Skip to main content

Table 4 Overview of the evaluation measurements and related results

From: EMG-driven control in lower limb prostheses: a topic-based systematic review

Ref.

Perform-based measurements

Biomechanical measurements

Averaged results

Control delay

Subjects number

Reported limitations

[64]

QA on locomotion performance

NI

NI

50–100 ms

> 1 TFA

NS

[29]

Cadence, Swing and Stance duration

Joint angles and moments, maximum knee flexion

Results in figures only

NS

1 TFA

Sensitive to movement artifacts

[24,25,26,27]

Error events analysis

Joint angle NRME, CC

Max error events amplitude = 42 (11 SD); NRME < 6.56 (1.85 SD)%; CC = 0.59 (0.9 SD)

NS

4 ABS

High maximum error amplitudes

[56]

Joint flexion/extension CA

Joint angle RMSE

CA = 92 (7 SD)%; RMSE = 6.2 (0.71 SD)\(^{\circ }\)

NS

2 TFA, 1 BTFA

NS

[14, 23, 62, 63]

\(t_{STRIDE}\); \(v_{WALKING}\)

Joint angle and joint stiffness; RMSE in joint angle trajectory tracking

\(t_{STRIDE}\) = 1.96s; \(v_{WALKING}\) = 2.9 km/h; able-bodied resemblance joint trajectories (figures only); RMSE statistical different only with haptic feedback and no-visual

1 ms

1 TFA; 2 ABS

Not appropriate swing control; lack of somatosensory feedback; sensitive to movement artifacts and skin perspiration

[57]

Motor task CA, MCT and MCP

NI

CA = 90.7 (5.0 SD)%; MCT = 1.26 (0.1 SD)s; CP = 96.3 (4.3 SD)%

< 700 ms

6 TFA

Sensitive to electrode shifts and impedance; extensive training necessary

[37]

NI

VAF, RMSE joint angle

VAF > 83%; RMSE < 5.4 (1.2 SD)\(^{\circ }\)

− 100 ms

3 TTA

Performance being tested only in constant velocity walking task

[16]

MCT

NI

MCT = 1.9 s

NS

5 ABS

Position controller unsatisfactorily during stance phase

[67, 68]

NP

Joint peak power and work respect to state-base controller

Statistical difference of evaluated parameters only with visual feedback (p-val = 0.02)

33 ms

5 TTA

Short training session, experienced high muscular fatigue

[76]

NI

Joint angle trajectories

Results in figures only

NS

10 ABS

No walking speed adaptation, no real-time

[11]

NI

Joint trajectory r-value and SNR

r-value = 0.64 (0.22SD); SNR = 7.42 (2.88SD)

3.3 ms

6 ABS

Position controller unsatisfactorily for limb dynamics

[39, 40]

Number of falling

Joint angle RMSE, EMG contraction level, mean joint torque during balance task

Falling events and applied torque decrease with training; final RMSE = 0.19 (8.78)\(^{\circ }\); no significant changes in muscle activation with training

10 ms

6 ABS; 6 TTA

Small sample population; study used visual feedback

[98]

NI

QA of EMG signals and joint angle

Figures only

NS

1 TFA

Sensitive to movement artifacts; sensitive to muscular mass changes

[70]

Locomotion classification accuracy

QA of ankle joint position and shank angular orientation

CA = 86.53 (8.5 SD)%; biomechanical measurements (figures only)

NS

> 1 ABS, > 1 TFA

NS

[5]

Stance time; gait symmetry

Toe-off angle; peak torque; joint trajectories

Qualitatively similar to biological ankle trajectories (figures only)

NS

1 BTTA

Asymmetry on knee flexion during late stance

[65, 66, 141, 144]

Locomotion and transitions CA

NI

Locomotion CA = 91.79–100%; transition CA = 100%

12 ms

5 TFA

Sensor fusion and sound leg instrumentation is necessary to increase accuracy

[53]

Locomotion CA

NI

CA \(\approx\) 80–100% (figures only)

NS

5 ABS

NS

[85]

Locomotion CA

NI

CA = 91.46%

NS

100 ms

NS

[15, 110]

Locomotion CA

NI

CA = 91.23%

NS

3 ABS

Tested only healthy subjects

[32, 33, 143]

Locomotion and transitions CA

NI

Locomotion CA \(\approx\) 98%; transitions CA > 99%

< 45.2 ms

4 TFA

Real-time test only on non-powered prosthesis; mechanical sensor feature are necessary

[86, 87]

Locomotion CA

NI

CA = 97.9 (1.39 SD)%

NS

5 ABS, 5 TFA

Only limited number of locomotions; major misclassification during gait transitions

[120]

Joint DoF motion CA

NI

1-DoF CA = 93.3 (0.5 SD)%; 3-DoF CA = 84.4 (0.8 SD)%;

50ms

5 ABS, 12 TTA

Best results only combining both tibia and thigh muscles

[58, 111]

Locomotion CA, NWB CA, falls occurrences

NI

with TMR: locomotion CA = 8.9%, NWB CA = 91.0 (4.7SD)%, falls occurrence = 0%; no-TMR: locomotion CA = 10.2%, NWB CA = 86.8 (3.0SD)%, falls occurrence = 2%

NS

4 TFA, 1 TMR

Control degradation over time due to fatigue, electrode shift and skin perspiration; necessity of mechanical sensors for high accuracy; small number of subjects

[59, 113, 114, 139, 140]

Locomotion and transitional CA, effects on classification errors

NI

Locomotion CA < 99%; transition CA = 87%; classification errors during stairs were more disruptive

< 20 ms

7 TFA

Control degradation over time due to fatigue, electrode shift and skin perspiration; testing is performed in only one sessions

[17]

Questioner on control comfort

Inclination CA

CA > 95%; comfort higher when no classification error (accepted error \(\le 5^\circ\))

NS

2 TFA

Experienced high muscular fatigue; lack of sensory feedback

[106]

NI

QA of EMG signals

Results in figures only

NS

1 ABS

Not tested on amputee; only walking activity control

[54, 55]

Locomotion CA

NI

CA = 99.06 (0.87SD)%

32 ms

5 ABS

Not tested on-line

[6]

NI

Joint angle trajectory frequency content

mean frequency = 5.4 (0.3 SD) Hz, qualitatively similar to biological ankle

NS

1 TTA

Temporal variation of the EMG signals are not accounted for

[61]

Subject qualitative report

QA of knee joint position and torque

Control interface did not feel natural; able-bodied resemblance joint trajectories (figures only)

NS

1 ABS with ABA

Experimental tuning of the parameters is necessary

[135, 136]

QA of joint trajectories with respect to ABS

Joint angle trajectory RMSE

Results in figures only

NS

1 ABS with ABA

Model parameters are tuned manually; control instable with biarticular muscles

[71, 72, 130]

NI

Toe-off angle, joint net work, peak power, joint torque vs angle

Net work performed higher than the biological norms; not substantial difference in joint moments between intrinsic controller and EMG-driven

2 ms

1 BTTA; 3 TTA, 3 ABS

Further studies on the real metabolic cost benefits are required

[19]

NI

Joint torque NRMSD with respect ABS

NRMSD \(\le\) 0.24 (0.11 SD)

< 50 ms

1 ABS

Complex subject-specific calibration; required validation on more subjects and hardware

  1. Fields include: paper reference; performance-based measurements (e.g. cadence, stance/swing time, stumble rates, etc.); biomechanical measurements (e.g. joint trajectory deviations, peak angles, net work, etc.); averaged results (averaged or worst subject-case results); control delay (time required for the generation of the high-level control output from the acquired relevant signals, including processing); subjects (number of subjects); reported limitations (limitations reported by the authors)
  2. NP not present, NI not implemented, NS not specified, QA qualitative analysis, CA classification accuracy, RMSE root mean squared error, NRMSD normalized root mean squared deviation, \(t_{STRIDE}\) time stride, \(v_{WALKING}\) walking self selected velocity, VAF variance accounted for, MCT motion completion time, MCP motion compilation percentage, DoF Degree of Freedom, TFA transfemoral amputee, BTFA bilateral transfemoral amputee, TTA transtibial amputee, BTTA bilateral transtibial amputee, ABS able-bodied subjects, ABA able-body adaptor