1 Decoding Neural Activity from an Intracortical Implant in Humans with Tetraplegia Chad E. Bouton Battelle Health and Life Sciences Division.

11

Decoding Neural Activity from an Intracortical Decoding Neural Activity from an Intracortical Implant in Humans with TetraplegiaImplant in Humans with Tetraplegia

Chad E. BoutonChad E. Bouton

BattelleBattelleHealth and Life Sciences DivisionHealth and Life Sciences Division

22

BackgroundBackground

• Cyberkinetics (founded in 2001) developed a Brain-Computer Interface (BCI) system called BrainGate

• At the core of the system was the Utah Electrode Array (brain implant)

• Implanted in the Primary Motor Cortex (M1), hand/arm area

• Purpose was to allow severely disabled individuals to communicate, use their PC’s, and even control a wheelchair by decoding their thoughts

33

Elbow Extension

BackgroundBackground

• Battelle developed user training methods and Neural Decoding Algorithms

• Processed signals from the electrode array (96 channels)

• Goal: isolate and predict arm/hand movements the patient is thinking about

Time Domain Signals from Electrode Array

Neural Decoding

Shoulder Abduction

Wrist Flexion

44

PatientsPatients

• 54 year old female stroke victim (“SCI 3”), quadriplegic and unable to speak

• 36 year old male, diagnosed with ALS, quadriplegic and unable to speak

• Conducted an intended motion survey to look at neural modulation (change in neuronal firing patterns) for various visually cued movements

– Shoulder Flexion and Extension

– Shoulder Abduction and Adduction

– Elbow Flexion and Extension

– Wrist Flexion and Hyperextension

– Wrist Radial and Ulnar Deviation

– Wrist Pronation and Supination

– Hand Open and Close

55

Initial Motion Survey with SCI3Initial Motion Survey with SCI3

OMAS, SCI3, Tasks 1A and 2A

0

2

4

6

8

10

12

14

16

18

ShoulderFlexion

ShoulderExtension

Shoulder Add. Shoulder Abd. WristExtension

Wrist Flexion Wrist RadialDev. with Palm

Down

Wrist UlnarDev. with Palm

Down

Num

. of U

nits

Iden

tifie

d

Series1

Zero Units (Neurons) Exhibited Differing Firing Patterns

-2000

200 -100-50050100150200

-150

-100

-50

0

50

100

150

Flexion/extension mixed (dark/med blue)

(Axes are unit-less and represent the first three principal components from a principal components / eigenvector analysis.)

66

Thought-Driven CursorThought-Driven Cursor

• Previous decoding methods (from non-human primate work)– Position based (assumed neural patterns correlated to position in space)– Early demonstration in a human subject:

BG SCI P003 - Center Out - 2006.01.23.wmv

• Marginally stable results - proprioceptive feedback is disrupted in a disabled user

• Introduced user training methods - used motion demonstration (mirror neuron activation) and ‘coaching’ periods

• Moved away from position-based decoding and utilized model-based approaches along with support vector machine methods

• Results for Battelle decoding algorithm - stabilized cursor motion by isolating 6 intended motions: rest, four wrist motions, and close hand:

– SCI P003 8-23-06 Great - excellent motion and diagonal.mov

77

Increasing Degrees-of-Freedom -12 Motions Increasing Degrees-of-Freedom -12 Motions (6 DOF) Recognized by Algorithm(6 DOF) Recognized by Algorithm

1 2 3 4 5 6 7 8 9 10 11 12 130

20

40

60

80

100

120

Imagined Movement

Detection Rate for State vs. All Others: 2006_8_23_16_52.mat

Det

ectio

n R

ate

(%)

97.3

90.8 97

.6

85.4 89.9

79.8

98.4

92.9

92.0

94.0

94.5

80.5

restShoulder Up

Shoulder Down

Shoulder Left

Shoulder RightElbow - Curl

Eblow - Straighten

Wrist UpWrist Down

Wrist Left

Wrist Right

Hand - CloseHand - Open

1 2 3 4 5 6 7 8 9 10 11 12 130

1

2

3

4

5

Imagined Movement

False Positive Rate for State vs. All Others:

Fal

se P

ositi

ve R

ate

(%)

1.0

1.6

0.9

1.6

1.0 1.2

0.8

1.3

1.3

1.1

0.7 0.8

restShoulder Up

Shoulder Down

Shoulder Left

Shoulder RightElbow - Curl

Eblow - Straighten

Wrist UpWrist Down

Wrist Left

Wrist Right

Hand - CloseHand - Open

88

Adaptive AlgorithmAdaptive Algorithm

• Subject ALS1 exhibited decreased and

inconsistent performance

• Incorporated performance feedback during

training (“dots/masses on springs”)

• Adapted decoding algorithm after each training

block

• Demonstration: ALS1_4DOF_matlab_screen.wmv

99

Communicating to OthersCommunicating to Others

• Locked-in syndrome

• 5 imagined movements to move and ‘click’

cursor over virtual keyboard: Aug_Communications_Wonderful_high_quality.wmv

1010

Wheelchair ControlWheelchair Control

• Mobility extremely important and can be

difficult/impossible for certain patients with existing

interfaces

• As aired on 60 Minutes (Nov. 2, 2008), Cathy

Hutchinson became the world’s first person to control

a wheelchair with a brain implant (using the Battelle

training and decoding methods)– Stroke Patient - Controlling Wheelchair.wmv

1111

AcknowledgementsAcknowledgements

• Cyberkinetics, Inc.

• Dr. John Donoghue

• Dr. Leigh Hochberg

• Mike Fritz

• Almut Branner

• Abe Caplan

• Ammar Shaikhouni

• Oystein Johnsen (Eagle)