METHOD AND APPARATUS FOR PREDICTION OF EPILEPTIC SEIZURES

Abstract

A system for predicting epileptic seizures includes sensors operable to record a wearer's brain activity. The sensors electronically communicate with a processor configured to receive and store output EEG oscillations and activities. A threshold electrical fluctuation level is identified as the level electrical activity experienced at the onset of a seizure event, and is then stored in the PDA memory as a predetermined threshold value. The processor analyzes the input EEG data logged for a recording period, and the logged data is broken into a number of data values across a series of individual set sampling periods. Convert collected data value readings for individual sampling periods as a non-linear measure value using fractal dimension, P&H and/or Lyapunov weighing. The calculated values for a predicted next time intervals extending the sampling period is projected forward and compared against the predetermined threshold value to indicate a likely seizure event.

Claims

1. A monitoring system for providing a user with advance warning of a likely seizure event, the system comprising: a signaling mechanism operable to provide at least one of an audible, visual or sensory warning signal to said user indicative of a predicted seizure event; a sensor assembly having at least one sensor operable to sense and output sensed signals representative of the user's electroencephalographic (EEG) wave forms over a baseline monitoring period, a computing device having a processor and memory, the computing device electronically communicating with said sensor assembly for receiving the output sensed signals at selected time intervals over said baseline monitoring period, storing in said memory sensed data values representative of said output sensed signals at spaced time intervals (x.sub.1, x.sub.2 . . . x.sub.n) over said baseline monitoring period, the processor including program instructions operable to perform the process steps of: A. select an initial data series sequence comprising associated one of said sensed data values over a selected timeline period comprising part of said baseline monitoring period, wherein said initial data series sequence is represented as
S.sub.L={x.sub.1,x.sub.2 . . . x.sub.L} B. compute a first non-linear measure value V(S.sub.L)=.sub.L of said initial data series sequence using at least one of fractal dimension, Lyapunov exponent and P&H; C. for each subsequent remaining sensed data value over the baseline monitoring period S.sub.L+1={x.sub.1, x.sub.2 . . . x.sub.L+1} compute an associated non-linear measure values using at least one of fractal dimension, Lyapunov exponent and P&H to provide a transformed data series,
(S.sup.m.sub.N)=(y.sub.L,y.sub.L+1,y.sub.L+2 . . . y.sub.N); D. for the transformed data series sequence S.sup.m(.sub.L,N) calculate a reference non-linear value V(S.sup.m.sub.N) of the transformed data series sequence (S.sup.m.sub.N) using at least one of fractal dimension, Lyapunov exponent and P&H; E. determine a normal distribution curve of the Y values between sequential data points (y.sub.L, y.sub.L+1, y.sub.L+2 . . . y.sub.N) in the transformed value data series sequence S.sup.m.sub.N; F. with said normal distribution curve centered on the last data value y.sub.N in said transformed data series (S.sup.m.sub.N), generate a plurality of random next data values (y.sup.j.sub.N+1) for a predicted next time interval (x.sub.n+1), as separate generated extended time series sequences (S.sup.j.sub.N+1); G. for each said generated extended time series sequence S.sup.j.sub.N+1, compute an associated non-linear measure value V(S.sup.j.sub.N+1) using at least one of fractal dimension, Lyapunov exponent or P&H; H. select the generated extended time series sequence having the associated non-linear measure value (V(S.sup.j.sub.N+1)) closest to the stored reference non-linear measure value V(S.sup.m.sub.N) as a next time series data sequence, and assigning the random data value for the selected extended time series sequence as a predicted data value (y.sub.N+1) for the predicted next time interval (x.sub.N+1); I. compare the predicted data value of the predicted next time interval with a predetermined threshold indicative of a likelihood of said seizure event, and wherein when at least one predicted data value exceeds a predetermined threshold, the computing device activating said signaling mechanism to output said warning signal to the user.

2. The monitoring system as claimed in claim 1, wherein the predicted data value is selected as a new last data value for an extended transformed data series, the processor further including program instructions to: repeat steps F through I to generate successive predicted data values for next time intervals of at least about ⅓ of the time of the baseline monitoring period, and preferably at least about sixteen minutes.

3. The monitoring system as claimed in claim 1, wherein the selected timeline period is selected at from about a first one quarter to about one half of the baseline monitoring period, and preferably at about a first one third of the baseline monitoring period.

4. The monitoring system as claimed in claim 2, wherein the baseline monitoring period is selected at between about 45 and 120 minutes and the selected timeline period is selected at between about 10 and 30 minutes.

5. The monitoring system as claimed in claim 1, wherein the time intervals comprise equally spaced time intervals over the baseline monitoring period selected at between about 90 and 240.

6. The monitoring system as claimed in claim 1, wherein the plurality of random next data values is selected at between about 5 and 50.

7. The monitoring system as claimed in claim 6, wherein the system further includes a random number generator for generating the random data values.

8. The system as claimed in claim 1, wherein the threshold is selected as a standard deviation greater than about two to about 15.

9. The system as claimed in claim 8, wherein the spaced time intervals comprise equally spaced intervals selected at between about 1 and 120 seconds, and the selected timeline period is selected at between about 15 and 30 minutes.

10. (canceled)

11. The system as claimed in claim 1, wherein said sensor assembly includes a plurality of said sensors, the output sensed signals comprising continuous electronic readings sampled over a plurality of constant time intervals, and the baseline period is selected as a time period consisting of at least two user experienced pre-seizure, seizure and post-seizure events.

12. The system as claimed in claim 1, wherein the computing device comprises a personal digital assistant, and said signaling mechanism comprises at least one of a visual display and an audio output, and wherein the warning signal comprises at least one of an audible signal to said user emitted by said audio output and a visual to said user signal visible on said visual display.

13. The system as claimed in claim 1, wherein the seizure event comprises a Tonic-clonic seizure.

14. A seizure monitoring system having a signaling mechanism for providing a user with advance warning of a predicted epileptic seizure event, the system comprising: a sensor assembly having a sensor operable to sense and output sensed signals representative of the user's electroencephalographic (EEG) activity over a monitored baseline period of time as sensed data, a computing device having a processor and memory, the computing device electronically communicating with said sensor assembly for receiving said sensed data, and operable to store in said memory a baseline time series comprising sensed data values (x.sub.1, x.sub.2, x.sub.3 . . . x.sub.N) representative of said sensed data at selected time intervals over the baseline period of time, the processor including program instructions operable to: A. compile an initial time series data sequence S.sub.L=(x.sub.1, x.sub.2, x.sub.3 . . . x.sub.L) comprising data values over an initial recording portion of said baseline period of time; B. compute an initial non-linear measure value V(S.sub.L)=y.sub.L of the initial time series data sequence using fractal dimension, Lyapunov exponent and/or P&H, C. compute successively, a non-linear measure value V(S.sub.L+1) . . . V(S.sub.N) for a time series data sequences comprising each subsequent data value in the baseline time series using fractal dimension, Lyapunov exponent and/or P&H as successive non-linear measure values [V(S.sub.L+1)=y.sub.L+1] . . . [V(S.sub.N)=y.sub.N]; D. store the non-linear measure values as a transformed value data series S.sup.m.sub.N, S.sup.m.sub.N=(y.sub.1, y.sub.2 . . . y.sub.N) and determining a non-linear measure value for the transformed value V(S.sup.m.sub.N) data series S.sup.m.sub.N as a reference value; E. determine a normal distribution curve of the non-linear measure values in the transformed value data series S.sup.m.sub.N; and F. with the normal distribution curve centered on a last transformed value y.sub.N thereof, generate from 5 to 50, and preferably about 10 random data signal values (y.sup.1, y.sup.2 . . . y.sup.N) for a predicted next time interval (x.sub.N+1), as part of an associated generated extended time series sequence S.sup.j.sub.N+1; G. for each said generated extended time series sequence (S.sup.j.sub.N+1), compute an associated non-linear measure value (V(S.sup.j.sub.N+1)), using at least one of fractal dimension, Lyapunov exponent and P&H; and H. select the generated extended time series sequence having the associated non-linear measure value (V(S.sup.j.sub.N+1)) which is closest to the reference value V(S.sup.m.sub.N) as a new time series data sequence; wherein the random data signal value of the selected generated extended time series is selected as the predicted next data value y.sub.N+1 for the next projected time interval x.sub.N+1, and I. when at least one predicted next data value exceeds a preselected threshold value by a preselected amount, the system being operable to output by the signaling mechanism a warning signal to the user indicative of the likelihood of a future onset of said seizure event.

15. The system as claimed in claim 14, wherein the system is operable to output said warning signal when at least three successive predicted data values differ from said threshold value.

16. The monitoring system as claimed in claim 14, wherein following the selection of the predicted next data value, selecting the predicted next data value as a new last transformed value y.sub.N associated with a new last time interval, the processor further including program instructions to: J. repeat steps F to I.

17. The monitoring system as claimed in claim 14, wherein the baseline period of time is selected at about 60 minutes, and the number of selected time intervals is selected at between about 150 to 200.

18. (canceled)

19. (canceled)

20. The system as claimed in claim 14, wherein the preselected threshold amount comprises a standard deviation of about 2.4±0.3.

21. The system as claimed in claim 14, wherein the baseline time series data sequence is divided into a plurality of said equally spaced time intervals, said initial time intervals being selected at between about 1 and 120 seconds.

22. The system as claimed in claim 14, wherein the initial recording portion of said baseline period of time selected at between about a first 10 and 30 minutes, said data signal values comprise electronic readings over a plurality of constant time intervals, and the baseline period of time is selected as a time period consisting of two or more of a user pre-seizure, a seizure and a user post-seizure event.

23. (canceled)

24. (canceled)

25. The system as claimed in claim 14, wherein the epileptic seizure event comprises a Tonic-clonic seizure, and wherein step J is performed to generate predicated next data values at next time intervals for a period of upto about ⅓ of the baseline period of time, and preferably is performed for at least about sixteen minutes.

26. (canceled)

27. An epileptic seizure monitoring and warning system for providing a user with advance warning of a likely seizure event, the system comprising: a signaling mechanism operable to provide a warning signal to said user indicative of a predicted epileptic seizure; a sensor assembly having at least one sensor operable to sense and output user sensed electroencephalographic (EEG) wave forms over an initial monitoring period, a computing device having a processor and memory, the computing device electronically communicating with said sensor assembly and operable to receive the output sensed signals over said initial monitoring period, and store in said memory sensed data values representative of said output wave forms at approximately equally spaced time intervals (x.sub.1, x.sub.2 . . . x.sub.N) over said initial monitoring period, the processor including stored program instructions operable to perform the process steps of: A. compile from said stored data values an initial data series sequence comprising associated ones of said sensed data values over a first timeline period, the first timeline period comprising between about 25% to 50%, and preferably about 33.3% of said initial monitoring period, wherein said initial data series sequence being represented as
S.sub.L={x.sub.1,x.sub.2 . . . x.sub.L} B. compute a first non-linear measure value V(S.sub.L)=y.sub.L of said initial data series sequence using at least one of fractal dimension, Lyapunov exponent and P&H; C. for a next and each subsequent remaining sensed data value over a remainder of the baseline monitoring period, compute an associated non-linear measure value (y.sub.L+1, y.sub.L+2 . . . y.sub.N) using at least one of fractal dimension, Lyapunov exponent and P&H, to form a transformed data series sequence,
(S.sup.m.sub.N)y.sub.L,y.sub.L+1,y.sub.L+2 . . . y.sub.N); D. for the transformed data series sequence S.sup.m.sub.N, calculate a reference non-linear value V(S.sup.m.sub.N) of the transformed data series sequence (S.sup.m.sub.L) using at least one of fractal dimension, Lyapunov exponent and P&H; E. determine a normal distribution curve of the Y values between each adjacent data point (y.sub.L, y.sub.L+1, y.sub.L+2 . . . y.sub.N) in the transformed value data series sequence S.sup.m.sub.N; F. with said normal distribution curve centered on the last data value y.sub.N, in said transformed data series (S.sup.m.sub.N), generate randomly a plurality of possible next data values (y.sup.j.sub.N+1) for a predicted next time interval (x.sub.N+1), as part of a separate generated extended time series sequences (S.sup.j.sub.N+1); G. for each said generated extended time series sequence S.sup.j.sub.N+, compute an associated non-linear measure value V(S.sup.j.sub.N+1) using at least one of fractal dimension, Lyapunov exponent or P&H; H. select the generated extended time series sequence having the associated non-linear measure value (V(S.sup.j.sub.N+i)) closest to the stored reference non-linear measure value V(S.sup.m.sub.N) as a next time series data sequence, and assigning the next data value for the selected extended time series sequence as a predicted data value (y.sub.N+1) for the predicted next time interval (x.sub.N+1); I. compare the predicted data value of the predicted next time interval with a predetermined threshold value indicative of said epileptic seizure, and J. repeat steps F through I for the selected extended time series data sequences, wherein the predicted data value of the selected extended time series data sequence is selected as a new last data value; and wherein when at least two consecutive said predicted data values exceed the predetermined threshold, the computing device activating said signaling mechanism to output said warning signal to the user

28. The monitoring system as claimed in claim 27, wherein the initial monitoring is selected at about 60 minutes, and the time intervals are selected at between about 10 and 60.

29. The monitoring system as claimed in claim 27, wherein step J is performed to generate predicted next data value at next time intervals of upto about one third the initial monitoring period.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Reference may be had with the following detailed description, taken together with the accompanying drawings, in which:

[0071] FIG. 1 shows schematically a system for predicting and outputting to a user an advance warning of a likely epileptic seizure in accordance with a preferred embodiment of the invention;

[0072] FIG. 2 illustrates graphically the user's recorded EEG data measured by the sensor of the system of FIG. 1 over a twenty minute monitoring and recording period, in accordance with the present invention;

[0073] FIG. 3 shows graphically the discretization of the recorded EEG measured data shown in FIG. 2 into a time series data sequence (x.sub.1, x.sub.2, x.sub.3 . . . x.sub.N) in accordance with the preferred embodiment of the invention;

[0074] FIG. 4 illustrates graphically the step of generating predicted next time data values used to effect timed series prediction to generate future predicted EEG values; and

[0075] FIG. 5 shows graphically the system output illustrating the predication of a future epileptic seizure, in accordance with the preferred embodiment of the invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0076] Reference may be had to FIG. 1 which illustrates a system 10 for use by a user 8 in predicting the likely occurrence of an epileptic seizure or event in accordance with a preferred embodiment of the invention. The system 10 includes a sensor assembly 12 having at least one electroencephalography (EEG) sensor 14 and personal digital assistant (PDA) 16.

[0077] As shown, the sensor 14 is adapted for placement in juxtaposed contact with a user's skull 18, and is operable to measure and record the electrical activity or electrical fluctuations of the user's brain. In one possible construction, the sensor assembly 12 may be provided as part of a smart glasses design, such as Google® glasses, or other such wearable technology. The sensor assembly 12 is operable to collect the EEG readings and wirelessly transmit them to the PDA 16 as a series of data readings or measurements taken over an initial sampling or monitoring period of from about ten to one-hundred and twenty minutes and preferably about twenty minutes. It is to be appreciated, that while FIG. 1 shows the system 10 as having a single EEG sensor 14, the invention could equally be provided with additional and/or different types of sensors 14 to detect and transmit to the PDA 16 more comprehensive data respecting the measured EEG readings.

[0078] The PDA 16 is provided with an internal antennae (not shown) adapted to wirelessly receive signals from the sensor assembly 12, and includes an internal memory 20, a processor 22, an audio output speaker 24 and a visual display screen 26. As will be described, the audio output speaker 24 and display screen 26 are operable in the use of the system 10 to provide the user 8 with both an audio and visual warning of a predicted likelihood of an impending epileptic seizure or event.

[0079] In a preferred mode of operation, the sensor 14 is operated to collect and transmit to the PDA 16 the user's EEG data over an initial sampling or monitoring period as time series T.sub.sample, where it is stored in the PDA memory 20 as measured relative EEG values. Preferably, EEG data is collected as a substantially continuous data file for the initial monitored period of time and thereafter, as for example is shown graphically in FIG. 2. More preferably, the monitoring period is chosen as a measured time period which is selected where the user 8 does not experience a seizure event such as a Tonic-clonic seizure, but also undergoes pre-seizure and/or post seizure EEG activities.

[0080] As will be described, the processor 22 includes programme instructions which are stored in memory, and which are operable to identify any measure the threshold value of transformed EEG signal values. The determined threshold value is then stored in the memory 20 as a preset threshold value which, based on the transformed historical data, provides a value above which is indicative of the occurrence a seizure event.

[0081] Once the initial monitored data is input and stored in memory 20, the processor 22 is used to transform the sensed data into a series of data values taken at equally spaced time intervals (i.e. preferably every twenty seconds). In particular, as shown in FIG. 3, the sensed data is used to generate a continuous baseline data sequence over period of 60 minutes, whereby data values are determined at each 20 second time interval (x.sub.1, x.sub.2, . . . x.sub.N) over the selected monitored time interval, as shown in FIG. 3. In generating the initial time series data sequence shown:

[0082] 1. The smallest time interval taken in the illustrated time series is the sampling time between x.sub.1 and x.sub.2, and the horizontal distance/time between points is always equal.

[0083] 2. In the data time series shown graphically, the point on the left is x.sub.1, with the final point or time interval on the right is x.sub.N (x.sub.N in the example shown in FIG. 3 is time sequence 180), with the subscription of each point increasing by 1 moving to x.sub.N; the sequence of points (x.sub.1, x.sub.2, . . . x.sub.N) in the initial measured time series data for the entire monitored 60 minute period may thus be expressed as S.sub.N or S.sub.N=(x.sub.1, x.sub.2, x.sub.N).

[0084] Following the establishment of the initial measured time series data sequence S.sub.N=(x.sub.1, x.sub.2 . . . x.sub.N), over the sixty (60) minute period a first twenty (20) minute interval L of data (S.sub.L) is chosen S.sub.L (x.sub.1, x.sub.2, x.sub.3 . . . x.sub.L).

[0085] A non-linear measure value V(S.sub.L) is then determined for the measured time series data sequence (S.sub.L) as a reference value. Preferably, the processor 22 is used to calculate the non-linear reference value for the series S.sub.L using one or more of “Fractal Dimension” or “Lyapunov” or “P&H”. The value V(S.sub.L) represents the measure of chaos in sample S.sub.L is plotted as a new value V(S.sub.L)=y.sub.L at time point L. As such, for the first initial interval V(S.sub.1)=y.sub.1.

[0086] The data window L is then moved to the right one data point. A next time series for a next interval S.sub.L+1=(x.sub.2, x.sub.3 . . . x.sub.L+1) is then chosen and a non-linear measure V(S.sub.L+1) computed using fractal dimension, the P&H value and/or Lyapunov exponent to generate a new V(S.sub.L+1) [i.e. V(S.sub.2)]y.sub.L+1 value [i.e. y.sub.2].

[0087] The process is then repeated for all of the remaining “x” values in the measured time series S.sub.N to generate a transformed data time series S.sub.LN (y.sub.L, y.sub.L+1 . . . y.sub.N) For the initial transformed data series S.sup.m.sub.N, possible mapping may be required, forming the new time series S.sup.m.sub.N.sup.={y.sub.L, y.sub.L+1, . . . y.sub.N}:

y.sub.i=V(S.sub.i−L+1,i),L≦i≦N where S.sub.i−L+1,i={y.sub.i−L+1,y.sub.i−L+2, . . . y.sub.N};

[0088] otherwise S.sup.m.sub.N=S.sub.N

where 0<L<N is the size of a sliding window used to compute the local level of chaos measured by V( ). Therefore, when the mapping is applied, the new considered time series S.sup.m.sub.N corresponds to the variation in time of the local non-linear measure in the initial time series S.sub.N.

[0089] V(S.sup.m.sub.N) is then determined as a reference value that will be used for predicting the next k values of the time series:

y.sub.N+i,1≦i≦k.

[0090] As will be described, based on the historical data collected during the initial 60 minute monitoring period T.sub.sample, the processor 22 is operable to read actual EEG then transform these values into values upon which future data values can be predicted. These predicted values may then be compared against the preselected threshold value to identify a likely epileptic event. In a simplified embodiment of the system 10, the PDA 16 outputs to the user 8 an alert signal or other identifier on the PDA speaker 24 and/or display 26, and preferably if three or more consecutive predicted future values exceed the preselected threshold value. Preferably, the PDA 16 is operable to provide a different warning or visual output signals to the user 8. Output warning signals may vary depending upon the resultant value of the predicted from the transformed EEG readings, and as it may relate to the probability of the seizure.

[0091] In a preferred operating mode, following the determination of the transformed time series S.sup.m.sub.N, a reference non-linear measure value of the transformed time series data V(S.sup.m.sub.N) is determined using Lyapunov exponent, P&H and/or fractural dimensions. The PDA processor 22 analyzes the transformed series S.sup.m.sub.N=y.sub.L, y.sub.L+1, . . . y.sub.N taking the value difference between y.sub.L+1 and y.sub.L+2, y.sub.L+3 and y.sub.L+4, (FIG. 3), and so on to y.sub.N, to calculate a normal distribution of the data series S.sup.m.sub.N on the Y-axis value N(y.sub.i, σ.sup.2). Using the normal distribution of values at each time period N(y.sub.i, σ.sup.2), a predicted next data point y.sub.N+ is calculated.

[0092] Preferably future time periods T.sub.L, T.sub.L+1, T.sub.L+2 . . . T.sub.N are chosen as equal constant intervals of time (s) over the same selected duration of between 5 and 60 seconds, and preferably about 20 seconds.

[0093] The processor 22 operates to generate and output predicted future data values, at such time periods based on transformed data values that are used to create predicted values for the time interval (x.sub.N+1) at point in time in the future, and preferably over a predicted future period of up to one third of the time covered by the measured baseline historical data points. Processor 22 performs a complex time series prediction based on an optimization process, whereby the processor 22 analyzes EEG data characteristics of the transferred time series S.sup.m.sub.N, and generates successively new predicted values y.sub.N+1 at successive points in time in the future, as continuing predicted time series. Further, as each new predicted data point is (y.sup.j.sub.N+i) generated, the processor 22 effects Lyapunov weighing and/or P&H methodology and/or fractal dimension to minimize the difference between the characteristic of the predicted new time series and the initial one.

[0094] A most preferred method for long-term time series prediction is shown graphically in FIG. 4. As shown, using the normal distribution calculated for the transformed initial time series S.sub.N described above, the distribution curve centered on the y.sub.N+i 1≦i≦k value of the last data point of the time series data sequence.

[0095] In particular, the parameter a of the normal distribution N(y.sub.i,σ.sup.2) 1≦i≦k of the transformed time series S.sup.m.sub.N is computed by computing the variation between every two consecutive values (i.e. y.sub.i to y.sub.i+1). This distribution represents the distribution of probability of value of y.sub.i, knowing y.sub.i−1 (FIG. 3).

[0096] Next, the processor 22 is used to generate randomly a number of potential predicted future values for the next time interval. The processor 22 preferably operates to generate, at least five to thirty, and preferably about ten new random values. In a simplified mode, random numbers are generated by way of a random number generator program for the next and as well be described, each subsequent time interval (y.sub.N+i+1) to be evaluated at the next and each subsequent point to be predicted for time T.sub.N+1+1. For predicting y.sub.N+i+1 Pos(y.sub.N+i+1), each randomly generated valve of the set of r random values, are plotted following the normal distribution N(y.sub.N+i, σ.sup.2) (FIG. 3). Therefore random r numbers are generated whereby:

Pos(y.sub.N+i+1)={y.sup.j.sub.N+i+1, 1≦j≦Nrand} is a parameter that can impact on the quality of the prediction, since having more values will increase the chance of finding an optimal value. However, it has been shown that in the analysis of EEG data significant improvement was not observed for the data when r was greater than ten.

[0097] For each of the random data values generated y.sup.j.sub.N+i+1, an associated extended generated time series is created (S.sup.m.sub.N+i+1=(y.sub.L, y.sub.L+1 . . . , y.sub.N, y.sub.N+1, . . . , y.sup.j.sub.N+i+1). The extended generated time series sequence in then used to compute an associated non-linear measure value V(S.sup.m.sub.N+i+1) using fractal dimension, P&H method and/or Lyapunov exponent. As such, for each separate data set containing each ten predicted point generated by the random number generator, a new “V” (i.e. V.sub.1, V.sub.2 . . . V.sub.10) value is established using the data sequence (y.sub.L, y.sub.L+1 . . . y.sub.N+i, y.sup.j.sub.N+i+1), where y.sup.j.sub.N+i+1 is one of the r new points generated by the random number generator.

[0098] The generated time series sequence having the associated non-linear measure value (V.sub.1, V.sub.2, V.sub.3 . . . V.sub.10) closest to the reference value V(S.sup.m.sub.N) is then chosen as the predicted next time series data sequence S.sup.m.sub.N+i+1=(y.sub.L, y.sub.L+1 . . . y.sub.N+i, y.sub.N+i+1). Further, the random data value y.sup.j.sub.N+i+1 for the selected next time series data sequence is assigned as the predicted data value for the next time interval T.sub.N+i+1.

y.sub.N+i+1 is thus computed by:

j.sub.min=arg min.sub.j(|V(S.sup.m.sub.N+i−1+y.sup.j.sub.N+i)−V(S.sup.m.sub.N)|) with (S.sup.m.sub.N+i−1+y.sup.j.sub.N+i={y.sub.1,y.sub.2, . . . ,y.sub.N+i−1,y.sup.j.sub.N+1})y.sub.N+1=y.sup.jmin.sub.N+i

[y.sup.jmin=minimum variance value between reference values and V values]

[0099] The value y.sup.j.sub.N+i.sup.i is chosen to make V(S.sup.m.sub.N+i+y.sup.j.sub.N+i+1) as close as possible to V(S.sup.m.sub.N).

Test Data

[0100] Preliminary testing suggests that the present method and system may achieve a high degree of accuracy in providing epileptic patients with advance warning of the likely onset of a seizure.

[0101] In preliminary testing, 21 patients diagnosed with epilepsy were monitored. In particular, EEG (electroencephalography) data from each patient was acquired using a Neurofile NT™ digital video EEG system with 128 channels, 256 Hz sampling rate, and a 16 bit analogue-to-digital convert. For each of the patients, there were datasets celled “ictal” and “interictal”. As shown in FIG. 3, the EEG signal was discretized as a time series vector, X={x.sub.1, x.sub.2, . . . x.sub.N} comprised of single electrical data readings at various time intervals and expressed as a series of individuals data points (single electrical readings by an electrode), where N is the total number of data points and the subscript indicates the time instant (FIG. 3). The P&H method was applied to the EEG time series to find the difference between seizure and seizure-free parts of EEG time series.

[0102] To evaluate performance of new method on prediction of epileptic seizure, the EEG time series measured by five electrodes, generating five different time series, for 21 patients were examined. For each EEG time series, the exact time of seizure was known and recorded. The P&H chaoticity values were predicted using GenericPred. The P&H chaoticity values were calculated on a constant-length (20 minutes) sliding window, with sliding time intervals of 20 seconds, of the EEG time series. During seizure, a peak in P&H values obtained from EEG time series appears. Based on the analysis of all 21 patients, a threshold for prediction of seizure determined at a preselected P&H value equal to 2.4 or greater (see FIG. 4) providing a reliable indication as a threshold EEG value indicative of the onset of a Tonic-clonic type seizure event. [0103] a. In undertaking testing to determine the prediction of a likelihood of epileptic seizures, 60 minutes of EEG data was selected as an initial base. This was based on the type of data observed and a best assessment. As a result, 180 data points (1 for every twenty (20) seconds) were analyzed as the initial time series data sequence; [0104] b. The time step or interval between data points/readings was chosen as a constant at 20 seconds; [0105] c. EEG data was transformed, to permit it to be used for predicting into the future; [0106] d. Time interval “L” was chosen at 20 minutes; and [0107] e. The total data series chosen S.sub.N=(x.sub.1, x.sub.2, . . . x.sub.N) was 60 minutes of data containing 180 data points.

[0108] The selection of 60 minutes of baseline data and the 20 minute time interval L were selected based on an expectation of reasonable values that would fit the case being evaluated, and were not provided as fixed values to be used in every application. As such, larger or shorter baseline and/or time interval data may be used. [0109] 1. The initial data was transformed to provide new predictive data series S.sup.m.sub.N={y.sub.L, y.sub.L+1, . . . y.sub.N} having new values as follows: [0110] i. Using “Fractal Dimension”, “Lyapunov” and/or “P&H”, calculate a “V” value for the data series S.sub.L comprising data points (x.sub.1, x.sub.2, . . . x.sub.L) over the first 20 minutes of data. This “V” value becomes the new data point/value y.sub.L=v(x.sub.1, x.sub.2, . . . x.sub.L) at time position “L” in the new transformed time series “S.sup.m.sub.L” [0111] ii. The 20 minute data interval was then shifted by one time interval to series (x.sub.2, x.sub.3, . . . x.sub.N+1), and a next “V” value is calculated for the shifted interval series which becomes the new data/point value at time position “y.sub.L+1=V(x.sub.2, x.sub.3, x.sub.N+1)” [0112] iii. The shifting of the 20 minute time interval is continued one data point or time interval position toward x.sub.N, and subsequent “V” values are calculated which become the new data/point value at that time position until the x.sub.N value has been transformed. This completes the data transformation to this point in time/data reading, providing an initial transformed data series S.sup.m.sub.N [0113] 2. For the initial transformed data series S.sup.m.sub.N created in 1 above (which runs from the L or 20 minute time point to the 60 minute time point and having 120 data points) calculate a “V” reference value using the same “Fractal Dimension”, “Lyapunov” and/or “P&H” as described above. This V(S.sup.m.sub.N) is stored as a “V” value used as a reference for predicting the next data point y.sub.N+1 values in the time series. [0114] 3. Now in the series S.sup.m.sub.N (which is from the 20 minute time mark to the 60 minute time mark) the value of the vertical difference between consecutive points y.sub.L, y.sub.L+1, y.sub.L+2 . . . y.sub.N is taken for all points and the normal distribution of these values calculated N(y.sub.i, σ.sup.2) [0115] 4. Using the normal distribution calculated, and centered on y.sub.N, create multiple and preferably about ten new random values y.sup.j.sub.N+i which are generated by a random number generator at the next projected time increment to be evaluated, for the next point to be predicted on time T.sub.N+i 1≦i≦k (and where for the first increment i=1) [0116] 5. For each of the new random values generated by the random number generator on time T.sub.N+i establish a new “V” value as above using its data sequence S.sup.j.sub.N+i=(y.sub.L, y.sub.L+1, . . . y.sub.N, y.sup.j.sub.N+i), where y.sup.j.sub.N+i is one of the 10 new points generated by the random number generator. As a result, (Vj) values V.sub.1, V.sub.2 . . . V.sub.10 are developed as V.sub.1=V(y.sub.L, y.sub.L+1, . . . y.sub.N+i, y.sup.1.sub.N+i), V.sub.2=V(y.sub.L, y.sub.L+1, . . . y.sub.N+i, y.sup.2.sub.N+), . . . V.sub.10=V(y.sub.L, y.sub.L+1, . . . y.sub.N, y.sup.10.sub.N+1). [0117] 6. Each of the V.sub.1 to V.sub.10 values are then compared to the references V(S.sup.m.sub.N) value, for the initial time series and the Vj with the value closest to V(S.sup.m.sub.N) is selected as the predictive time series, and the associated randomly generated value y.sup.j.sub.N+1 is chosen as the next new predicted point y.sub.N+i+1=y.sup.j.sub.N+i+1. [0118] 7. Calculations #3 to #6 are then repeated to establish the next data value prediction at a next time interval using y.sub.N+i+1 instead of y.sub.N+i. This is repeated until 20 minutes of data is projected into the future. [0119] 8. Next the calculation starts again after the next new raw data point x.sub.N+1 is received into the system.

[0120] It is to be appreciated that establishing “V” values using “Fractal Dimension”, Lyapunov” and/or “P&H” are based on what is more appropriate for the application. It may also be acceptable to calculate “V” values using a combination of values of two or more such methods (“Fractal Dimension”, “Lyapunov” or “P&H”).

[0121] Using the P&H threshold value, the current method was shown to predict future epileptic seizures with a high degree of sensitivity and specificity up to 17 minutes in advance (see Table 1 below). Further, different ranges of EEG time series were considered before and after seizure (we considered 10 ranges during seizure-free part of EEG time series for each patient) and there was no peak predicted by the current method in any case.

[0122] For each patient, one positive and 10 negative samples were constructed. The positive sample contains one epileptic seizure event, and the 10 negative samples are seizure-free. Therefore, there are 21 positive and 210 negative samples in total that were used to compute the specificity and the sensitivity accuracy levels.

TABLE-US-00001 TABLE 1 Sensitivity and specificity of epileptic seizure prediction for 21 patients for different lengths of prediction. Length of prediction before seizure Sensitivity Specificity 16 minutes ± 7 seconds 100%  100% 17 minutes ± 7 seconds 100%  100% 18 minutes ± 13 seconds 85% 100% 19 minutes ± 13 seconds 57% 100% 20 minutes ± 43 seconds 43% 100%
The same results were obtained by considering the data of any five electrodes independently. This is believed to represent an improvement over other predictive method, which typically achieves accuracy levels of 73% sensitivity and 67% specificity for 10 patients within a 1-19 minute range.

[0123] It is not anticipated that the current method will provide 100% sensitivity and specificity in all instances. Preliminary testing has, however, suggested that the system and method of the present invention shows strong promise in providing a good indicator of the likelihood of the onset on an epileptic event.

Further Applications

[0124] Although the detailed description describes the current method and system as most preferably being used for predicting epileptic seizures, the current system shows promise for a wide variety of different applications.

[0125] In an alternate mode, the method of the present invention may be used to predict the possible onset of a heart attack or stroke, as for example, by assessing chaotic variability of blood pressure changes, heart beat or heart arrhythmia. In yet another embodiment, the system 10 may be adapted for use as a medical warning device, as a predictor for the likelihood of the onset of seizure heart attack. The system 10 may include electro-cardiogram (ECG) in place of electroencephalogram (EEG) sensors to provide data representative of a patient's heart palpitations or arrhythmia over a historical or monitored time period.

[0126] In yet a further alternate possible application, the system 10 may be used as a predictor for future angina attacks. In particular, a patient's blood pressure data may be monitored over a selected period of time and input into the processor memory 20. By the aforementioned process, the processor 22 is activated to identify the future times where a potentially critical high blood pressure event is likely, and which correlates to a patient angina attack.

[0127] Again, on predicting the possible onset of such an occurrence, the system 10 could be used to provide either visual or audible warning to a user or medical practioner via the display 26. Alternately, if provided as part of an automatic drug dispensation system, the processor 22 could be used to output control signals to effect an adjustment of a pacemaker or an automated drug dispensation apparatus to alter the medical dosage of a patient's heart medication in anticipation of the possible angina event.

[0128] In a further non-limiting embodiment, the system 10 may be used to establish predictive environmental models. In one embodiment, data representing past measured amounts of vegetative growth of a particular plant or algae may be input for a selected historical time period. Using the foregoing method, the processor 22 may provide output data which is predictive of when a selected plant species may dominate or be subordinated relative to other species within a particular geographic area.

[0129] It is noted that establishing “V” values using “Fractional Dimension”, “Lyapunov” or “P&H” are based on what is more appropriate for the application. It may also be acceptable to calculate “V” values using the averaged values of one, two or all of these methods.

[0130] Each of the new non-linear data values (V.sub.1, V.sub.2 . . . V.sub.N) are compared with the earlier calculated (V.sup.m.sub.N) reference value, and generated time series with the closest corresponding value is selected, with its associated random data value chosen as the prediction for the next predicted time interval value in the time sequence. Using the generated time series sequence, the next subsequent predicted data value is determined by repeating steps of randomly generating and selecting data points by their approximation to the initial reference value. The process calculations may continue to be used to generate new predicted data values or points. Most preferably, number of new data points created in the sequence does not exceed one third of the total number of historic data points (N/3).

[0131] As a result, with the present method historical data may be rapidly updated. Most preferably, instead of making a shift of N data points at a time, a shift of a single data point is undertaken. That means that just one new real point value is measured (N+1) and then the new historical data to be taken into account are (2, 3, . . . , N+1), and the new prediction begin at N+2.

[0132] In the preferred mode, the reference value is always V(S.sup.m.sub.N), and which is obtained based on the value of the transformed non-linear data series from the original time interval. Therefore, with to the present method it is advantageous to keep the value of a transformed non-linear measure steady as much as possible during prediction (see FIG. 3). As used, the new predicted value is chose from a set of potential values generated from a distribution of probability in an acceptable selection range.

[0133] With the current system 10, prediction is performed using the complete time series whereas, in traditional approaches, after computation of the model, prediction is performed only using the model and no longer the original time series. Therefore, the current model allows for constant adjustment of information about the current time series, whereas classical predictive methods apply the model without taking into account the accordance between the original time series properties and the predicted ones. Moreover, the optimization step allows making choice among a set a potentially good predictive values, compared to the traditional models which only generate one value. Another advantage of the present invention is that it does not rely on a complex model of the original time series and it is therefore very general. Having no specialized model for prediction makes new method less restricted to a specific domain.

[0134] The system 10 most preferably incorporates built-in diagnostics software operable to verify that all aspects of the system 10 are functioning properly, and outputting via the display a green light signal confirming same. In the event that the system 10 encounters a functional problem, the user is alerted both by visual signal and audible signal of system malfunction, along with screen display as to the nature of the malfunction.

[0135] The present method shows a strong improvement compared to traditional methods over different situations and other chaotic time series in term of accuracy both for short and long term prediction. Moreover, the present method shows ability to predict the trend of evolution of other chaotic time series is much better than those of existing methods. Its performances are also more stable, with a standard deviation of the error measure appearing lower than those of the other methods. The method provides step toward an accurate and comprehensive time series long-term prediction.

[0136] It should be noted that preferred embodiment of the present method is not customized for a specific application, using a similar non-linear criterion may have the same function for a variety of applications. Further, by involving knowledge from other fields, it may be possible to provide a universal method for predicting a variety of non-linear time series. In another embodiment, the present method could utilize several non-linear measures simultaneously, instead of using just one measure, to identify and preserve the complexity of time series more efficiently.

[0137] Although the preferred embodiment describes the system and process for use in the predictive analysis of epileptic seizures, it is to be appreciated that the present process and system is equally applicable across a number of other possible applications. Such applications could include without restriction, applications in predicting macrogeographic events and trends; the predictive modeling of pandemics and pathogenic outbreaks; weather and meteorological modeling; and/or earthquake and geological event modeling. In addition, the system and method may further be used in the prediction and/or analysis of other complex data of non-linear events, including heart attack and/or stroke, as well as part of a health monitoring or warning system to provide an advance indication of other types of likely health events.

[0138] Although the disclosure describes and illustrates various preferred embodiments, the invention is not so limited. Many modifications and variations will now occur to persons skilled in the art. For a definition of the invention, reference may be had to the appended claims.