So you want to read minds?

Today

1. Introduction to EEG
2. Walk through EEG terms and technology using OpenVibe
3. Given enough pointers for people trying their own experiments

Hi!

• Former: software engineer / maths student.
• Now: Going into neuroscience, specializing in data processing.
• Member of Vancouver NeuroTechX
• Code: http://github.com/padster
• Blog: https://padsterprogramming.blogspot.com

Intro: What are brains?

• Nodes = Neurons, ~100 billion (?)
• Edges = Synapses, ~100 trillion (?)
• Signal = electrical current

What is EEG?

• Non-invasive!
• Good temporal resolution: Fast
• Bad spatial resolution: measure voltage on scalp
• Lots of research into what can be identified.
• Sensitive: voltage is around µV, susceptible to movement, ...

Example!

• X axis = Time
• Y axis = Voltage, one row per sensor

Measure example: OpenBCI

Process example: OpenVibe

Installing OpenVibe...

• Visit the OpenVibe website downloads page
• At time of writing, v1.3.0 supports windows and linux

...Run the Designer

~/openvibe$ ./dist/openvibe-designer.sh

...What is going on‽‽

Option: use existing data

Load and Visualize

Rewrite signal values into CSV format

  import scipy.io as sio
  import csv

  SUBJECT = 'aak'
  IN_FILE = 'data/musicbci_VP%s.mat' % SUBJECT
  OUT_FILE = 'dataCSV/%s_signal.csv' % SUBJECT

  print "Loading matlab file %s..." % IN_FILE
  data = sio.loadmat(IN_FILE)
  readings = data['data'][0]['X'][0]

  # First line: Row count, then channel names, then sample rate
  channelNames = [str(c[0]) for c in data['data'][0]['clab'][0][0]]
  sampleRate = data['data'][0]['fs'][0][0][0]

  print "# Readings: %d" % readings.shape[0]
  print "# Channels: %d" % len(channelNames)
  print "Sample rate: %d hz" % sampleRate

  # sampleRate samples/sec = 1000 / sampleRate ms/sample
  secPerSample = 1.0 / sampleRate

  print "Writing to %s..." % OUT_FILE
  with open(OUT_FILE, 'wb') as csvfile:
      writer = csv.writer(csvfile)
      writer.writerow(['Time (s)'] + channelNames + ['Sampling Rate'])
      # First row has extra rate:
      writer.writerow([0.0] + readings[0].tolist() + [sampleRate])
      for r in range(1, readings.shape[0]):
          writer.writerow([r * secPerSample] + readings[r].tolist())

  print "Done!"

Run conversion...

~/openvibe/musicbci$ time python signalToCSV.py
Loading matlab file data/musicbci_VPaak.mat...
# Readings: 2277240
# Channels: 64
Sample rate: 200 hz
Writing to dataCSV/aak_signal.csv...
Done!

real	2m30.080s
user	2m25.902s
sys	0m3.841s

View in Designer

Concept: Frequency Analysis

Power bands in designer

1. Read signal from CSV, as before.
2. Use Channel selector box to pick some channels.
3. Convert to frequencies with Spectral Analysis (FFT)
4. Draw to screen with Power spectrum display

Result

Example: Alpha Waves

• Add Temporal filter to restrict to 8-15Hz
• Display the result with the usual signal display
• Also add a Spectrum average and display with Matrix display

Concept: Events

Rewrite CSV format

  import scipy.io as sio
  import csv

  SUBJECT = 'aak'
  IN_FILE = 'data/musicbci_VP%s.mat' % SUBJECT
  OUT_FILE = 'dataCSV/%s_events.csv' % SUBJECT

  print "Loading matlab file %s..." % IN_FILE
  data = sio.loadmat(IN_FILE)

  trials = data['data'][0]['trial'][0]
  values = data['data'][0]['y'][0]
  assert trials.shape == values.shape

  eventCount = trials.shape[1]
  sampleRate = int(round(data['data'][0]['fs'][0][0][0]))
  print "# Events: %d" % eventCount

  print "Writing to %s..." % OUT_FILE
  with open(OUT_FILE, 'wb') as csvfile:
  writer = csv.writer(csvfile)
  writer.writerow(['Time (s)', 'Identifier', 'Duration'])
  # First row has extra rate:
  for i in range(eventCount):
  msOffset = int(round(trials[0, i])) * 1000 / sampleRate
  # eID = values[0, i]
  eID = 33024 + values[0, i] # HACK: http://openvibe.inria.fr/stimulation-codes/
  writer.writerow([msOffset / 1000.0, eID, 1])

  print "Done!"

In Designer

• Right-Click 🡺 Modify Outputs 🡺 Output Stream 🡺 Stimulations
• Purple arrow = correct (see here)

Concept: Event Related Potentials (ERPs)

• P = Positive voltage (= down...)
• 300 = 300 milliseconds after stimulus

• Two CSV readers, load signal and events
• Filter for some channels (for simplicity)
• Stimulation-based epoching
• OVTK_StimulationId_Label_03 used
• Draw with signal display, plus ERP plot

Extra

• Epoch Variance box allows the calculation of the mean and variance across multiple epochs.
• ERP plot can also save the result to file on a particular signal.

Concept: ICA

• Noise reduction: find the 'blink' component, and remove it
• Virtual channels: e.g. reduce to 'frontal' signal, 'left hemisphere', ...

Localization

In Designer

• Apply channel selection for P8;P10;F7;F9
• Run through Independant Component Analysis box

Concept: Localization

In Designer, 2D

• Load data from CSV, as before, and exclude "EOGvu" (reference).
• Load locations from .txt, using localization file reader.
• Display with 2D Topographic Map.

Bonus! '3D'

Live streaming!

Acquisition Server

~/openvibe$ ./dist/openvibe-acquisition-server.sh

Muse (via LSL)

1) Muse -> LSL

$~/muse$ muse-io --device DEVICE_ID --lsl-eeg muse 

2) LSL -> Aquisition Server

1. Select signal stream to the name above, and marker to None.
2. Connect to your device, make sure it's successful
3. Play to start the streaming

View in designer

• Acquisition client box to read from the server
• Connect to the stream visualizer (or whatever you like)

OpenBCI

1) OpenBCI to 'PC' mode and plug in USB

2) Aquisition server:

Saving...

🢂

More on OpenVibe...

• Connectivity box for synchronicity between various parts of the brain.
• Automatic Classification trainer and processor: use machine learning techniques (SVM, LDA) to learn how to identify various events, if possible

Thanks!

More on NeuroTechX...

• Join the slack: https://neurotechx.herokuapp.com/
• Keep an eye on future events in Vancouver, posted on Meetup