brain computer interfaces
play

Brain Computer Interfaces Physiologic basis for feature selection, - PowerPoint PPT Presentation

Brain Computer Interfaces Physiologic basis for feature selection, and decoding techniques Brain Computer Interfaces For dexterous motor control Hochberg 2012 Brain Computer Interfaces Control of end effectors Brain Computer


  1. Brain – Computer Interfaces Physiologic basis for feature selection, and decoding techniques

  2. Brain – Computer Interfaces For dexterous motor control Hochberg 2012

  3. Brain – Computer Interfaces Control of end effectors

  4. Brain – Computer Interfaces Control of end effectors Communication

  5. Brain – Computer Interfaces Control of end effectors Communication Neuromodulation to replace lost senses

  6. Brain – Computer Interfaces Control of end effectors Communication Neuromodulation to Other Neuromodulation / Biofeedback replace lost senses

  7. Brain – Computer Interfaces Consumer BCI Control of end effectors Communication Neuromodulation to Other Neuromodulation / Biofeedback replace lost senses

  8. Brain – Computer Interfaces Many Applications -> Many Engineering Requirements -> Many Architecture Considerations

  9. Brain – Computer Interfaces Many Applications -> Many Engineering Requirements -> Many Architecture Considerations But in general: need to isolate , translate , and utilize a neural signal

  10. Brain – Computer Interfaces Many Applications -> Many Engineering Requirements -> Many Architecture Considerations But in general: need to isolate , translate , and utilize a neural signal

  11. Architecture of a BCI Classification / Regression

  12. Architecture of a BCI Classification / Regression

  13. BCI Signal Types Non-invasive EEG (scalp) ECoG (brain surface) Intracortical (Picture credit: Wadsworth Center) Electrodes Invasive

  14. BCI Signal Types Signal Cell count Raw Feature Z Spatial Signal Magnitude (depends) Specificity Stability EEG (non- > 1M ~50 uV 3-5 1-5 cm Long-term? invasive) ECoG 500K ~500 uV 10-20 3-10 mm Months (semi- invasive?) Intracortical 1-??? 10s of mV Very high < 300 um Days (invasive) Appropriate modality choice depends on application. Consider subject population. Research/Clinical goals. Stimulation requirements.

  15. Architecture of a BCI Classification / Regression

  16. Feature extraction, intracortical recordings

  17. Feature extraction, intracortical recordings The quest for single units mean firing rates

  18. Feature extraction, intracortical recordings The quest for single units Ensemble spiking mean firing rates

  19. Feature extraction, intracortical recordings The quest for single units Local Field Potentials (LFPs) Ensemble spiking mean firing rates

  20. Feature extraction, ECoG and LFPs

  21. Feature extraction, ECoG and LFPs Spectral Estimation: STFFT Wavelets Band filtering and envelope detection Auto-regressive model

  22. Feature extraction, EEG Signal spreads as it passes through meat

  23. Feature extraction, EEG Signal spreads as it passes through meat 1) Correct for spatial spreading Common Spatial Patterns – Linear combination of electrodes maximizing two class discriminability

  24. Feature extraction, EEG Signal spreads as it passes through meat 1) Correct for spatial spreading Common Spatial Patterns – Linear combination of electrodes maximizing two class discriminability Use of spherical head model as solution to forward model

  25. Feature extraction, EEG Signal spreads as it passes through meat 1) Correct for spatial spreading Common Spatial Patterns – Linear combination of electrodes maximizing two class discriminability Use of spherical head model as solution to forward model Subject specific MRI as solution to forward model

  26. Feature extraction, EEG Signal spreads as it passes through meat 1) Correct for spatial spreading Common Spatial Patterns – Linear combination of electrodes maximizing two class discriminability Use of spherical head model as solution to forward model Subject specific MRI as solution to forward model 2) Apply same spectral estimation techniques used in ECoG (50 Hz and below) for SMR and SSVEP Or Simple LPF for EPs

  27. Architecture of a BCI Classification / Regression

  28. Decoding, intracortical recordings Translation of neural signal to one or more continuous variables

  29. Decoding, intracortical recordings Translation of neural signal to one or more continuous variables Population Vector   − ∑ r r =   ˆ 0 d p   i   r i max i

  30. Decoding, intracortical recordings Translation of neural signal to one or more continuous variables Population Vector Kalman Filter Estimate Update   − ∑ r r =   ˆ 0 d p   i   r i max i Bonus: Incorporates effector kinematics

  31. Decoding, intracortical recordings Translation of neural signal to one or more continuous variables Population Vector Kalman Filter Estimate Update   − ∑ r r =   ˆ 0 d p   i   r i max i Bonus: Incorporates effector kinematics Many Others: Neural Networks, ARMA Models, etc

  32. Decoding, ECoG Translation of neural signal to one or more continuous variables, High SNR allows (causes  ) us to be lazy.

  33. Decoding, ECoG Translation of neural signal to one or more continuous variables, High SNR allows (causes  ) us to be lazy. mean std dy/dt = (x-mu) / std

  34. Decoding, EEG Much harder computational problem, because of low SNR Neural signal typically translated to discrete variable with pre-defined (and pre-trained) number of states SVM, Naïve Bayes, Decision Trees, LDA Random Forest, Neural Network, on and on… BCI competition Non-linear transform + LDA

  35. An Inherent Problem

  36. The Underlying Model visual feedback effector position effector decoder effector movement movement target position visual feedback effector effector Inverse position decoder Task goal effector movement model target movement position Separately capable of adaptation

  37. Closed-loop decoder adaptation

Recommend


More recommend