managing activities and facilitating targeted therapies to enhance quality of life.

[0006] Detectable changes in brain activity often precede the onset of seizures in patients. However, existing techniques for forecasting seizures are limited by short historical data and / or user input, hindering their reliability. Despite recent progress, traditional seizure forecasting methods remain insufficient for assessing probabilistic forecasts. While there are challenges in accurately forecasting seizures, predicting seizure clusters could offer valuable insights and aid in more effective seizure management and treatment planning.

[0007] Summary

[0008] It is thereby an object of the present inventive concept to improve seizure forecasting, especially when the seizure is part of a seizure cluster, in different ways. This and other objects are achieved by the features set out in the appended independent claims, with embodiments set out in the dependent claims.

[0009] Accordingly, a first aspect of the inventive concept is provided by a method for seizure clustering forecasting. The method for seizure clustering forecasting comprises: continually monitoring an electroencephalography (EEG) of a user; and detecting when and what type of cluster seizures occur.

[0010] If the seizure detected was a start-cluster seizure or an intra-cluster seizure, the method further comprises a step of notifying the user and / or a third party a forecast that additional seizures are more likely to occur within the next 24 hours, within 12 hours, within 6 hours, within 2 hours, within 1 hour, or within 30 minutes. The timespan of the forecast may further be user-specific, such as being determined based on user history or set by the user.

[0011] If the seizure detected was a terminal cluster seizure or an isolated seizure, the method further comprises a step of notifying the user and / or the third party a forecast that additional seizures are less likely to occur within the next 24 hours, within 12 hours, within 6 hours, within 2 hours, within 1 hour, or within 30 minutes. The timespan of the forecast may further be user-specific, such as being determined based on user history or set by the user. A cluster of seizures / seizure cluster are two or more seizures in relatively close proximity in time, e.g. within minutes or hours such as within 24 hours, within 12 hours, within 6 hours, within 2 hours, within 1 hour, or within 30 minutes.

[0012] A clustered seizure represents a seizure whose probability of occurrence is influenced by the occurrence of a prior seizure (rather than being random).

[0013] Seizure clusters describe a set or series of seizures that are grouped consecutively, typically with short (or shorter than normal) inter-ictal periods (i.e. a time between seizures). However, the precise definition of seizure clusters regarding time interval and number of seizures has not been well established yet.

[0014] One definition of a seizure cluster is more than one seizure within 24 hours, another is more than one seizure within 12 or 6 hours. A third definition of a seizure cluster is more than three seizures within 24 hours.

[0015] A type of cluster seizure may be considered from a clustering point of view. Different types of seizures include start-cluster seizure, intra-cluster seizure, terminal cluster seizure, and isolated seizure. An isolated seizure is not part of a cluster of seizures. A cluster of seizures starts with a start-cluster seizure as the first seizure of the seizure cluster and ends with a terminal cluster seizure as the last seizure of the seizure cluster. The other seizures in a seizure cluster, between the start-cluster seizure and the terminal cluster seizure, are considered as intra-cluster seizures.

[0016] It is noted that an isolated seizure is not part of a seizure cluster. Therefore, if a new seizure is detected that is not preceded by a seizure that would form a cluster with the new seizure, such a new seizure is either a start-cluster seizure or an isolated seizure. Similarly, if a new seizure is detected that is preceded by one or more seizures forming a cluster with the new seizure, such a new seizure is either an intra-cluster seizure or a terminal cluster seizure.

[0017] The EEG may be sensed using an EEG sensor, such as an EEG sensor device comprising subcutaneous electrodes for (continuously) sensing EEG data. The EEG may alternatively be sensed by an EEG sensor comprising external electrodes or intracranial electrodes.

[0018] Detecting when a seizure occurs may comprise any number of methods, such as a real-time seizure detection algorithm or the user manually inputting that a seizure has occurred.

[0019] Assuming a seizure has been detected and when the seizure occurred is known, the method then detects what type of cluster seizure has occurred based on data available before (i.e. in a pre-ictal period), during (i.e. in an ictal period), and / or immediately after (i.e. in a post-ictal period) the seizure.

[0020] Notifying a user and / or a third party may comprise a smartphone notification, a text message, a tactile feedback, an audible sound, and / or an indicator light or display. The third party may be a caretaker, a family member or some other person.

[0021] For example, a text message may be sent to say that additional seizures are more likely to occur within the next 24 hours or that additional seizures are less likely to occur within the next 24 hours.

[0022] Alternatively, there may be a first indication showing that additional seizures are more likely to occur within the next 24 hours and a second indication showing that additional seizures are less likely to occur within the next 24 hours. In that way, either the first or second indication may always be active.

[0023] As an alternative, there may only be a single type of indication showing that additional seizures are more likely to occur within the next 24 hours. If additional seizures are no longer or less likely to occur within the next 24 hours, the indication may be inactivated. Inactivating such an indication may be considered a notification in itself.

[0024] Detecting when and what type of cluster seizures occur may comprise measuring a power spectral density (PSD), of the EEG during a pre-ictal period before each seizure and determining the type of cluster seizure by analyzing the measured PSD data and comparing the measured PSD data with a predetermined value. The PSD data is different during the pre-ictal period for the different types of seizures. The pre-ictal period may be 30 or 60 minutes before the seizure, or may be even longer such as 2 hours, 6 hours, 12 hours, or 24 hours before the seizure.

[0025] The measured PSD data may be analyzed using machine learning and / or a neural network.

[0026] The neural network may be trained using PSD data corresponding to different types of seizures as well as baseline data. The neural network may thereby be used to classify the type of cluster seizure based on the measured PSD data.

[0027] Machine learning may be used to extract features from PSD data to classify the seizure type.

[0028] Analyzing the PSD data may comprise calculating PSD from timeseries data using fast Fourier transforming (FFT) and / or wavelet algorithms, or any other method to calculate PSD; and filtering the PSD data to delta and theta frequency bands.

[0029] The delta frequency band is 0.5-4 Hz. The theta frequency band is 4-8 Hz. The inventor has realized that these frequency bands have the highest specificity and accuracy for determining what type of cluster seizure is about to occur.

[0030] Other frequency bands are alpha and beta, and any combination of one or more or two or more frequency band(s) may be used to determine what type of cluster seizure is about to occur.

[0031] Filtering may comprise separating the different frequency bands into different sources / channels and / or selecting the most relevant frequency band(s).

[0032] Detecting when and what type of cluster seizures occur may comprise measuring a duration of each seizure and comparing the measured duration with a predetermined value.

[0033] Start-cluster seizures and intra-cluster seizures have a shorter duration than terminal cluster seizures and isolated seizures. Accordingly, by determining an average duration of isolated seizures, this can be used as a predetermined value. If the measured duration is sufficiently shorter than this value, additional seizures are more likely to occur within the next few hours.

[0034] Detecting when and what type of cluster seizures occur may comprise logging each seizure and fitting a seizure cycle to the logged seizures.

[0035] It may be assumed that the seizures follow some sort of cycle that may be determined by fitting a seizure cycle to the logged seizures. This may further enable long-term forecasting beyond 24 hours, such as days or weeks in the future. Detecting when and what type of cluster seizures occur may comprise detecting inter-ictal epileptic discharges, lEDs, of the EEG during a pre-ictal period before each seizure and determining the type of cluster seizure of each seizure by counting the lEDs and comparing the number of lEDs with a predetermined value.

[0036] The IED data is different during the pre-ictal period for the different types of seizures. The pre-ictal period may be 30 or 60 minutes before the seizure, or may be even longer such as 2 hours, 6 hours, 12 hours, or 24 hours before the seizure. The inventor has found that isolated seizures have fewer lEDs than start-cluster seizures and intra-cluster seizures for some subjects, a count above a predetermined value may thereby indicate that additional seizures are more likely to occur within the next few hours.

[0037] Detecting when and what type of cluster seizure occur may comprise detecting and monitoring sleep stages of the EEG during a pre-ictal period before each seizure and determining the type of cluster seizure of each seizure by considering the pre-ictal sleep stage and / or comparing the sleep stage with a predetermined and / or subject-specific stage condition.

[0038] By using sleep stages, a more accurate detection may be made.

[0039] Detecting when and what type of cluster seizures occur may comprise measuring a heart rate of the user and comparing a heart rate variation between 1 and 4 hours before each seizure with a predetermined (and potentially patientspecific) value.

[0040] The heart rate variation may be determined by calculating a standard deviation value of the measured heart rate.

[0041] The heart rate may be measured using oximetry, be manually input, or using any other means for measuring heart rate.

[0042] The inventor has realized that start-cluster seizures and intra-cluster seizures have a higher heart rate variation between 1 to 4 hours or 2 to 4 hours before the start cluster and intra-cluster seizures than terminal cluster seizures and isolated seizures.

[0043] It is noted that continuous measurement is not necessary to measure a heart rate variation with high enough fidelity to determine a significant difference between the different types of seizures. A measurement every few minutes, such as every 5 or every 10 or every 15 minutes may be sufficient. Detecting when and what type of cluster seizures occur may comprise calculating a likelihood and / or confidence of the detection; and notifying the user and / or the third party may comprise displaying the likelihood and / or confidence of the forecast based on the likelihood and / or confidence of the detection.

[0044] By showing the user and / or the third party a likelihood and / or confidence of the detection, they are provided with more information to base decisions on.

[0045] A second aspect of the inventive concept is provided by an EEG sensor device comprising: subcutaneous electrodes configured to continuously measure an EEG; and a processing circuit configured to execute the method according to the first aspect.

[0046] This provides a single device performing the method of the first aspect.

[0047] Subcutaneous electrodes are relevant for continuously measuring EEG as they enable the user to continue living their normal life while measuring without disturbing the electrodes.

[0048] A third aspect of the inventive concept is provided by an EEG sensor system comprising: an EEG sensor device comprising subcutaneous electrodes configured to continuously measure an EEG; and a processing unit in communication with the EEG sensor device and configured to execute the method according to the first aspect.

[0049] This provides a simple system performing the method of the first aspect that is compatible with many EEG sensor devices.

[0050] Subcutaneous electrodes are relevant for continuously measuring EEG as they enable the user to continue living their normal life while measuring without disturbing the electrodes.

[0051] The processing unit may e.g. be a smartphone, a computer, a cloud-based processing service, and / or a dedicated device. The connection between the processing unit and the EEG sensor device may e.g. be Bluetooth®, Wi-Fi, or other suitable fast, low energy alternatives.

[0052] A fourth aspect of the inventive concept is provided by a method for seizure forecasting. The method comprises steps of: continually monitoring an EEG of a user; detecting a seizure; calculating a power spectral density, PSD, in delta and theta frequency bands of the EEG during a pre-ictal period before the seizure; analyzing the PSD data in the delta and theta frequency bands of the EEG to detect indication of an upcoming seizure; and notifying the user and / or a third party that a seizure is likely to occur (within a set time, such as 30 minutes, 1 hour, 2 hours, 6 hours, 12 hours, or 24 hours).

[0053] Improvements and alternatives to the first aspect of the inventive concept apply mutatis mutandis to the fourth aspect of the inventive concept.

[0054] Detecting a seizure may make use of the continually monitored EEG, and / or by user input.

[0055] The inventor has realized that forecasting any seizure is simpler than forecasting a type of cluster seizure and methods for forecasting seizure clusters may be used in a similar way to forecast any seizure with increased precision.

[0056] Analyzing the PSD data may comprise comparing the PSD data to corresponding data from a pre-ictal period before a seizure.

[0057] The PSD data during the pre-ictal period is indicative of seizures. The pre-ictal period may be 30 or 60 minutes before the seizure, or may be even longer such as 2 hours, 6 hours, 12 hours, or 24 hours before the seizure.

[0058] The corresponding data may be from a database, previously measured data of the user, and / or online sensing.

[0059] Analyzing the PSD data may comprise filtering the PSD data to delta and theta frequency bands.

[0060] The delta frequency band is 1 -4Hz. The theta frequency band is 4-8 Hz. The inventor has realized that these frequency bands have the highest specificity and accuracy for determining that a seizure is about to occur.

[0061] Other frequency bands are alpha and beta, and any combination of one or more or two or more frequency band(s) may be used to determine that a seizure is about to occur.

[0062] The PSD data may be analyzed using a neural network.

[0063] The neural network may be trained using PSD data with and without seizures.

[0064] The neural network may thereby be used to classify between a seizure and not based on the measured PSD data.

[0065] Machine learning may be used to extract features from PSD data to classify the seizure.

[0066] Analyzing the filtered PSD data to detect indication of an upcoming seizure may comprise calculating a likelihood and / or confidence of the detection; and notifying the user and / or the third party may comprise displaying the likelihood and / or confidence of the forecast based on the likelihood and / or confidence of the detection. By showing the user and / or the third party a likelihood and / or confidence of the detection, they are provided with more information to base decisions on.

[0067] Description of drawings

[0068] The invention will be described in further detail with reference to preferred aspects and the accompanying drawing, in which:

[0069] Figs. 1a-b show different seizure types according to an embodiment;

[0070] Figs. 2a-b show changes in the distribution of power spectral density (PSD) data in different frequency bands of the EEG during a pre-ictal period before each seizure according to an embodiment;

[0071] Fig. 3 shows a seizure cycle according to an embodiment;

[0072] Fig. 4 shows inter-ictal epileptic discharges (lEDs) of the EEG during a pre- ictal period before each seizure according to an embodiment;

[0073] Fig. 5 shows a heart rate variation according to an embodiment;

[0074] Fig. 6 shows a schematic view of an EEG sensor system according to an embodiment;

[0075] Fig. 7 shows a flowchart of a method for seizure clustering forecasting according to an embodiment; and

[0076] Fig. 8 shows a flowchart of a method for seizure forecasting according to an embodiment.

[0077] Description of embodiments

[0078] In the following, terms such as a / an / the and comprising are intended to be interpreted as non-limiting. Any processor or computing unit may be implemented as an electronic circuit and electronic connections may be wired or wireless unless otherwise explicitly stated. Unless explicitly specified, a wireless connection may be implemented in any number of standards known to a person skilled in the art, such as Wi-Fi, Bluetooth®, Zigbee, 4G / LTE, 5G and so on.

[0079] Fig. 1 a shows a schematic view of a timeline of seizures of different types. There is no set scale of the timeline, and the timeline is not intended to be in scale.

[0080] Fig. 1a shows a few seizure clusters 50. A seizure cluster 50 is a group of seizures close together on the timeline. The timeline also shows isolates seizures 60, which are not part of a seizure cluster. A person experiencing seizures, such as a person with epilepsy, may experience a mix of isolated seizures and seizure clusters.

[0081] In a seizure cluster 50, different types of seizures may be identified. The first seizure of a seizure cluster 50 is a start-cluster seizure 51 , the last seizure of a seizure cluster 50 is a terminal cluster seizure 53, and all other (intermediary) seizures in the cluster 50 are intra-cluster seizure(s) 52.

[0082] Some seizure clusters 50 do not have intra-cluster seizures 52, such as the far-right seizure cluster 50 in Fig. 1 a.

[0083] Fig. 1 a further shows time spans, such as pre-ictal periods before a seizure or seizure cluster. Another time span shown in Fig. 1 a is a baseline, which is a period of time removed from any seizures. Measurements made during a baseline period may be used to normalize measurements.

[0084] Other time spans not shown in Fig. 1 a is an inter-ictal period, which lasts between the isolated seizure(s) 60 and / or seizures within the seizure cluster(s) 50, and a post-ictal period, which is after the isolated seizure(s) 60 or and / or seizures within the seizure cluster(s) 50.

[0085] Fig. 1 b shows another schematic view of a timeline of seizures of different types. Seizure clusters 50 are marked and the different types of seizures (startcluster seizure 51 , intra-cluster seizure 52, terminal cluster seizure 53, and isolated seizures 60) are shown with different patterns.

[0086] Figs. 2a-b show changes in the distribution of power spectral density (PSD) data in different frequency bands of the EEG during a pre-ictal period before each seizure. Fig. 2a shows box plots and Fig. 2b shows violin plots arranged in a similar manner. Figs. 2a and 2b use different but corresponding processing methods.

[0087] Figs. 2a-b shows that the different seizure types have different PSD distributions during the pre-ictal period. This difference is especially clear in the delta and theta frequency bands of Fig. 2b.

[0088] Using this, the type of cluster seizure may be determined by analyzing the measured PSD data and comparing the measured PSD data with a predetermined value.

[0089] The different frequency bands may be treated differently, such as one band being ignored, another being compared to a band-specific predetermined value, another band being fed to a neural network / machine learning unit to train, and another band being used to receive a trained analysis from the neural network / machine learning unit. Any combination of these approaches may be used in solitaire or combination.

[0090] Fig. 3 shows a fitting of a seizure cycle. The crosses and dots represent seizures over time and a cycle has been fitted to the seizures so that the seizures are concentrated (as much as possible) along the top of the cycle.

[0091] Such a cycle may be used to determine a periodicity of a patient’s seizures, which may be used to predict seizures.

[0092] Further, the crosses represent clustered seizures and Fig. 3 shows that these cohere more closely to the top of the curves. This means that clustered seizures may be more accurately precited using a seizure cycle fitting, hence a seizure cycle may be used to further improve seizure type determination as well as prediction of both isolated seizures and clustered seizures.

[0093] Fig. 4 shows a plot comparing the number of pre-ictal inter-ictal epileptic discharges (lEDs) during clustered and isolated seizures for different patients. There is a clear trend that clustered seizures have a higher number of pre-ictal lEDs.

[0094] As such, the type of cluster seizure of each seizure may be determined by counting the pre-ictal lEDs and comparing the number with a predetermined value.

[0095] Fig. 5 shows a plot of a p-value of heart rate variation against the time before a seizure cluster compared to an isolated seizure. The heart rate variation has been calculated using a standard deviation value of the measured heart rate. The plot in Fig. 5 uses the hypothesis that clustered seizures have a higher standard deviation value than isolated seizures.

[0096] The p-value is below 0.052-5 hours before seizure clusters, which means that in this time-span the heart rate variation is statistically significantly higher for seizure clusters than for isolated seizures. Thereby, the type of cluster seizure may be determined by measuring the heart-rate variation hours before each seizure.

[0097] It is noted that the higher values in time have naturally less data since these values are rarer.

[0098] Fig. 6 shows a schematic view of an EEG sensor system 30. The EEG sensor system 30 comprises an EEG sensor device 10 in communication with a processing unit 20 or comprising a processing circuit 12.

[0099] The processing unit 20 and processing circuit 12 are dashed in Fig. 6 to indicate that they are optional, only one is needed to execute the method in Fig. 7. The EEG sensor device 10 further comprises subcutaneous electrodes 11 to continuously measure an EEG.

[0100] Fig. 7 shows a flowchart of a method 100 for seizure clustering forecasting. The steps in dashed lined boxes are optional, and the steps may be performed in a different order or simultaneously.

[0101] The first step of Fig. 7 is continually S110 monitoring an EEG of a user. This is performed by an EEG sensor device using electrodes to measure EEG, such as the EEG sensor device in Fig. 6.

[0102] Next in Fig. 7 is a step of detecting S120 when and what type of cluster seizures occur. The types are start-cluster seizure 51 , intra-cluster seizure 52, terminal cluster seizure 53, and isolated seizures 60. Detecting when seizures occur may comprise logging seizures after they have happened based on EEG measurements and / or user input.

[0103] Steps S121-S126 are different ways to detect S120 different types of seizures. They can be used in isolation or combination in any order or simultaneously. They are performed by a processing circuit, either comprised in the EED sensor device or another device.

[0104] Next in Fig. 7 is a step of measuring S121 a power spectral density (PSD) of the EEG during a pre-ictal period before each seizure and determining the type of cluster seizure by analyzing the measured PSD data and comparing the measured PSD data with a predetermined value. Isolated seizures may in general have a significantly different, such as a higher, PSD than clustered seizures.

[0105] The analysis may be made using machine learning and / or a neural network.

[0106] Analyzing the PSD data comprises filtering the PSD data to delta and / or theta frequency bands.

[0107] Next in Fig. 7 is a step of measuring S122 a duration of each seizure and comparing the measured duration with a predetermined value. Isolated seizures may in general have a longer duration than clustered seizures.

[0108] Next in Fig. 7 is a step of logging S123 each seizure and fitting a seizure cycle to the logged seizures. Clustered seizures may in general fit better to seizure cycles.

[0109] Next in Fig. 7 is a step of detecting S124 inter-ictal epileptic discharges (lEDs) of the EEG during a pre-ictal period before each seizure and determining the type of cluster seizure of each seizure by counting the lEDs and comparing the number of lEDs with a predetermined value. Clustered seizures may in general have a higher number of pre-ictal lEDs than isolated seizures.

[0110] Next in Fig. 7 is a step of measuring S125 a heart rate of the user and comparing a heart rate variation between 1 to 4 hours before each seizure with a predetermined value. Clustered seizures may in general have a higher heart rate variation than isolated seizures in this timespan.

[0111] Next in Fig. 7 is a step of monitoring S126 sleep stages of the EEG during a pre-ictal period before each seizure and determining the type of cluster seizure of each seizure by considering the pre-ictal sleep stage and / or comparing the sleep stage with a predetermined and / or subject-specific stage condition.

[0112] Next in Fig. 7 is a step of calculating S127 a likelihood and / or confidence of the detection of steps S120-S126. This calculation may be done in any number of ways known to the skilled person and may depend on which step(s) and / or how many steps of S121-S126 were used to detect what type of cluster seizures occur, as well as the individual results of each step, such as a difference to the predetermined value.

[0113] The final step of Fig. 7 is a step of notifying S130 the user and / or a third party a forecast that additional seizures are more or less likely to occur. If the seizure was detected (based on steps S120-S126) as a start-cluster seizure or an intra-cluster seizure, the forecast is that additional seizures are more likely to occur within the next 24 hours. If the seizure was detected (based on steps S120-S126) as a terminal cluster seizure or an isolated seizure, the forecast is that additional seizures are less likely to occur within the next 24 hours.

[0114] This notification may be done using any means of communication known to the skilled person, such as a phone call or pop-up notification.

[0115] Such a notification may include the likelihood and / or confidence of the forecast if step S127 has been performed.

[0116] Fig. 8 shows a flowchart of a method 100 for seizure forecasting. The steps in dashed lined boxes are optional, and the steps may be performed in a different order or simultaneously.

[0117] Fig. 8 is similar to Fig. 7 and may use similar steps and devices.

[0118] The first step of Fig. 8 is continually monitoring S210 an EEG of a user. This is performed by an EEG sensor device using electrodes to measure EEG, such as the EEG sensor device in Fig. 6. Next in Fig. 8 is a step of detecting S220 a seizure. Detecting when seizures occur may comprise logging seizures after they have happened based on EEG measurements and / or user input.

[0119] Next in Fig. 8 is a step of measuring S221 a power spectral density (PSD) in delta and theta frequency bands of the EEG during a pre-ictal period before the seizure.

[0120] Measuring S221 the PSD data comprises filtering the PSD data to delta and / or theta frequency bands.

[0121] Next in Fig. 8 is a step of analyzing S222 the PSD data in the delta and / or theta frequency bands to detect indication of an upcoming seizure. A significantly different PSD may indicate an upcoming seizure, such as a higher PSD.

[0122] The analysis may be made using machine learning and / or a neural network.

[0123] Next in Fig. 8 is an optional step of calculating S227 a likelihood and / or confidence of the detection of an indication of an upcoming seizure in step S222. This calculation may be done in any number of ways known to the skilled person and may depend on the contents of the PDS data in the delta and / or theta frequency bands.

[0124] The final step of Fig. 8 is a step of notifying S230 the user and / or a third party a forecast that a seizure is (more or less) likely to occur. This forecast is based on the indication of an upcoming seizure in step S222.

[0125] This may be done using any means of communication known to the skilled person, such as a phone call or pop-up notification.

[0126] Such a notification may include the likelihood and / or confidence of the forecast if step S227 has been performed.

[0127] The preceding description has been a non-exhaustive disclosure of different exemplifying embodiments of the present inventive concept. This is not to be misconstrued as limiting the scope of the protection sought for the inventive concept, which is defined by the appended claims.

Claims

CLAIMS1 . A method (100) for seizure clustering forecasting, the method (100) comprising steps of: continually monitoring (S110) an EEG of a user; and detecting (S120) when and what type of cluster seizures occur; if the seizure detected was a start-cluster seizure (51 ) or an intra-cluster seizure (52), the method further comprises a step of notifying (S130) the user and / or a third party a forecast that additional seizures are more likely to occur within the next 24 hours; and if the seizure detected was a terminal cluster seizure (53) or an isolated seizure (60), the method further comprises a step of notifying (S130) the user and / or the third party a forecast that additional seizures are less likely to occur within the next 24 hours.

2. The method according to claim 1 , wherein detecting (S120) when and what type of cluster seizures occur comprises measuring (S121 ) a power spectral density, PSD, of the EEG during a pre-ictal period before each seizure and determining the type of cluster seizure by analyzing the measured PSD data and comparing the measured PSD data with a predetermined value.

3. The method according to claim 2, wherein the measured PSD data are analyzed using machine learning and / or a neural network.

4. The method according to claim 2 or 3, wherein analyzing the PSD data comprises filtering the PSD data to delta and / or theta frequency bands.

5. The method according to any one of the preceding claims, wherein detecting (S120) when and what type of cluster seizures occur comprises measuring (S122) a duration of each seizure and comparing the measured duration with a predetermined value.

6. The method according to any one of the preceding claims, wherein detecting (S120) when and what type of cluster seizures occur comprises logging (S123) each seizure and fitting a seizure cycle to the logged seizures.

7. The method according to any one of the preceding claims, wherein detecting (S120) when and what type of cluster seizures occur comprises detecting (S124) inter-ictal epileptic discharges, lEDs, of the EEG during a pre-ictal period before each seizure and determining the type of cluster seizure of each seizure by counting the lEDs and comparing the number of lEDs with a predetermined value.

8. The method according to any one of the preceding claims, wherein detecting (S120) when and what type of cluster seizures occur comprises measuring (S125) a heart rate of the user and comparing a heart rate variation between 1 to 4 hours before each seizure with a predetermined value.

9. The method according to any one of the preceding claims, wherein detecting (S120) when and what type of cluster seizure occur comprises detecting and monitoring (S126) sleep stages of the EEG during a pre-ictal period before each seizure and determining the type of cluster seizure of each seizure by considering the pre-ictal sleep stage and / or comparing the sleep stage with a predetermined and / or subject-specific stage condition.

10. The method according to any one of the preceding claims, wherein detecting (S120) when and what type of cluster seizures occur comprises calculating (S127) a likelihood and / or confidence of the detection; and notifying (S130) the user and / or the third party comprises displaying the likelihood and / or confidence of the forecast based on the likelihood and / or confidence of the detection.11 . A method (200) for seizure forecasting, the method (200) comprising steps of: continually (S210) monitoring an EEG of a user; detecting (S220) a seizure: measuring (S221 ) a power spectral density, PSD, in delta and theta frequency bands of the EEG during a pre-ictal period before the seizure;analyzing (S222) the PSD data in the delta and / or theta frequency bands to detect indication of an upcoming seizure; and notifying (S230) the user and / or a third party a forecast that a seizure is likely to occur.

12. The method according to claim 11 , wherein analyzing (S220) the PSD data comprises comparing the PSD data to corresponding data from a pre-ictal period before a seizure.

13. The method according to claim 11 or 12, wherein the PSD data are analyzed (S222) using machine learning and / or a neural network.

14. The method according to any one of the claims 11-13, wherein analyzing (S222) the filtered PSD data to detect indication of an upcoming seizure comprises calculating (S227) a likelihood and / or confidence of the detection; and notifying (S230) the user and / or the third party comprises displaying the likelihood and / or confidence of the forecast based on the likelihood and / or confidence of the detection.

15. An EEG sensor device (10) comprising: subcutaneous electrodes (11 ) configured to continuously measure an EEG; and a processing circuit (12) configured to execute the method (100) according to any one of claims 1 -14.

16. An EEG sensor system (30) comprising: an EEG sensor device (10) comprising subcutaneous electrodes (11 ) configured to continuously measure an EEG; and a processing unit (20) in communication with the EEG sensor device (10) and configured to execute the method (100) according to any one of claims 1-14.

Images