Determining the Optimal Window Length for Pattern Recognition-Based Myoelectric Control: Balancing the Competing Effects of Classification Error and Controller Delay

Feinberg School of Medicine, Northwestern University, Chicago, IL 60611 USA.
IEEE transactions on neural systems and rehabilitation engineering: a publication of the IEEE Engineering in Medicine and Biology Society (Impact Factor: 3.19). 12/2010; 19(2):186-92. DOI: 10.1109/TNSRE.2010.2100828
Source: PubMed


Pattern recognition-based control of myoelectric prostheses has shown great promise in research environments, but has not been optimized for use in a clinical setting. To explore the relationship between classification error, controller delay, and real-time controllability, 13 able-bodied subjects were trained to operate a virtual upper-limb prosthesis using pattern recognition of electromyogram (EMG) signals. Classification error and controller delay were varied by training different classifiers with a variety of analysis window lengths ranging from 50 to 550 ms and either two or four EMG input channels. Offline analysis showed that classification error decreased with longer window lengths (p < 0.01 ). Real-time controllability was evaluated with the target achievement control (TAC) test, which prompted users to maneuver the virtual prosthesis into various target postures. The results indicated that user performance improved with lower classification error (p < 0.01 ) and was reduced with longer controller delay (p < 0.01 ), as determined by the window length. Therefore, both of these effects should be considered when choosing a window length; it may be beneficial to increase the window length if this results in a reduced classification error, despite the corresponding increase in controller delay. For the system employed in this study, the optimal window length was found to be between 150 and 250 ms, which is within acceptable controller delays for conventional multistate amplitude controllers.

12 Reads
  • Source
    • "Each gait cycle was further divided in 256 ms data segments for continuous decision making. A recent report of a real time prosthetic control system indicated that the optimal window length was between 150 and 250 ms [29]. We selected 256 ms, a value closer to the higher reported range to avoid large variance (or classification error ) due to shorter window size. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Electromyogram (EMG) signal representation is crucial in classification applications specific to locomotion and transitions. For a given signal, classification can be performed using discriminant functions or if-else rule sets, using learning algorithms derived from training examples. In the present work, a spectrogram based approach was developed to classify (EMG) signals for locomotion mode. Spectrograms for each muscle were calculated and summed to develop a histogram. If-else rules were used to classify test data based on a matching score. Prior knowledge of locomotion type reduced class space to exclusive locomotion modes. The EMG data were collected from seven leg muscles in a sample of able-bodied subjects while walking over ground (W), ascending stairs (SA) and the transition between (W-SA). Three muscles with least discriminating power were removed from the original data set to examine the effect on classification accuracy. Initial classification error was <20% across all modes, using leave one out cross validation. Use of prior knowledge reduced the average classification error to <11%. Removing three EMG channels decreased the classification accuracy by 10.8%, 24.3%, and 8.1% for W, W-SA, and SA respectively, and reduced computation time by 42.8%. This approach may be useful in the control of multi-mode assistive devices. Copyright © 2015 IPEM. Published by Elsevier Ltd. All rights reserved.
    Medical Engineering & Physics 04/2015; 37(5). DOI:10.1016/j.medengphy.2015.03.001 · 1.83 Impact Factor
  • Source
    • "It is possible to separate these methods into two groups. First group is pattern recognition based and second one is non pattern recognition based [1] [2] [3] [4] [5]. "
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper, a technique to classify seven different forearm movements using surface electromyography (sEMG) data which were received from 8 able bodied subjects was proposed. A 2-channel sEMG system was used for data acquisition and recording, then this raw electromyography (EMG) signals were applied to the wavelet denoising. In the next step, time-frequency feature is extracted calculating wavelet packet transform (WPT) coefficients for the offline classification. Feature vector of EMG signals were formed using only node energy of the WPT coefficients. In conclusion, seven forearm movements were separated by Adaptive Neuro-Fuzzy Inference System (ANFIS) classifier and 92% success ratios over 500 samples were obtained.
    2014 IEEE International Symposium on Innovations in Intelligent Systems and Applications (INISTA); 06/2014
  • Source
    • "The LDA classifier was trained to recognize the EMG signal patterns in four commonly used time domain features (Mean Absolute Value, Number of Zero Crossings, Slope Sign Changes, and Waveform Length [28]) and six autoregressive coefficients (the first 6 parameters ai from the autoregressive model of the EMG activity [29]) that were extracted in analysis windows of 250 ms with an overlap of 200 ms. Using analysis windows of up to 250 ms has been shown to produce an acceptable trade-off between controller delay and classification accuracy [30]. Proportional control was derived in each window to provide an output speed to the corresponding class [31]. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Background Pattern recognition (PR) based strategies for the control of myoelectric upper limb prostheses are generally evaluated through offline classification accuracy, which is an admittedly useful metric, but insufficient to discuss functional performance in real time. Existing functional tests are extensive to set up and most fail to provide a challenging, objective framework to assess the strategy performance in real time. Methods Nine able-bodied and two amputee subjects gave informed consent and participated in the local Institutional Review Board approved study. We designed a two-dimensional target acquisition task, based on the principles of Fitts’ law for human motor control. Subjects were prompted to steer a cursor from the screen center of into a series of subsequently appearing targets of different difficulties. Three cursor control systems were tested, corresponding to three electromyography-based prosthetic control strategies: 1) amplitude-based direct control (the clinical standard of care), 2) sequential PR control, and 3) simultaneous PR control, allowing for a concurrent activation of two degrees of freedom (DOF). We computed throughput (bits/second), path efficiency (%), reaction time (second), and overshoot (%)) and used general linear models to assess significant differences between the strategies for each metric. Results We validated the proposed methodology by achieving very high coefficients of determination for Fitts’ law. Both PR strategies significantly outperformed direct control in two-DOF targets and were more intuitive to operate. In one-DOF targets, the simultaneous approach was the least precise. The direct control was efficient in one-DOF targets but cumbersome to operate in two-DOF targets through a switch-depended sequential cursor control. Conclusions We designed a test, capable of comprehensively describing prosthetic control strategies in real time. When implemented on control subjects, the test was able to capture statistically significant differences (p < 0.05) in control strategies when considering throughputs, path efficiencies and reaction times. Of particular note, we found statistically significant (p < 0.01) improvements in throughputs and path efficiencies with simultaneous PR when compared to direct control or sequential PR. Amputees could readily achieve the task; however a limited number of subjects was tested and a statistical analysis was not performed with that population.
    Journal of NeuroEngineering and Rehabilitation 05/2014; 11(1):91. DOI:10.1186/1743-0003-11-91 · 2.74 Impact Factor
Show more