Targeted brain stimulation to modulate memory in humans...

1
Targeted brain stimulation to modulate memory in humans 353.14/HHH21 Data Collection and Analysis: intracranial EEG / SR = 500,1000 or 1600ms / wavelet decomposition / 50 log-spaced freqs 1-200Hz / Record-only N = 102 Stimulation details: 50Hz pulse frequency / 4600ms duration / biphasic 300 μs / onset -200ms relative to word presentation. Stimulation N = 52 This work was supported by the DARPA Restoring Active Memory (RAM) program (Cooperative Agreement N66001-14-2-4032). The views, opinions, and/or findings contained in this material are those of the authors and should not be interpreted as representing the official views or policies of the Department of Defense or the U.S. Government. Experimental Design Memory success depends on neural processes engaged at the time that items are encoded. Encoding operations vary in their efficiency from moment to moment (Kahana 2012). We asked how direct electrical stimulation modulates brain states conducive to successful memory encoding. Using stimulation of intracranial electrodes, we attempt to modulate neural activity and consequently memory performance (Halgren et al. 1985; Suthana et al. 2012). How does stimulation’s effect on physiology predict its effect on memory? References Halgren at al. (1985). Brain and Cognition / Haufe et al (2014) NeuroImage Kahana (2012). Foundations of Human Memory / Suthana et al. (2012). NEJM Task Details: 1600ms presentation / 750-1000ms jittered ISI / 20s distractor / 30s recall 25 lists per session / 12 words per list Classifier Details: L2 penalized logistic regression / spectral power over 1600ms encoding interval / z-scored input patterns / model evaluation leave-one-out by session Background Summary Overall Effects of Stimulation on Behavior Multivariate Analysis Discriminates Memory States Classifier Predicts Behavioral Effects of Stimulation [email protected] Stimulation increases memory performance if delivered in poor encoding states. Suggests strategies for optimally applying stimulation to affect memory function. Evidence for the spectral tilt predicted stimulation’s effects on neural activity. 1 Dept of Psychology, University of Pennsylvania; 2 Dept of Neurological Surgery, University of California San Francisco Medical Center; 3 Depts of Neurology and 4 Radiology, Thomas Jefferson University Hospital; 5 Dept of Neurology, Mayo Clinic; 6 Dept of Neurology, 7 Neurosurgery and 8 Radiology, Hospital of the University of Pennsylvania; 9 Dept of Neurosurgery, Emory University; 10 Dept of Neurosurgery, University of Texas Southwestern; 11 Dept of Neurology, Dartmouth-Hitchcock Medical Center; 12 Dept of Neurosurgery, Columbia University; 13 Surgical Neurology Branch, National Institutes of Health Youssef Ezzyat 1 , James Kragel 1 , John Burke 2 , Deborah Levy 1 , Logan O’Sullivan 1 , Paul Wanda 1 , Michael Sperling 3 , Greg Worrell 5 , Michal Kucewicz 5 , Kathryn Davis 6 , Timothy Lucas 7 , Cory Inman 9 , Bradley Lega 10 , Barbara Jobst 11 , Sameer Sheth 12 , Kareem Zaghloul 13 , Joel Stein 8 , Sandhitsu Das 6 , Richard Gorniak 4 , Daniel Rizzuto 1 & Michael Kahana 1 Word Encoding Trial Classifier Output Probability 0 0.2 0.4 0.6 0.8 1 List 1 List 2 List 3 List 4 List 5 List 6 List 7 List 8 False Alarm Rate 0.2 0.4 0.6 0.8 1 Hit Rate 0.2 0.4 0.6 0.8 1 AUC = .76 Region IFG MFG SFG MTLC Hipp TC IPC SPC OC Frequency (Hz) 2 3 5 8 13 23 40 68 116 200 Model-Based Activation -0.5 -0.25 0 0.25 0.5 False Alarm Rate 0.2 0.4 0.6 0.8 1 Hit Rate 0.2 0.4 0.6 0.8 1 Mean AUC: 0.63 t(101) = 19.2, P < 10 -10 Region IFG MFG SFG MTLC Hipp TC IPC SPC OC Frequency (Hz) 2 3 5 8 13 23 40 68 116 200 Subsequent Memory (t-stat) -0.8 -0.4 0 0.4 0.8 LOCK FUR GOLD ... ... Low High Low-High State Difference Region IFG MFG SFG MTL Hipp TC IPC SPC OC Frequency (Hz) 2 3 5 8 13 23 40 68 116 200 Mean t-stat -8 -6 -4 -2 0 2 4 6 8 P < 0.003 P < 0.03 P < 0.05 Stim-NoStim Classifier (t-stat) -5 0 5 N 0 2 4 6 8 10 12 Mean = 0.67 P < 0.01 Low States -5 0 5 N 0 2 4 6 8 10 12 Mean = -1.52 P < 0.001 High States Stim-NoStim Classifier (t-stat) Low High Stim-NoStim Normalized Recall -250 -150 -50 50 150 Preceding State Serial Position 4 8 12 Probability of Recall 0 0.4 0.8 Lag -5 -3 -1 1 3 5 Cond. Response Probability 0 0.4 0.8 Stim Lists NoStim Lists Record-Only Normalized Stim-No Stim Recall -75 -50 -25 0 25 50 75 # Patient Stimulation Sessions 0 4 8 12 16 Mean = -6.8% P < 0.04 Spectral Tilt (HFA - LFA t-stat) -2 -1.5 -1 -0.5 0 0.5 1 Classifier Output (Stim-NoStim) -0.2 -0.1 0 0.1 0.2 0.3 r(37) = 0.54 P < 0.001 Use classifier decoding to estimate brain state pre-stimulation Stimulation increases memory performance and memory state estimates if delivered in low states Within patient, classifier significantly discriminates successful and unsuccessful memory states Classifier significant across patients Important features for classification reflect the ‘spectral tilt’ pattern Stimulation has modest behavioral effects across the group, but some larger effects within individual patients PreStim Data Recalled Forgotten TAPE MOLE DAD FLOOD 2+3+1 Encoding Stimulation Sessions Frequency Electrode “MOLE” LOCK FUR ... Frequency Electrode ... ... Encoding Classification Pre-stimulation Post-stimulation GOLD HEN NoStim control NoStim control Stim Distractor Recall Sessions 25 NoStim Lists 20 Stim / 5 NoStim ... Record Stimulation ... “FLOOD” ... Method: Haufe et al (2014) Subsequent Mem. Analysis

Transcript of Targeted brain stimulation to modulate memory in humans...

Page 1: Targeted brain stimulation to modulate memory in humans ...memory.psych.upenn.edu/files/pubs/EzzyEtal16.poster.pdfFoundations of Human Memory / Suthana et al. (2012). NEJM Task Details:

Targeted brain stimulation to modulate memory in humans 353.14/HHH21

Data Collection and Analysis: intracranial EEG / SR = 500,1000 or 1600ms / wavelet decomposition / 50 log-spaced freqs 1-200Hz / Record-only N = 102Stimulation details: 50Hz pulse frequency / 4600ms duration / biphasic 300μs / onset -200ms relative to word presentation. Stimulation N = 52

This work was supported by the DARPA Restoring Active Memory (RAM) program (CooperativeAgreement N66001-14-2-4032). The views, opinions, and/or findings contained in this materialare those of the authors and should not be interpreted as representing the official views orpolicies of the Department of Defense or the U.S. Government.

Experimental Design

Memory success depends on neural processes engaged at the time that items are encoded.

Encoding operations vary in their efficiency from moment tomoment (Kahana 2012).

We asked how direct electrical stimulation modulates brain statesconducive to successful memory encoding.

Using stimulation of intracranial electrodes, we attempt tomodulate neural activity and consequently memory performance(Halgren et al. 1985; Suthana et al. 2012).

How does stimulation’s effect on physiologypredict its effect on memory?

ReferencesHalgren at al. (1985). Brain and Cognition / Haufe et al (2014) NeuroImage Kahana (2012). Foundations of Human Memory / Suthana et al. (2012). NEJM

Task Details: 1600ms presentation / 750-1000ms jittered ISI / 20s distractor / 30s recall 25 lists per session / 12 words per listClassifier Details: L2 penalized logistic regression / spectral power over 1600ms encoding interval / z-scored input patterns / model evaluation leave-one-out by session

Background

Summary

Overall Effects of Stimulation on Behavior

Multivariate Analysis Discriminates Memory States Classifier Predicts Behavioral Effects of Stimulation

[email protected]

Stimulation increases memory performance if delivered in poorencoding states.

Suggests strategies for optimally applying stimulation to affectmemory function.

Evidence for the spectral tilt predicted stimulation’seffects on neural activity.

1Dept of Psychology, University of Pennsylvania; 2Dept of Neurological Surgery, University of California San Francisco Medical Center; 3Depts of Neurology and 4Radiology, Thomas Jefferson University Hospital; 5Dept of Neurology, Mayo Clinic; 6Dept of Neurology, 7Neurosurgery and 8Radiology, Hospital of the University of Pennsylvania; 9Dept of Neurosurgery, Emory University; 10Dept of Neurosurgery, University of Texas Southwestern; 11Dept of Neurology, Dartmouth-Hitchcock Medical Center; 12Dept of Neurosurgery, Columbia University; 13Surgical Neurology Branch, National Institutes of Health

Youssef Ezzyat1, James Kragel1, John Burke2, Deborah Levy1, Logan O’Sullivan1, Paul Wanda1, Michael Sperling3, Greg Worrell5, Michal Kucewicz5, Kathryn Davis6, Timothy Lucas7,Cory Inman9, Bradley Lega10, Barbara Jobst11, Sameer Sheth12, Kareem Zaghloul13, Joel Stein8, Sandhitsu Das6, Richard Gorniak4, Daniel Rizzuto1 & Michael Kahana1

Word Encoding Trial

Cla

ssifi

er O

utpu

tPr

obab

ility

00.20.40.60.8

1

List 1 List 2 List 3 List 4 List 5 List 6 List 7 List 8

False Alarm Rate0.2 0.4 0.6 0.8 1

Hit

Rat

e

0.20.40.60.8

1

AUC = .76

RegionIFG

MFGSFG

MTLCHipp TC IPCSPC OC

Freq

uenc

y (H

z)

2358

13234068

116200

Mod

el-B

ased

Act

ivat

ion

-0.5

-0.25

0

0.25

0.5

False Alarm Rate0.2 0.4 0.6 0.8 1

Hit

Rat

e

0.2

0.4

0.6

0.8

1

Mean AUC: 0.63t(101) = 19.2, P < 10-10

RegionIFG

MFGSFG

MTLCHipp TC IPCSPC OC

Freq

uenc

y (H

z)

2358

13234068

116200

Subs

eque

nt M

emor

y (t-

stat

)

-0.8

-0.4

0

0.4

0.8

LOCK FUR GOLD ......

Low HighLow-High State Difference

RegionIFG

MFGSFG

MTLHipp TC IPC

SPC OC

Freq

uenc

y (H

z)

2358

13234068

116200

Mea

n t-s

tat

-8

-6

-4

-2

0

2

4

6

8

P < 0.003

P < 0.03 P < 0.05

Stim-NoStim Classifier (t-stat)-5 0 5

N

0

2

4

6

8

10

12Mean = 0.67P < 0.01

Low States

-5 0 5

N

0

2

4

6

8

10

12Mean = -1.52

P < 0.001

High States

Stim-NoStim Classifier (t-stat)

Low High

Stim

-NoS

tim N

orm

aliz

ed R

ecal

l

-250

-150

-50

50

150

Preceding State

Serial Position4 8 12

Prob

abilit

y of

Rec

all

0

0.4

0.8

Lag-5 -3 -1 1 3 5C

ond.

Res

pons

ePr

obab

ility

0

0.4

0.8

Stim ListsNoStim Lists

Record-Only

Normalized Stim-No Stim Recall-75 -50 -25 0 25 50 75

# Pa

tient

Stim

ulat

ion

Sess

ions

0

4

8

12

16 Mean = -6.8%P < 0.04

Spectral Tilt (HFA - LFA t-stat)-2 -1.5 -1 -0.5 0 0.5 1

Cla

ssifi

er O

utpu

t (St

im-N

oStim

)

-0.2

-0.1

0

0.1

0.2

0.3

r(37) = 0.54P < 0.001

Use classifier decodingto estimate brain

state pre-stimulation

Stimulation increases memory performance andmemory state estimates if delivered in low states

Within patient, classifier significantly discriminatessuccessful and unsuccessful memory states

Classifier significantacross patients

Important features for classificationreflect the ‘spectral tilt’ pattern

Stimulation has modest behavioral effects acrossthe group, but some larger effects within individual patients

PreStimData

Recalled Forgotten

TAPEMOLE

DADFLOOD

2+3+1

Encoding

Stimulation Sessions

Freq

uenc

y

Electrode

“MOLE”

LOCK

FUR

...

Freq

uenc

y

Electrode

... ...

Encoding Classification

Pre-stimulation

Post-stimulation

GOLD

HEN

NoStim control

NoStim control

Stim

Distractor

Recall

Sessions

25 N

oStim

Lis

ts

20 S

tim /

5 N

oStim

...

Record Stimulation

...

“FLOOD”

...

Method: Haufe et al (2014) Subsequent Mem. Analysis