Understanding Limb Position and External Load Effects on Real-Time Pattern Recognition Control in Amputees

Teh, Yuni; Hargrove, Levi J.

doi:10.1109/tnsre.2020.2991643

Cited by 30 publications

(25 citation statements)

References 42 publications

(52 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For example, performing dynamic arm movements during training data collection instead of maintaining a static position signi cantly improves LDA classi cation performance across different limb positions [33,34]. By increasing training data variability, classi ers are encouraged learn discriminative features that are consistent across sources of variance, thus preventing over tting.…”

Section: Introductionmentioning

confidence: 99%

Data Augmentation and Deep Learning Techniques to Improve Interface Noise Tolerance of Myoelectric Pattern Recognition Controllers

Teh¹,

Hargrove

2022

Preprint

Self Cite

View full text Add to dashboard Cite

Background: Clinically available myoelectric PR controllers deteriorate under conditions that generate interface noise, such as electrode liftoff or wire failure. Previous solutions relied on additional processing steps like signal denoising and controller adaptation to mitigate these negative effects. However, there are no clinically practical controllers that are inherently robust to interface noise. This paper investigated the use of data-driven methods to build clinically practical, noise tolerant PR control strategies for transradial prostheses.Methods: We developed a data augmentation protocol to increase training data variability and two deep neural networks (MLP and CNN) to compress redundancies across a six-channel EMG array to a latent space that is robust to noise. Using data from fourteen intact limb subjects and seven below-elbow amputee subjects, we evaluated five control strategies: an LDA classifier (LDA), an existing adaptive LDA classifier that recalibrates after discarding noisy EMG channels (LDA-), an LDA classifier trained with an augmented data set (LDA+), an LDA classifier trained with MLP-aligned latent variables of an augmented data set (MLP-LDA), and an LDA classifier trained with CNN-aligned latent variables of an augmented data set (CNN-LDA). We computed the classification accuracies on clean signals and signals that had interface noise in up to four channels.Results: Compared to the baseline LDA method, LDA+ and MLP-LDA significantly increased classification accuracy on noisy data but decreased accuracy on clean data. On the other hand, LDA- and CNN-LDA improved noise tolerance while maintaining high performance with clean data. Conclusion: The CNN-LDA control strategy is robust to multichannel interface noise and can improve the reliability and usability of PR-based upper limb prostheses.

show abstract

Section: Introductionmentioning

confidence: 99%

Data Augmentation and Deep Learning Techniques to Improve Interface Noise Tolerance of Myoelectric Pattern Recognition Controllers

Teh¹,

Hargrove

2022

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Beyond the algorithm itself, the data used to train the algorithm affects prosthesis control. Training an algorithm in multiple arm positions can improve prosthesis control whether the position data is used to train the algorithm [32,44] or not [45], although incorporating positional data may not benefit linear algorithms [46]. For continuous controllers without a discrete set of classes, using training data with simultaneous movements of more than one degree-of-freedom may expose nonlinearities in EMG data and preferentially benefit nonlinear algorithms with greater learning capacities.…”

Section: Introductionmentioning

confidence: 99%

Activities of daily living with bionic arm improved by combination training and latching filter in prosthesis control comparison

Paskett

Brinton

Hansen

et al. 2021

J NeuroEngineering Rehabil

View full text Add to dashboard Cite

Background Advanced prostheses can restore function and improve quality of life for individuals with amputations. Unfortunately, most commercial control strategies do not fully utilize the rich control information from residual nerves and musculature. Continuous decoders can provide more intuitive prosthesis control using multi-channel neural or electromyographic recordings. Three components influence continuous decoder performance: the data used to train the algorithm, the algorithm, and smoothing filters on the algorithm’s output. Individual groups often focus on a single decoder, so very few studies compare different decoders using otherwise similar experimental conditions. Methods We completed a two-phase, head-to-head comparison of 12 continuous decoders using activities of daily living. In phase one, we compared two training types and a smoothing filter with three algorithms (modified Kalman filter, multi-layer perceptron, and convolutional neural network) in a clothespin relocation task. We compared training types that included only individual digit and wrist movements vs. combination movements (e.g., simultaneous grasp and wrist flexion). We also compared raw vs. nonlinearly smoothed algorithm outputs. In phase two, we compared the three algorithms in fragile egg, zipping, pouring, and folding tasks using the combination training and smoothing found beneficial in phase one. In both phases, we collected objective, performance-based (e.g., success rate), and subjective, user-focused (e.g., preference) measures. Results Phase one showed that combination training improved prosthesis control accuracy and speed, and that the nonlinear smoothing improved accuracy but generally reduced speed. Phase one importantly showed simultaneous movements were used in the task, and that the modified Kalman filter and multi-layer perceptron predicted more simultaneous movements than the convolutional neural network. In phase two, user-focused metrics favored the convolutional neural network and modified Kalman filter, whereas performance-based metrics were generally similar among all algorithms. Conclusions These results confirm that state-of-the-art algorithms, whether linear or nonlinear in nature, functionally benefit from training on more complex data and from output smoothing. These studies will be used to select a decoder for a long-term take-home trial with implanted neuromyoelectric devices. Overall, clinical considerations may favor the mKF as it is similar in performance, faster to train, and computationally less expensive than neural networks.

show abstract

“…Pattern recognition algorithms can also struggle to generalize to new contexts. For example, classifiers often issue incorrect predictions while supporting the prosthesis weight in untrained arm positions ( 22 ). One study found that more robust classifiers can mitigate this issue ( 23 ).…”

Section: Introductionmentioning

confidence: 99%

Surgically Implanted Electrodes Enable Real-Time Finger and Grasp Pattern Recognition for Prosthetic Hands

Vaskov

North

et al. 2020

Preprint

View full text Add to dashboard Cite

Currently available prosthetic hands are capable of actuating anywhere from five to 30 degrees of freedom (DOF). However, grasp control of these devices remains unintuitive and cumbersome. To address this issue, we propose directly extracting finger commands from the neuromuscular system via electrodes implanted in residual innervated muscles and regenerative peripheral nerve interfaces (RPNIs). Two persons with transradial amputations had RPNIs created by suturing autologous free muscle grafts to their transected median, ulnar, and dorsal radial sensory nerves. Bipolar electrodes were surgically implanted into their ulnar and median RPNIs and into their residual innervated muscles. The implanted electrodes recorded local electromyography (EMG) with Signal-to-Noise Ratios ranging from 23 to 350 measured across various movements. In a series of single-day experiments, participants used a high speed pattern recognition system to control a virtual prosthetic hand in real-time. Both participants were able to transition between 10 pseudo-randomly cued individual finger and wrist postures in the virtual environment with an average online accuracy of 86.5% and latency of 255 ms. When the set was reduced to five grasp postures, average metrics improved to 97.9% online accuracy and 135 ms latency. Virtual task performance remained stable across untrained static arm positions while supporting the weight of the prosthesis. Participants also used the high speed classifier to switch between robotic prosthetic grips and complete a functional performance assessment. These results demonstrate that pattern recognition systems can use the high-quality EMG afforded by intramuscular electrodes and RPNIs to provide users with fast and accurate grasp control.

show abstract

Understanding Limb Position and External Load Effects on Real-Time Pattern Recognition Control in Amputees

Cited by 30 publications

References 42 publications

Data Augmentation and Deep Learning Techniques to Improve Interface Noise Tolerance of Myoelectric Pattern Recognition Controllers

Data Augmentation and Deep Learning Techniques to Improve Interface Noise Tolerance of Myoelectric Pattern Recognition Controllers

Activities of daily living with bionic arm improved by combination training and latching filter in prosthesis control comparison

Surgically Implanted Electrodes Enable Real-Time Finger and Grasp Pattern Recognition for Prosthetic Hands

Contact Info

Product

Resources

About