Alexander Boschmann

Reducing the limb position effect in pattern recognition based myoelectric control using a high density electrode array

Pattern recognition based myoelectric control schemes are an active field of research. However, there are numerous disparities between current research findings and actual clinical results. In literature, electromyographic signals are usually recorded in a fixed position and used for both, training and testing of the classifier. This supports the test subject in performing repeatable contractions throughout the trials of the experiment and generally results in a high classification accuracy. In clinical testing however, subjects have to perform various activities of daily living, causing the limb to move in different positions. Recent studies have shown that these variations in limb positions significantly decrease robustness and usability of myoelectric control systems. This so-called limb position effect has been previously studied but remains an unsolved problem. In this study we investigate if increasing the number of electrode channels and recording locations can improve the degraded classification accuracy caused by the limb position effect. In our experiment we use a 96 channel high density electrode array to distinguish 11 different hand and wrist movements recorded in three different limb positions. Our results show that training in multiple positions in combination with an increasing number of channels helps reducing the limb position effect.

Reducing classification accuracy degradation of pattern recognition based myoelectric control caused by electrode shift using a high density electrode array

The robustness and usability of pattern recognition based myoelectric control systems degrade significantly if the sensors are displaced during usage. This effect inevitably occurs during donning, doffing or using an upper-limb prosthesis over a longer period of time. Electrode shift has been previously studied but remains an unsolved problem. In this study we investigate if increasing the number of electrode channels and recording locations can improve the degraded classification accuracy caused by electrode shift. In our experiment we use a 96 channel high density electrode array to distinguish 11 different hand and wrist movements. Our results show that for electrode shifts up to 1 cm an array of about 32 sensors in combination with state-of-the-art pattern recognition algorithms is sufficient to compensate the electrode displacement effect.

FPGA-based acceleration of high density myoelectric signal processing

In recent years, advances in electromyographic (EMG) sensor technology and machine learning algorithms have led to an increased research effort into high density EMG based pattern recognition methods for prosthesis control. With the goal set on an autonomous multi-movement prosthesis that is capable of performing training and classification of an amputees EMG signals, the focus of this paper lies in the acceleration of the embedded signal processing chain. Using the Xilinx Zynq as a low-cost off-the-shelf reconfigurable processing platform, we present a solution that is able to compute prosthesis control signals from multi-channel EMG input with up to 256 channels with a maximum processing delay of less than a single millisecond. While the presented system is able to perform training as well as classification, most of our efforts were focused on the acceleration of the feature extraction units, achieving a speed-up of 6.7 for feature extraction alone, and 4.8 for the total processing time as compared to a software only solution.

A novel immersive augmented reality system for prosthesis training and assessment

In recent years, the field of prosthetics developed immensely, along with a variety of control methods and computer interfaces for prosthetic training. In this work, we present an architecture for an augmented reality training system enabling the user to control a virtual prosthetic hand displayed as an extension of the residual limb using EMG pattern recognition in a stereoscopic augmented reality scene. Validated in online experiments with four able-bodied subjects, the novel system provided a more realistic experience compared to the classic 2D implementation and resulted in improved subject performance.

Towards robust HD EMG pattern recognition: reducing electrode displacement effect using structural similarity.

Even small changes of electrode recording sites after training a classifier heavily influence robustness and usability of traditional pattern recognition-based myoelectric control schemes. This effect occurs during donning and doffing of the prosthesis or when changing the arm position and generally leads to a significant decrease of classification accuracy. On the other hand, image representations taken from high density electromyographic (EMG) signals offer high spatial resolution and only seem to change slightly during electrode shift, preserving most structural information. In this paper, we present a simple one-against-one nearest neighbor classifier based on the Structural Similarity Index (SSIM). SSIM quantifies visual similarity of two images based on decomposition into three components: luminance, contrast and structure. Our experimental results indicate that an SSIM-based classifier can outperform an LDA-based classifier using structural information taken from high density EMG signals during simulated electrode shift.

Improving transient state myoelectric signal recognition in hand movement classification using gyroscopes

Pattern recognition of myoelectric signals in upper-limb prosthesis control has been subject to intense research for several years. However, few systems have yet been successfully clinically implemented. One possible explanation for this discrepancy is that published reports mostly focus on classification accuracy of myoelectric signals recorded under laboratory conditions as the metric for the systems performance. These data are usually acquired only during the static state of the contraction in a fixed seated position. This supports the test subject in performing repeatable contractions throughout the experiment and generally results in an unrealistically high classification accuracy. In clinical testing however, subjects have to perform various activities of daily living, causing the limb to move in different positions. These variations in limb positions can significantly decrease robustness and usability of myoelectric control systems. Recent reports have shown that the so-called limb position effect can be resolved for the static state of the signal by adding accelerometer data to the feature vector. Including data from the transient state of the signals for classifier training generally significantly increases the classification error so it is mostly not considered in published reports. In this paper, we investigate the classification accuracy of transient EMG data, taking into account the limb position effect. We demonstrate that a classifier trained with features from EMG, accelerometer and gyroscope outperforms classifiers using only EMG or EMG and accelerometer data when classifying transient EMG data.

Zynq-based acceleration of robust high density myoelectric signal processing

Abstract Advances in electromyographic (EMG) sensor technology and machine learning algorithms have led to an increased research effort into high density EMG-based pattern recognition methods for prosthesis control. With the goal set on an autonomous multi-movement prosthesis capable of performing training and classification of an amputee’s EMG signals, the focus of this paper lies in the acceleration of the embedded signal processing chain. We present two Xilinx Zynq-based architectures for accelerating two inherently different high density EMG-based control algorithms. The first hardware accelerated design achieves speed-ups of up to 4.8 over the software-only solution, allowing for a processing delay lower than the sample period of 1 ms. The second system achieved a speed-up of 5.5 over the software-only version and operates at a still satisfactory low processing delay of up to 15 ms while providing a higher reliability and robustness against electrode shift and noisy channels.

A Zynq-based dynamically reconfigurable high density myoelectric prosthesis controller

The combination of high-density electromyographic (HD EMG) sensor technology and modern machine learning algorithms allows for intuitive and robust prosthesis control of multiple degrees of freedom. However, HD EMG real-time processing poses a challenge for common microprocessors in an embedded system. With the goal set on an autonomous prosthesis capable of performing training and classification of an amputees HD EMG signals, the focus of this paper lies in the acceleration of the computationally expensive parts of the embedded signal processing chain: the feature extraction and classification. Using the Xilinx Zynq as a low-cost off-the-shelf system, we present a solution capable of processing 192 HD EMG channels with controller delays below 120 milliseconds, suitable for highly responsive real-world prosthesis control, achieving speed-ups up to 2.8 as compared to a software-only solution. Using dynamic FPGA reconfiguration, the system is able to trade off increased controller delay against improved classification accuracy when signal quality is decreased due to noisy channels. Offloading feature extraction and classification to the FPGA also reduced the systems power consumption, making it more suitable to be used in a battery-powered setup. The system was validated using real-time experiments with online HD EMG data from an amputee to control a state-of-the-art prosthesis.