Monitoring an EEG

• US 5,816,247 A
• Filed: 06/13/1996
• Issued: 10/06/1998
• Est. Priority Date: 06/13/1995
• Status: Expired due to Term

• Current Assignee: RDM Consultants, Ltd.
• Original Assignee: RDM Consultants, Ltd.
• Inventors: Maynard, Douglas E.
• Primary Examiner(s): Bahr, Jennifer
• Assistant Examiner(s): Ruddy, David M.
• Application Number:
  US08/663,385
• Time in Patent Office: 845 Days
• Field of Search: 128/731, 128/734, 395/21
• US Class Current: 600/544
• CPC Class Codes: A61B 5/374 Detecting the frequency distribution of signals, e.g. detecting delta, theta, alpha, beta or gamma waves
• CPC Class Codes: A61B 5/7267 involving training the classification device
• CPC Class Codes: A61B 5/742 using visual displays A61B5/339 takes precedence
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems
• CPC Class Codes: G16H 50/70 for mining of medical data, e.g. analysing previous cases of other patients
• CPC Class Codes: Y10S 706/924 Medical

Abstract

An apparatus and method for EEG monitoring provides multi-dimensional classification of EEG samples, using a neural network having multiple outputs trained upon a training set of samples to define an n-dimensional space in which to classify the samples and provide to the user a display of that space.

Claims

1. Apparatus for electroencephalographic (EEG) patient monitoring, the apparatus comprising:

input circuitry for receiving signals from patient measurements, including signals representative of patients'"'"' brain waves, the input means including amplifier means and signal processing means for amplitude and spectral analysis of the signals to provide for each of a plurality of epochs of the signals a set of signals including signals representative of a plurality of amplitude and frequency properties of the brain waves;

means for applying statistical processing to the set of signals to derive therefrom a set of values representative of the brain waves in the associated signal epoch;

input means for defining for each of a plurality of said epochs n where n is at least equal to 2 classifying values so that each epoch is categorised by n values and wherein of those n values there are x where x is greater than which are multivalued to define an x-dimensional space;

neural network means including means defining adjustable weightings and having a training mode for deriving, from training input data comprising a plurality of said sets of values and from target outputs comprising the associated defined classifying values, output classifying values representative of patient conditions and a classifying mode for deriving from one of said sets of values n output classifying values obtained from the neural network means as conditioned by training in the training mode,

the neural network means comprising;

means for calculating in said training mode weighting errors between output classifying values and the target outputs corresponding to the input data from which the output classifying values are derived;

means for defining as a substantially zero weighting error any calculated weighting error which is less than a predefined distance from the classifying value of the epoch concerned; and

means for adjusting said weightings of the neural network means in dependence upon determined weighting error;

display means for displaying in x-dimensional form the x output classifying values of the neural network means; and

memory means for storing data values obtained from the input circuitry.

2. Apparatus for electroencephalographic (EEG) patient monitoring of patient states, the apparatus comprising:

input circuitry for receiving signals including signals representative of brain waves of the patients;

signal processing means for amplitude and for spectral analysis of the signals to provide for each patient a set of signals, including signals representative of a plurality of amplitude and frequency properties of the brain waves;

means for applying statistical processing to each set of signals to derive therefrom respective data sets of values;

means defining an n-dimensional where n is at least equal to 2 space;

input means for defining categories for said data set by inputting coordinate sets defining coordinates in said space to categorise respective states of patients;

display means for displaying a zone containing at least a first and a second multivalued dimension of said n-dimensional space;

supervised learning neural network means having a plurality of layers, with each layer including adjustable weighting, the neural network means having;

(a) a training mode for acting upon a plurality of said sets of values obtained from respective patient measurements and from a corresponding plurality of target outputs which are defined coordinates defining the category of those patients to adjust said weightings to produce therefrom output classifying values in said space and representative of the category of the patients; and

(b) a classifying mode for deriving, from a set of said values and said adjusted weightings, output classifying coordinate values in, and for display by the input means in, said space; and

memory means for storing data values obtained from the input circuitry and signal processing means.

3. Apparatus as defined in claim 2 wherein the neural network means has:

means for calculating weighting errors between the outputs of a layer and target outputs corresponding to the associated input data set;

means for adjusting the weightings in dependence upon any calculated weighting error; and

means for defining as a zero weighting error any calculated weighting error less than an error value corresponding to a predetermined distance value which is at least 0.25 of the distance from the position in said space of said set and nearest neighbours thereof.

4. Apparatus according to claim 3 wherein said error value is a predetermined distance between 0.4 and 0.5 of said distance.

5. A method of determining EEG categories of patients, the method comprising:

obtaining a plurality of sets of training data from patient data including EEG records of a plurality of patients;

determining categories for classifying said sets, each category being defined by n where n is an integer greater than 1 classifying items;

defining an n-dimensional space in which said determined categories are represented as distinct coordinates of which at least two dimensions are multivalued;

training a neural network with n outputs, utilising said plurality of sets as input data and said defined categories as target outputs; and

applying a set of EEG data from a patient to the neural network to derive therefrom a position in said space, whereby a patient category can be assessed.

6. A method according to claim 5 wherein obtaining training data comprises:

making EEG measurements upon a plurality of patients to obtain respective patient records;

obtaining for each of a plurality of sections of said records a first set of values representing statistical relationships between amplitudes of said records;

dividing each such section into bands of different frequency ranges;

obtaining from those ranges for each section a second set of values representing statistical relationships between magnitudes in each band over the associated section; and

said sets of training data comprise combinations of the first and second sets of values.

7. A method according to claim 5, and comprising effecting the step of statistical analysis on the EEG records to define outliers in respect of said categories.

8. A method according to claim 6 wherein n is greater than 2 and wherein the categories of data include a classifying item representing a coordinate in one of the dimensions of said space other than the dimensions of said at least two coordinates.

9. A method according to claim 5 wherein the neural network has a plurality of layers with each layer including adjustable weighting and the training of the neural network comprises:

calculating weighting errors between the outputs of a layer and target outputs corresponding to the associated input data set;

adjusting the weightings in dependence upon any calculated weighting error; and

defining as a zero weighting error any calculated weighting error less than an error value corresponding to a predetermined distance value which is at least 0.25 of the distance from the position in said space of said set and nearest neighbours thereof.

10. A method according to claim 5 wherein the training of the neural network comprises:

calculating weighting errors between the outputs of a layer and target outputs corresponding to the associated input data set;

adjusting the weightings in dependence upon any calculated weighting error; and

defining as a zero weighting error any calculated weighting error at a given distance from the position in said space of said set;

defining a negative error within said distance.

11. A method according to claim 5, wherein the training of the neural network comprises:

defining for each example of said input data a boundary about its target output;

calculating weighting errors in dependence upon the distances between the outputs of a layer and the boundaries corresponding to the target outputs; and

adjusting said weightings in dependence upon any calculated weighting errors.

12. A method according to claim 11, wherein the calculating step comprises calculating the weighting errors so as to generate negative errors within the boundaries, thus to tend to drag each item of input data towards its boundary.

13. Apparatus for electroencephalographic (EEG) patient monitoring of patient states, the apparatus comprising:

input circuitry for receiving from a plurality of electrodes attached to sectors of patients'"'"' heads signals representative of brain waves of the patients;

signal processing means for amplitude and for spectral analysis of the signals to provide for each patient a set of signals, including signals representative of a plurality of amplitude and frequency properties of the brain waves;

means for applying statistical processing to each set of signals to derive therefrom respective data sets of values;

means defining an n-dimensional where n is at least equal to 2 space;

input means for defining categories for said data set by inputting coordinate sets defining coordinates in said space to categorise respective states of patients;

display means for displaying a zone containing at least a first and a second substantially continuous dimension of said n-dimensional space;

supervised learning neural network means having a plurality of layers, with each layer including adjustable weighting, the neural network means having;

(a) a training mode for acting upon a plurality of said sets of values obtained from respective patient measurements and from a corresponding plurality of target outputs which are defined coordinates defining the category of those patients to adjust said weightings to produce therefrom output classifying values in said space and representative of the category of the patients;

(b) a classifying mode for deriving, from a set of said values and said adjusted weightings, output classifying coordinate values in, and for display by the input means in, said space; and

(c) means for calulating a delta-logistic parameter for weighting adjustments of layers as a vector; and

memory means for storing data values obtained from the input circuitry and signal processing means.

14. A method of determining EEG categories of patients, the method comprising:

the step of obtaining a plurality of sets of training data from patient data including EEG records of a plurality of patients;

the step of determining categories for classifying said sets, each category being defined by n where n is an integer greater than 2 classifying items;

the step of defining an n-dimensional space in which said determined categories are represented as distinct coordinates of which at least two coordinates are substantially continuous;

the step of training a supervised learning neural network with n outputs, utilising said plurality of sets as input data and said defined categories as target outputs, the training of the neural network comprising;

calculating weighting errors between the outputs of a layer and target outputs corresponding to the associated input data set;

adjusting the weightings in dependence upon any calculated weighting error;

defining as a zero weighting error any calculated weighting error at a predefined distance from the position in said space of said set and said given distance is a function of the standard deviation from a population mean of the associated category; and

defining a negative error within said distance; and

the step of applying a set of EEG data from a patient to the neural network to derive therefrom a position in said space, whereby a patient category can be assessed.

15. An apparatus for EEG patient classification from data conveyed by measurements carried out upon a patient, the measurements comprising EEG measurements of brain waves and the apparatus comprising:

input means arranged for the inputting to the apparatus of;

(a) a plurality of sets of training data representative of measurements obtained from a plurality of patients;

(b) categories for classifying said sets and each category being defined by n where n is an integer greater than 2 classifying items of which at least a first and a second are multivalued and a further classifying item categorises the sets into at least two types; and

(c) a set of data obtained from measurements carried out upon a patient;

means arranged to define a category space having n-dimensions in which space said categories are represented as distinct coordinates;

means for displaying at least a portion of said space defined by dimensions corresponding to at least said first and second of said n-dimensions; and

means defining a supervised learning neural network with n outputs and having;

(a) a training mode for utilising said plurality of sets as input data and said categories as target outputs to derive a trained version of said neural network; and

(b) a classifying mode for using the trained neural network to derive a patient category in said n-dimensional space from said set of data obtained from a patient.

16. An apparatus according to claim 15, and comprising means responsive to input data in which the further classifying item defines the presence and absence of outliers in data obtained from measurements carried out upon a patient so that the display means is able to display a portion of said space substantially free of such outliers.