Development of Implantable

Interfaces to Restore Motor Function

Karunesh Ganguly, MD PhD

Osher Mini Medical School “UCSF Scientists Outline What’s to Come”November 22, 2019

Introduction• Concept of bio-interactive neural

interfaces dates to early 20th century

• Successful translation of − Cochlear implants− Deep brain-stimulation (DBS)− Responsive stimulation (RNS)

• Neural Interface for paralysis and rehabilitation

− ‘Brain-Machine Interfaces’/’Brain-Computer Interfaces’

MECHANIX ILLUSTRATED (1957)

3

Cochlear Implants

• Auditory nerve stimulation research starting in the 1950s

• ~22,000 adults and ~15,000 children live in the US with cochlear implants

Image credit: UCSF Benioff Children’s Hospital

4

Deep-Brain Stimulation

Source: NIH, Neurology

Neural Interfaces for Communication and Movement

Motor Disability in the US

www.cdrf.org

Locked-In (e.g. TBI, Stroke, ALS)

Upper Limb paralysis

VS/MCS (cardiopulmonary arrest, TBI, etc)

Quadriplegia, SCI (CP,ALS, MS, etc)

Comatose

Rehabilitation Needs Vary

7

Paraplegia

Below ~C5/C6

Above ~C5

Loss

of I

ndep

ende

nce

8

Patient Rehabilitation Goals

Anderson, J Neurotrauma (2004)

9

Motor Dysfunction

Brain-Computer Interface (BCI)

Move left..

Also known as “Neural Interface” and “Brain-Machine Interface/BMI”

Extracellular Recording of Activity

11

Spikes Electrode

Neurons

ECoGElectrode

Field potentials

Spikes

12

Recording neural activity from the brainfMRI, MEG

Schwartz et al. Neuron, 2006

13

Recording neural activity from the brain

Buzsaki et al., Nat Neuro Reviews, 2012

Principles Underlying BCI Control

14

15

Volitional control of brain activity

Fetz, E, Science, 1969

1 cm

M1

16Georgopoulos AP, J. Neurosci (1982); Dum & Strick (2002); Schwartz, J. Physiology (2006)

Volitional control of movementsA B

BCIs grounded by > 40 years of research into movement control!

1 cm

PPSMA

PMdS1 M1

Distributed neural population activity

From Wessberg et al., 2000; but see also Chapin et al., 1999, Serruya et al., 2002; Taylor et al., 2002; Carmena et al., 2003; Velliste et al., 2008; Pohlmeyer et al., 2009

Recording from neural populationsM1 (L)M1(R) PMd (L)

1.2mVpp

0.5mVpp]

1.3mVpp

0.3mVpp

19

Real-time decoding of hand position

Ganguly et al., 2009

Actual hand movementsDecoder predictions

“Closed-Loop” Brain-Computer Interfaces

Neural Signals

“Actuator”• Prosthetics• FES• Stimulation

“Decoders”

Feedback

Sensors

Neural plasticity key for stable BCI control

Ganguly et al., PLoS Biology, 2009; Ganguly et al., Nat Neuro 2011; Gulati et al, Neuro Neuro 2014, 2017 Nat Neuro; Kim et al., Cell 2019

Examples of BCI Control in Human Subjects

22

Development of BCI Control

• Real-time recording of ensembles Chapin et al., 1999; Wessberg et al., 2000

• Multiple demonstrations of ‘closed-loop’ control over an external deviceSerruya et al., 2002; Taylor et al., 2002; Carmena et al., 2003; Velliste et al., 2008; Pohlmeyer et al., 2009

• Human subjects can learn direct neural control of a computer cursorKennedy et al., 2000; Leuthardt et al., 2004; Hochberg et al., 2006

• Clear demonstration that a subject’s intentions can be translated

24

EEG-interface for communication• Two patients with advanced ALS (ventilated, PEG tube for > 4

years) could learn to communicate• Patients were ‘locked-in’ (no voluntary muscle movements)• EEG signals could be used to type a message (but very slow!)

Birbaumer et al., Nature (1999)

Case Report: A 51yo molecular biologist with ALS…

25Sellars et al., Amyotroph Lateral Scler. (2010)

BCI Control of a Computer Cursor

26FROM Collinger et al., Lancet (2013)See also Hochberg et al., 2006, 2012; Wodlinger et al., 2014

27

BCI Control in a tetraplegic subject

FROM Pandarinath et al., eLife(2017).

Translational Challenges

28

• Signals are not stable à Utah Array

• Daily training required because of signal stability

• Complex setups that require lot of support

• Currently not wireless (this will be solved soon)

Ongoing UCSF studies- ECoG BCI Trial- Neural Interfaces for Stroke

29

ECoG based chronic implant

30** Experimental Device à FDA Investigational Device Exemption (IDE)

• FDA cleared testing of a chronic ECoG based device

• Addresses a downside of current trials• Signal instability• Daily training

• Two primary goals are control of a typing interface and a complex robotic arm

• Leverage stability of ECoG signals to engage learning and plasticity

‘Plug-and-Play’ BCI Control

31

‘Neural maps’ Across Days

32

Impaired hand function after stroke

• ~700,000 strokes/year

• >$15K for rehabilitation per patient– Limits of rehab (ICARES Trial, 2018)

• ~50% with hand impairments, limits independence

Towards the Development of a Closed-Loop Neural Interface for Stroke

Neuralsignal

EMG

Stimulation

What is required for closed-loop modulation?- Electrophysiological targets?- Predictor of good/poor recovery?- Can targeted stimulation help chronic deficits?

34

Synchronous Network Activity During Skilled Movements

Lemke et al., Nature Neuroscience, 2019

M1

DLS

M1

DLS

Muscle

M1

DLS

Time (sec)-1 1

Neur

ons

Neur

ons

Changes in Speed/Consistency with Learning

1

Tria

l #

600Accuracy (%)

Time (sec)

0 70

0 1.2

M1-DLS Coherence

Lemke et al., Nature Neuroscience, 2019

Reemergence of synchronous activity

0 100Accuracy

Tria

l #

250

Accuracy (%)Tr

ial #

260

1

0Time (s)

-.5 1.5

PLC electrode

Stroke

1.5-4 Hz LFP PowerReach

Ramanathan*, Guo*, Gulati* et al. Nature Medicine (2018)

Responsive Stimulation

37

Onset ofmovement

Limb position

Temporal electrical stimulation

Session

Stim (400 μA)No-stim

S1 S5

Accu

racy

(%) 75

25

Responsive Stimulation

SummaryInjured CNSIntact CNS

VA/VL

DLS

PLC

GPi

GPe

M2

DMS

S1/S2

Cerebellum

STROKE

Limb positionSpikes

Field Potential

Neural Activity

Temporally precise stimulation

ECoG in Stroke

40

Fugl-Meyer of 35

Ramanathan*, Guo*, Gulati* et al. Nature Medicine (2018)

Reduced task-related slow-oscillations

41

PLC electrode

Stroke

a

Elec

trode

im

plan

t +

PT s

troke

Time (sessions)

Reach onlyReach with neural recording

b c

B S1 S10

B = BaselineS = Session # Ac

cura

cy (%

)

100Stroke

0

d

e gf

r = 0.52***-10

80

-0.2 0.8Power change (z-sc)

Succ

. cha

nge

(%)

h

0.2

Time from reach (s)-1 0 1.50.08

SFC

EarlyLate

1.5-4Hz SFC

i j

k l

1.5-4Hz power

2.5

-0.25

Freq

. (H

z)

1.5

6

-0.05 0.45Time (s)

Early power spatial map

Late power spatial map

0.05

Time from reach (s)

20

-1

15

10

5

1.5 0 1.5

Freq

uenc

y (H

z)

-1 0 1.5

0.2

SFC

Early SFC Late SFC

-0.2

Time from reach (s)

20

-1

15

10

5

1.50 1.5

Freq

uenc

y (H

z)

-1 0 1.5

0.6

z-sc

ore

Early power Late power0.5

Time from reach (s)-1 0 1.5-0.1

z-sc

ore

Time from reach (s)

Uni

ts

-1 1.50

Firin

g ra

te (z

-sc)

-2

4

-1 1.50

Early unit firing Late unit firing1 1

170 219

Time from reach (s)-1 1.50

LFP

Uni

tsPE

TH

RetractReachGrasp1

2630

01.2

-1.2

Tria

l #

0-1 1.5Time from reach (s) Succ. (%)

0 100

S1

S11

1

255

1.5-4Hz Power

RatHuman

Ramanathan*, Guo*, Gulati* et al. Nature Medicine (2018)

Thanks for your attention!

42

Research Funding