Support Feature Machine for Classification of Abnormal Brain Activity - PowerPoint Presentation

Slide1

Support Feature Machine for Classification of Abnormal Brain Activity

W. Art ChaovalitwongseRutgers University*Joint work with Y.J. Fan (Rutgers) and R.C. Sachdeo (Jersey Shore University Hospital)

This work is supported in part by research grants from NSF CAREER Grant CCF 05-46574 and Rutgers Research Council Grant-02028.

Poster #16, Mon Aug 13

Slide2

Agenda

Research ObjectivesResearch BackgroundEpilepsy and seizuresElectroencephalogram (EEG) time series Signal processingSupport Feature MachineEmpirical Results

Concluding Remarks

Slide3

Objectives:

Develop a new pattern recognition and classification framework for multi-dimensional time series data – as a decision making support.Application in EpilepsyIdentification of Seizure Pre-Cursor: Classification of seizure susceptibility periods

Quick Screening Tool: Classification of epilepsy and non-epilepsy patientsSeizure Prediction: Anomaly (Seizure Pre-Cursor) detectionGeneralize the framework to other applications that have to deal with multi-dimensional time series data.

Slide4

How many people having epilepsy?

“The Swamp”: Seating capacity ~90,000

Slide5

Epilepsy and Seizures

Nearly 3 million people in the U.S. (1% of population) have epilepsy. Anyone, at any age, can develop it.Epilepsy is defined as recurring seizures – sudden, brief changes in the way the brain works.Seizures cause temporary disturbances of brain functions such as motor control, responsiveness and recall which typically last from seconds to a few minutes.

Seizures usually occur spontaneously, in the absence of external triggers.Seizures may be followed by a post-ictal period of confusion or impaired sensorial that can persist for several hours.

Slide6

Intracranial EEG Acquisition

Slide7

Electroencephalogram (EEG)

…is a traditional tool for evaluating the physiological state of the brain. …offers good spatial and excellent temporal resolution to characterize rapidly changing electrical activity of brain activation

…captures voltage potentials produced by brain cells while communicating. In an EEG, electrodes are implanted in deep brain or placed on the scalp over multiple areas of the brain to detect and record patterns of electrical activity and check for abnormalities.

Slide8

10-second EEGs: Seizure Evolution

Normal

Pre-Seizure

Seizure Onset

Post-Seizure

Slide9

Open Problems

Seizure pre-cursors exist?Seizure is a state transition process?Can we discriminate normal EEGs from pre-seizure EEGs (seizure susceptibility period)?

Slide10

Data Transformation Using Chaos Theory

Measure the brain dynamics from time series:Stock MarketCurrency Exchanges (e.g., Swedish Kroner)Apply dynamical measures (based on chaos theory) to non-overlapping EEG epochs of 10.24 seconds = 2048 points.

Maximum Short-Term Lyapunov Exponentmeasure the average uncertainty along the local eigenvectors and phase differences of an attractor in the phase spacemeasure the stability/chaoticity of EEG signalsIasemidis, Shiau, Chaovalitwongse, Sackellares & Pardalos, IEEE Transactions on Biomed (2003)

Slide11

Measure of Chaos

Slide12

Classification of Physiological States

Slide13

Support Vector Machine VS Support Feature Machine

Feature 1

Feature 2

Feature 3

Feature 1

Feature 2

Feature 3

Slide14

Nearest Neighbor for Time Series

Normal

Pre-Seizure

A

d

1

: Average distance to

blues

.

d

2

: Average distance to

reds

.

d

2

<

d

1

, so new point is classified as

red

.

Slide15

Similarity Measures

Dynamic Time Warping (DTW) DistanceEuclidean DistanceT-Statistical Distance

X

Y

D(X, Y)

STLmax

1, 2, 3, ….. , 30

Electrode 1

2

3

.

.

.

.

.

26

STLmax

1, 2, 3, ….. , 30

Slide16

Support Feature Machine

Given an unknown epoch of EEG signals A, we calculate statistical distances between the EEG epoch and the groups of Normal and Pre-Seizure EEGs in our data baseline.

Euclidean distanceT-statistical distanceDynamic Time WarpingEEG sample A will be classified in the group of patient’s state (normal or pre-seizure) that yields the minimum statistical distance.Multiple Electrodes = Multiple DecisionsAveraging

Majority Voting: selects action with maximum number of votesCan we select/optimize the selection of a subset of electrodes that maximizes number of correctly classified samples.

Chaovalitwongse et al.,

Submitted to

Operations Research

Slide17

Decision Rule: Basic Ideas

Two different average distances for each sample at each electrode are calculated:Intra-Class: Average distances from each sample to all other samples in the same class at Electrode jInter-Class:

Average distances from each sample to all other samples in different class at Electrode jIf for Sample i at Electrode j (Averaging VS Voting)

Average distance to the same class

Average distance to different class

<

Then Sample

i

is

correctly

classified.

Slide18

Optimization Model I: Averaging

Intra-Class

Inter-Class

Chaovalitwongse et al.,

Submitted to

Operations Research

Slide19

Model I: Averaging Formulation

Chaovalitwongse et al.,

Submitted to KDD, 2007 and Operations Research

Slide20

Optimization Model II: Voting

Precision matrix,

A

contains elements of

Chaovalitwongse et al.,

Submitted to

Operations Research

Slide21

Decision Rule: Basic Ideas

Two different average distances for each sample at each electrode are calculated:Intra-Class: Average distances from each sample to all other samples in the same class at Electrode jInter-Class:

Average distances from each sample to all other samples in different class at Electrode jIf for Sample i at Electrode j (Averaging VS Voting)

Average distance to the same class

Average distance to different class

<

Then Sample

i

at

Electrode

j

is

correctly

classified.

Chaovalitwongse et al.,

Submitted to

Operations Research

Slide22

Model II: Voting Formulation

Chaovalitwongse et al.,

Submitted to

KDD, 2007 and Operations Research

Slide23

Data Selection and Sampling

EEG Dataset Characteristics

Patient ID

Seizure types

Duration of EEG(days)

# of seizures

1

CP, SC

3.55

7

2

CP, GTC, SC

10.93

7

3

CP

8.85

22

4

,SC

5.93

19

5

CP, SC

13.13

17

6

CP, SC

11.95

17

7

CP, SC

3.11

9

8

CP, SC

6.09

23

9

CP, SC

11.53

20

10

CP

9.65

12

Total

 

84.71

153

CP: Complex Partial; SC subclinical; GTC: Generalized Tonic/Clonic

Randomly and uniformly sample 3 EEG epochs per seizure from each of normal and pre-seizure states.

For example, Patient 1 has 7 seizures.

There are 21 normal and 21 pre-seizure EEG epochs sampled.

Seizure

Seizure

Duration of EEG

30 minutes

30 minutes

8 hours

8 hours

8 hours

8 hours

Pre-seizure

Normal

Slide24

Sensitivity and Specificity

Sensitivity measures the fraction of positive cases that are classified as positive. Specificity measures the fraction of negative cases classified as negative.

Sensitivity = TP/(TP+FN)Specificity = TN/(TN+FP)

Sensitivity can be considered as a detection (prediction or classification) rate that one wants to maximize.

False positive rate can be considered as 1-Specificity which one wants to minimize.

Slide25

5-Fold Cross Validation Result

Chaovalitwongse et al., Submitted to Operations Research

81.29% 72.86%

Optimize the number of neighbors

Chaovalitwongse et al.,

IEEE Trans Systems, Man, and Cybernetics: Part A

, 2008

Slide26

DTW

Euclidean

T-Statistics

Slide27

DTW

Euclidean

T-Statistics

Slide28

 

Com

 

User

Interface

Technology

Multichannel

Data Acquisition

Pattern

Recognition

Initiate a variety of

therapies (e.g., electrical stimulation, drug injection)

VNS

Automated Seizure Prediction Paradigm

Drug

Feature Extraction/ Cluster Analysis

Slide29

Concluding Remarks

Overview of a Real Life Medical Problem in Spatio-Temporal Data MiningApplications of Data Mining and Optimization TechniquesPotential Applications in Medical DiagnosisAutomated seizure warning system

Monitoring devices for clinical use in epilepsy monitoring units (EMUs) and intensive care units (ICUs)Other monitoring procedures in trauma and operation roomsImprovement of the Nearest Neighbor Classification in Time Series Classification - New Classification Framework

Slide30

W. Chaovalitwongse

, Y.J. Fan, R.C. Sachdeo. Novel Optimization Models for Multidimensional Time Series Classification: Application to the Identification of Abnormal Brain Activity. Submitted to Operations Research.Y.J. Fan, W. Chaovalitwongse, L.D. Iasemidis

, R.C. Sachdeo. Multi-Dimensional Time Series Classification for Identification of Epilepsy Patients. Submitted to KDD 2007.W. Chaovalitwongse, Y.J. Fan, and R.C. Sachdeo. On the K-Nearest Dynamic Time Warping Neighbor for Abnormal Brain Activity Classification. To appear in IEEE Transactions on Systems, Man, and Cybernetics, Part A: Systems and Humans

, 2008.W. Chaovalitwongse and P.M. Pardalos. On the Time Series Support Vector Machine using Dynamic Time Warping Kernel for Brain Activity Classification. To appear in

Cybernetics and Systems Analysis

, 2007

W. Chaovalitwongse

, P.M. Pardalos, and O.A. Prokopyev. Electroencephalogram (EEG) Time Series Classification: Applications in Epilepsy.

Annals of Operations Research

, 148: 227-250, 2006.

W. Chaovalitwongse

, L.D.

Iasemidis

, P.M. Pardalos, P.R. Carney, D.-S. Shiau, and J.C. Sackellares. A Robust Method for Studying the Dynamics of the Intracranial EEG: Application to Epilepsy.

Epilepsy Research, 64, 93-133, 2005

.

W. Chaovalitwongse

, P.M. Pardalos, L.D.

Iasemidis

, D.-S. Shiau, and J.C. Sackellares. Dynamical Approaches and Multi-Quadratic Integer Programming for Seizure Prediction.

Optimization Methods and Software, 20 (2-3): 383-394, 2005

.

Reference

Slide31

Acknowledgements

Comprehensive Epilepsy Center, St. Peter’s University HospitalRajesh C. Sachdeo, MD

Rutgers Ph.D. StudentYa-Ju Fan, MS

Industrial and Systems Engineering, University of Florida

Panos M. Pardalos, PhD

Bioengineering, Arizona State University

Leonidas D.

Iasemidis

, PhD