Method and system for predicting outcomes based on spatio/spectro-temporal data

1. A computer-implemented method of creating an evolving spiking neural network personalized model Mx for determining a future outcome or a classification, for a scenario of interest, that is specific to a subject x having input data SSTDx, wherein the input data SSTDx of subject x comprises a number of static and temporal/spatio-temporal variables relating to the scenario of interest and wherein the model Mx comprises a three-dimensional spiking neural network reservoir and a spiking neural network classifier, the method comprising:

extracting, from a global dataset D of samples, the samples having the same static and temporal/spatio-temporal variables relating to the scenario of interest as SSTDx and for which an outcome or classification is known, a subset Kx of samples having static variables that most closely accord to the static variables of the input data SSTDx,

encoding the temporal/spatio-temporal variables of Kx into spike sequences, wherein the spike sequences represent individual input patterns;

constructing the three-dimensional spiking neural network reservoir to comprise spatially located neurons;

mapping the temporal/spatio-temporal variables of Kx to spatially located reservoir input neurons according to: (i) a known spatial location of the spatio-temporal variables of Kx, whereby spatially close spatio-temporal variables are mapped to spatially close input neurons; or (ii) a temporal similarity of the temporal variables of Kx, whereby high correlated spike sequences are mapped to spatially close input neurons;

training the spiking neural network reservoir to learn the spike sequences;

constructing the spiking neural network classifier; and

training the spiking neural network classifier to recognise and classify similar spike sequences in relation to the known outcomes or classifications.

2. The method of claim 1 , wherein the step of mapping spatio-temporal variables into the reservoir comprises mapping spatio-temporal brain data, including EEG, fMRI and other brain data, using a standard brain structural template.

3. The method of claim 1 , wherein the spiking neural network classifier is also trained, when recalled on new data, to classify spike sequences before the entire spike sequence has propagated through the spiking neural network reservoir.

4. The method of claim 1 , wherein the subject is an individual.

5. A method of determining a future outcome or classification, for a scenario of interest, specific to subject x, the method comprising:

inputting SSTDx data into the model Mx created according to the method of claim 1 to determine from the model Mx the future outcome and/or the classification for subject x.

6. A computer-implemented method of generating a prediction of stroke occurrence for an individual using the method of claim 5 .

7. A computer-implemented method of classifying unclassified SSTD brain data (STBD) using the method of claim 5 .

8. The method of claim 7 , wherein the input data SSTDx is or comprises STBD of the subject x and wherein the global dataset D of samples comprises an archive of STBD, and wherein the model Mx is a brain data classification model, wherein the method comprises:

training the brain data classification model using the archive brain data, the archive brain data comprising STBD with known classifications, wherein training the brain data classification model comprises the steps of:

providing the archive brain data as inputs to the spiking neural network reservoir to train the spiking neural network reservoir to provide similar spiking activity patterns for similar input spike sequences to the spiking neural network classifier; and

using the known classifications to train the spiking neural network classifier to recognise and classify similar spiking activity patterns; and

using the brain data classification model to classify the unclassified brain data SSTDx of subject x.

9. A computer-implemented method of controlling a state of an apparatus using brain signals of the subject x using the method of claim 7 , wherein the scenario of interest is a desired state of the apparatus, wherein the method comprises:

determining the desired state of the apparatus based on the classified brain data; and

sending data indicative of the desired state of the apparatus to a controller, wherein the controller is configured to control the apparatus in accordance with the desired state of the apparatus.

10. The method of claim 9 wherein the method of controlling a state of an apparatus is applied to a method of rehabilitation, wherein the subject x is a patient, the patient being equipped with one or more rehabilitation actuator(s) adapted to induce movement of a body part, the brain data corresponding to a desired movement of the body part, the method comprising:

controlling the rehabilitation actuator(s) using brain activity of the patient to produce the desired movement of the body part; and

if necessary, repeating the method until the patient improves his/her brain signals and body activity to produce the desired movement without the rehabilitation actuator.

11. A computer-implemented method of generating a prediction of an ecological event based on spatio-temporal data using the method of claim 5 .