Scalp EEG Separation During RSVP Tasks

Scalp EEG Separation During RSVP Tasks

Kaushik Majumdar

Indian Statistical Institute

Bangalore Center

Done with Robert M. Frank in EGI & University of Oregon in 2006

Neurotechnology for Intelligent Analysts

• Funded by DARPA of USA in 2006.

• 8 teams involved.

The RSVP Task

• Grey scale satellite images were presented at 3/s rate.

• # interesting object : # non-interesting objects ≈ 1 : 3.

• 256 channel geodesic sensor net of the EGI.

Target Objects

• Tank

• Truck

• Boat

Scalp Potential

Parra et al., NeuroImage 28 : 328 – 341, 2005.

Weighted Average EEG

Parra et al., NeuroImage 28 : 328 – 341, 2005.

Raster Plot of Two Channels

Parra et al., NeuroImage 28 : 328 – 341, 2005.

FP1 Channel of a Subject

Huang et al., http://www.bme.ogi.edu/~huangca/papers/C8.pdf

FP1 in 10 – 20 System

http://en.wikipedia.org/wiki/10-20_system_(EEG)

ERP Scalp Image of FP1 of a Subject

Huang et al., http://www.bme.ogi.edu/~huangca/papers/C8.pdf

96 tank trials versus 342 non-tank trials for subject 1

PLR Fisher Two cluster

Tank 0.9571 0.7321 0.9433

Non-tank 0.5491 0.8386 0.6036

ROC area 0.8912 0.8300 0.7807

96 tank trials versus 57 truck trials for subject 1

PLR Fisher Two cluster

Tank 0.5692 0.9975 0.6946

Truck 0.4737 0.9860 0.8519

ROC area 0.5227 0.9975 0.7743

Logistic Regression

Parra et al., NeuroImage 28 : 328 – 341, 2005.

Fisher’s Linear Discriminant

Two Cluster Discriminant

THANK YOU

Slides available at http://www.isibang.ac.in/~kaushik/ISSR-2010.ppt