Table 2 Real-time performance comparison between ReFIT neural network (RN), neural network (NN), and ReFIT Kalman filter (RK) decoders

From: Real-time brain-machine interface in non-human primates achieves high-velocity prosthetic finger movements using a shallow feedforward neural network decoder

 

Monkey N

Monkey W

 

RK

NN

RN

RK

NN

RN

NN vs. RK

   Throughput (bps)

1.70 ± 0.04

2.15 ± 0.05

 

0.84 ± 0.04

1.23 ± 0.09

 

   Acquisition time (ms)

1550 ± 40

1240 ± 40

 

3310 ± 130

2680 ± 160

 

   Time to target (ms)

1110 ± 30

950 ± 20

 

2410 ± 110

1680 ± 110

 

   Dwell time (ms)

440 ± 30

290 ± 20

 

790 ± 100

1000 ± 120

 

   Successful (total) trials

540 (543)

536 (537)

 

113 (279)

107 (133)

 

Number of Test days

2

 

1

 

Mos. post-implantation

13

 

3

 

RN vs. NN

   Throughput (bps)

 

1.51 ± 0.04

2.15 ± 0.05

 

1.20 ± 0.06

1.43 ± 0.05

   Acquisition time (ms)

 

1880 ± 50

1320 ± 30

 

2610 ± 120

2220 ± 70

   Time to target (ms)

 

1230 ± 30

840 ± 20

 

1740 ± 80

1430 ± 50

   Dwell time (ms)

 

650 ± 30

480 ± 30

 

860 ± 90

780 ± 60

   Successful (total) trials

 

616 (629)

772 (772)

 

185 (237)

441 (483)

Number of Test days

 

2

 

1

Mos. post-implantation

 

19

 

3

RN vs. RK

   Throughput (bps)

1.41 ± 0.03

 

2.29 ± 0.05

   

   Acquisition time (ms)

1940 ± 50

 

1270 ± 30

   

   Time to target (ms)

1330 ± 30

 

790 ± 10

   

   Dwell time (ms)

610 ± 40

 

470 ± 30

   

   Successful (total) trials

559 (614)

 

737 (737)

   

Number of Test days

2

   

Mos. post-implantation

19

   
  1. Of note, there was a gap of 6 months in data collection between NN vs. RK and other tests with Monkey N because of a lab shutdown as a result of the COVID-19 pandemic.
  2. bps bits per second.
Back to article page