Pulvino-cortical interactions

2026-05-21

Update on pilot + Next steps

• 2 EMU staff members know how to help set up: Siri, Jason (the supervisor)

• TODO document the setup steps
  • Hospital sEEG software
    • Manually add trigger channels (DC01-16)
    • Select electrode montage (all REF)
  • Hardware connections
    • hospital Nihon-Kohden sEEG amplifier
    • parallel port cable
    • RippleNeuroMed sEEG acquisition system
    • Behrmann lab sEEG laptop + MATLAB

sEEG logistics at Presby: Setup

• Write bytes to parallel port with MATLAB's Xippmex `digout`
• Need to send ON + OFF pulses manually (I use 5 ms width)
• Not all trigger channels work reliably across all hospital sEEG systems used for different participants, so I send the same pulse across channels for redundancy [can't encode any info besides 1 bit]

sEEG logistics at Presby: Triggers

• Received access to their data storage server: Jorge's lab manager Jay Chen
• Data files: given start/end timestamps, Jay provides EDF file + MRI + CT

• Electrode referencing/ground channel are still unclear
• Channels are named inconsistently: some weird names (ST5, SP4, ... instead of A6, B5)
  • Can be corrected/imputed based on other info

sEEG logistics at Presby: sEEG data

• Last we discussed...
  • standard sEEG experiment
    • present visual stimuli with a simple task
    • record from pulvinar + cortex
  • intra-operative experiment (shelved for later)
    • present simple ON/OFF flashes via goggles
    • record from pulvinar + cortex

Main experiment: From last time

• Does pulvinar activity precede/modulate V1?
• Is this affected by emotional valence, e.g. only for threatening stimuli?
• What about more general pulvino-cortical interactions?
  • e.g. visual cortex
  • e.g. association cortex

(I've been slowly going through the papers you cited in the grant proposal)

Main experiment: Questions

• 1 stimulus = 2 s video + 1 s blank
• 7 stimulus conditions
  • 3 types of faces [identity-matched]
    • neutral
    • smiling
    • fearful
  • 4 types of objects/scenes
    • negative threats
    • evolutionary threat
    • neutral
    • positive
• 4 stimuli per condition
• 2 repetitions per stimulus

Main experiment: Task and stimuli

• 2 tasks (no fixation)
  • 1-back
  • 2-AFC valence detection
• 4 runs per task
  • all conditions mixed, randomized
  • distinct stimuli across runs

• Each run is ~3 min
• ~24 min total

• Small subset of stimuli repeated 10x to estimate channel reliability?
• Do we want particularly shocking, maximally salient stimuli?
• If V1, do we want fixation? pRF mapping?

1-back task (no fixation)

• 8 stimulus conditions
  • 4 types of faces [identity-matched]
    • neutral / surprise; happiness; fear; anger
  • 4 types of objects/scenes
    • neutral; positive; evolutionary threats; modern threats
• In each run [all conditions mixed; distinct stimuli in each run]
  • 1.5 s per stimulus [1 s video + 0.5 s blank, eased in/out]
  • 4 stimuli per condition @ 2 repetitions
    +1 stimulus per condition @  8 repetitions
  • = 5 stimuli per condition
  • = 128 trials + 16 1-back repeats (8:1 ratio)
  • ~ 3.6 min  [144 trials]
• Scale number of runs depending on patient (8 runs ~ 30 min)

Bonus task: pRF mapping (with fixation) [Note: seizures!]

Post-discussion update

Main Qs

• Does pulvinar activity precede/modulate V1 (and other visual/association cortex)?
• Is this affected by emotional valence, e.g. for threatening stimuli?

Bonus Qs

• How does the channel response change over repeated stimulus presentations?
• Can we characterize pRFs with iEEG?

Measurements

• sEEG
• behavior (limited, attention check only)
• eye-tracking (if possible)

visual threat processing is rapid for "ancestral"/"evolutionary" threats

visual threat detection in subcortex?

More generally, one might perform an Intracranial EEG Investigation of Human Pulvinar and Cortical Dynamics During Visual and Cognitive Processing

Goal

- show naturalistic videos

- record brain responses (cortex + subcortex)

- see if there's a latency difference

- (possibly modulated by stimulus type)

Stimulus types

- faces (neutral, happy, angry, scared)

- scenes (neutral, positive, ancestral threat, modern threat)

Stimulus curation

clip and crop 1 s videos from pexels.com or existing databases

normalize inter-eye distances

+ study more general pulvinar-cortex interactions

• JavaScript and web development
• video processing
  • ffmpeg
  • Handbrake/Kdenlive
• UniFace Python library for face processing

What is a confound, even?

"well ackchyually, [X] cares about these low-level confounds, not snakes more generally"

But X these low-level features covary with semantics---maybe the brain leverages them to detect high-level things!

Maybe we throw the baby out with the bathwater if we control for every possible "confound"

e.g. Partialling out early DNN features

(But also, I'm worried about this)