Across-user adaptation for a powered lower limb prosthesis.

Pattern recognition algorithms have been used to control powered lower limb prostheses because they are capable of identifying the intent of the amputee user and therefore can provide a method for seamlessly transitioning between the different locomotion modes of the prosthesis. However, one major limitation of current algorithms is that they require subject-specific data from the individual user. These data are difficult and time-consuming to collect and consequently these algorithms do not generalize well across users. We have developed an adaptive pattern recognition algorithm that automatically learns new subject-specific data acquired from a novel user during ambulation. We tested this adaptive algorithm with one transfemoral amputee subject walking on a powered knee-ankle prosthesis. Before adaptation, the algorithm, which was initially trained with data from two other transfemoral amputee subjects, made critical errors that prevented continuous ambulation. With adaptation, error rates dropped from 4.21% before adaptation to 1.25% after adaptation, and allowed the novel amputee user to complete all mode transitions. This study demonstrates that adaptation can decrease error rates over time and can allow pattern recognition algorithms to generalize to novel users.

Delaying Ambulation Mode Transition Decisions Improves Accuracy of a Flexible Control System for Powered Knee-Ankle Prosthesis.

Powered lower limb prostheses can assist users in a variety of ambulation modes by providing knee and/or ankle joint power. This study's goal was to develop a flexible control system to allow users to perform a variety of tasks in a natural, accurate, and reliable way. Six transfemoral amputees used a powered knee-ankle prosthesis to ascend/descend a ramp, climb a 3- and 4-step staircase, perform walking and standing transitions to and from the staircase, and ambulate at various speeds. A mode-specific classification architecture was developed to allow seamless transitions at four discrete gait events. Prosthesis mode transitions (i.e., the prosthesis' mechanical response) were delayed by 90 ms. Overall, users were not affected by this small delay. Offline classification results demonstrate significantly reduced error rates with the delayed system compared to the non-delayed system (p < 0.001). The average error rate for all heel contact decisions was 1.65% [0.99%] for the non-delayed system and 0.43% [0.23%] for the delayed system. The average error rate for all toe off decisions was 0.47% [0.16%] for the non-delayed system and 0.13% [0.05%] for the delayed system. The results are encouraging and provide another step towards a clinically viable intent recognition system for a powered knee-ankle prosthesis.

Detection of and Compensation for EMG Disturbances for Powered Lower Limb Prosthesis Control.

Myoelectric pattern recognition algorithms have been proposed for the control of powered lower limb prostheses, but electromyography (EMG) signal disturbances remain an obstacle to clinical implementation. To address this problem, we used a log-likelihood metric to detect simulated EMG disturbances and real disturbances acquired from EMG containing electrode shift. We found that features extracted from disturbed EMG have much lower log likelihoods than those from undisturbed signals and can be detected using a single threshold acquired from the training data. We designed a linear discriminant analysis (LDA) classifier that uses the log likelihood to decide between using a combination of EMG and mechanical sensors and using mechanical sensors only, to predict locomotion modes. When EMG contained disturbances, our classifier detected those disturbances and disregarded EMG data. Our classifier had significantly lower errors than a standard LDA classifier in the presence of EMG disturbances. The log-likelihood classifier had a low false positive threshold, and thus did not perform significantly differently from the standard LDA classifier when EMG did not contain disturbances. The log-likelihood threshold could also be applied to individual EMG channels, enabling specific channels containing EMG disturbances to be appropriately ignored when making locomotion mode predictions.

Delaying ambulation mode transitions in a powered knee-ankle prosthesis.

Powered knee and ankle prostheses have the potential to improve the mobility of individuals with a lower limb amputation. As the number of different ambulation modes the prosthesis can be configured for increases, so too does the challenge of how to best transition the prosthesis between these modes. Pattern recognition systems have been suggested as a means to provide seamless and natural transitions, although error rates need to be reduced for these systems to be clinical viable. Delaying mode transitions by a small window may be one way to reduce error rates and improve reliability. The goal of this study was to develop and test a system for powered lower limb prostheses that introduced a delay between mode transitions. Three transfemoral amputees used a knee-ankle prosthesis to stand, walk on level ground, ascend/descend a ramp, and ascend/descend stairs. On Day 1 mode transitions occurred at a gait event (e.g., heel contact), and on Day 2 mode transitions occurred 90 ms following a gait event. A mode-specific pattern recognition system was trained and tested on each day. The 90 ms transition delay did not negatively affect users' performance ambulating with the prosthesis. Offline classification error results showed that the 90 ms delay reduced overall classification errors from 1.30% [0.29%], mean [SD], for the non-delayed system to 0.42% [0.22%] for the delayed system. These results demonstrate that delaying mode transitions by a small window of time can reduce overall errors, which moves these systems one step closer to clinical viability.

Preliminary results for an adaptive pattern recognition system for novel users using a powered lower limb prosthesis.

Powered prosthetic legs are capable of improving the gait of lower limb amputees. Pattern recognition systems for these devices allow amputees to transition between different locomotion modes in a way that is seamless and transparent to the user. However, the potential of these systems is diminished because they require large amounts of training data that is burdensome to collect. To reduce the effort required to acquire these data, we developed an adaptive pattern recognition system that automatically learns from subject-specific data as the user is ambulating. We tested our proposed system with two able-bodied subjects ambulating with a powered knee and ankle prosthesis. Each subject initially ambulated with a pattern recognition system that was not trained with any data from that subject (making each subject a novel user). Initially, the pattern recognition system made frequent errors. With the adaptive algorithm, the error rate decreased over time as more subject-specific data were incorporated. When compared to a non-adaptive system, the adaptive system reduced the number of errors by 32.9% [8.6%], mean [standard deviation]. This study demonstrates the potential improvements of an adaptive pattern recognition system over non-adaptive systems presented in prior research.

User intent prediction with a scaled conjugate gradient trained artificial neural network for lower limb amputees using a powered prosthesis.

Powered lower limb prostheses have the ability to provide greater mobility for amputee patients. Such prostheses often have pre-programmed modes which can allow activities such as climbing stairs and descending ramps, something which many amputees struggle with when using non-powered limbs. Previous literature has shown how pattern classification can allow seamless transitions between modes with a high accuracy and without any user interaction. Although accurate, training and testing each subject with their own dependent data is time consuming. By using subject independent datasets, whereby a unique subject is tested against a pooled dataset of other subjects, we believe subject training time can be reduced while still achieving an accurate classification. We present here an intent recognition system using an artificial neural network (ANN) with a scaled conjugate gradient learning algorithm to classify gait intention with user-dependent and independent datasets for six unilateral lower limb amputees. We compare these results against a linear discriminant analysis (LDA) classifier. The ANN was found to have significantly lower classification error (P<;0.05) than LDA with all user-dependent step-types, as well as transitional steps for user-independent datasets. Both types of classifiers are capable of making fast decisions; 1.29 and 2.83 ms for the LDA and ANN respectively. These results suggest that ANNs can provide suitable and accurate offline classification in prosthesis gait prediction.

A strategy for labeling data for the neural adaptation of a powered lower limb prosthesis.

Pattern recognition algorithms that use EMG signals have been proposed to help control powered lower limb prostheses. These algorithms do not automatically compensate for disturbances in EMG signals, resulting in deterioration of algorithm accuracies. Supervised adaptive pattern recognition algorithms can solve this problem, but require correct labeling of new data. Information from embedded mechanical sensors can be compared to the characteristic gait profiles of the different modes to identify the mode of the user's most recent stride and provide a label for new data. The purpose of this study was to develop a gait pattern estimator (GPE) that could automatically make such a comparison. The GPE output was used to supervise an adaptive EMG-based pattern recognition algorithm. Our results indicate that using GPE-based adaptation helped prevent classification errors that would otherwise occur between experimental sessions. The GPE could accurately label new data with a low error rate of approx. 2%. The low error rate of the GPE was reflected in the accuracy of an adapted pattern recognition algorithm. The error rate of the adapted algorithm that was supervised by the GPE was not significantly different from one that used perfect supervision.