Forskning

We investigated the feasibility of creating a clinically viable, self-contained prosthetic system that allowed for direct skeletal attachment, neural control, and direct neural sensory feedback via neurostimulation.

Osseoperception is the sensation arising from the mechanical stimulation of a bone-anchored prosthesis. Here we show that not only touch, but also hearing is involved in this phenomenon. Using mechanical vibrations ranging from 0.1 to 6 kHz, we performed four psychophysical measures (perception threshold, sensation discrimination, frequency discrimination and reaction time) on 12 upper and lower limb amputees and found that subjects: consistently reported perceiving a sound when the stimulus was delivered at frequencies equal to or above 400 Hz; were able to discriminate frequency differences between stimuli delivered at high stimulation frequencies (~1500 Hz); improved their reaction time for bimodal stimuli (i.e. when both vibration and sound were perceived). Our results demonstrate that osseoperception is a multisensory perception, which can explain the improved environment perception of bone-anchored prosthesis users. This phenomenon might be exploited in novel prosthetic devices to enhance their control, thus ultimately improving the amputees’ quality of life.

Decoding the firing of individual spinal motor neurons enables the offline control of prosthetic limbs.

Myoelectric pattern recognition (MPR) can be used for intuitive control of virtual and robotic effectors in clinical applications such as prosthetic limbs and the treatment of phantom limb pain. The conventional approach is to feed classifiers with descriptive electromyographic (EMG) features that represent the aimed movements. The complexity and consequently classification accuracy of MPR is highly affected by the separability of such features. In this study, classification complexity estimating algorithms were investigated as a potential tool to estimate MPR performance. An early prediction of MPR accuracy could inform the user of faulty data acquisition, as well as suggest the repetition or elimination of detrimental movements in the repository of classes. Two such algorithms, Nearest Neighbor Separability (NNS) and Separability Index (SI), were found to be highly correlated with classification accuracy in three commonly used classifiers for MPR (Linear Discriminant Analysis, Multi-Layer Perceptron, and Support Vector Machine). These Classification Complexity Estimating Algorithms (CCEAs) were implemented in the open source software BioPatRec and are available freely online. This work deepens the understanding of the complexity of MPR for the prediction of motor volition.

Phantom limb pain is a condition where pain is felt in a missing limb, for example after amputation. In this study patients received a treatment where machine learning and augmented reality was used to visualize and improve their ability to move their phantom limb. Patients showed significant improvements of their pain, suggesting potential value in phantom motor execution as a treatment for phantom limb pain.

Despite the technological progress in robotics, mechatronic limbs still lack functionality, reliability and comfort. This work builds upon osseointegration and implanted neuromuscular interfaces for the realization of an embedded system for artificial upper limb control. The controller allows for bioelectric signals acquisition, processing, decoding and prosthetic control. It includes a neurostimulator to provide direct neural feedback aimed for restoration of tactile sensations. It was designed to be reliably used in activities of daily living, as well as a research platform to monitor prosthesis usage and training, machine learning based control techniques and neural stimulation paradigms. The system has passed bench tests and its functionality was proven in a first pilot patient.

To measure electric signals from the muscles electrodes commonly made of metal containing silver are placed on the skin above the muscle, but electrodes made out textiles might be better for long-term use. To ensure that the signal quality is still good we compared textile electrodes with common metal electrodes. We found that the textile electrodes provided as good signals as the metal electrodes, meaning that textile electrodes can be an alternative to metal electrodes

Bioelectric potentials provide an intuitive source of control in human-machine interfaces. In this work an open source low-cost system for bioelectric signals acquisition and processing was developed. A single module can acquire up to 8 differential or single-ended channels, and several modules can be daisy-chained together. Opto-isolated USB communication was included in the design to interface safely with a personal computer. Embedded digital processing was used for float conversion and filtering. The high-level software was implemented as a complementary part of BioPatRec, an existing open source project. Source files for the PCB, firmware, and high-level software are available online (GitHub: ADS_BP). This integration provides a low-cost, open source and complete system for research on intuitive myoelectric control.

Muscular activity plays an important role in clinical practice and rehabilitation research. Researchers use features of myoelectric signals to decode movements. In this study, a feature called “cardinality” was found to consistently outperform other features, including those that are more complicated and computationally expensive. This technology is useful for prosthesis control and the rehabilitation of patients with motor impairments where myoelectric signals are viable.

Compact and low-noise Analog Front-Ends (AFEs) are becoming increasingly important for the acquisition of bioelectric signals in portable system. In this work, we compare two popular AFEs available on the market, namely the ADS1299 (Texas Instruments) and the RHA2216 (Intan Technologies). This work develops towards the identification of suitable acquisition modules to design an affordable, reliable and portable device for electromyography (EMG) acquisition and prosthetic control. Device features such as Common Mode Rejection (CMR), Input Referred Noise (IRN) and Signal to Noise Ratio (SNR) were evaluated, as well as the resulting accuracy in myoelectric pattern recognition (MPR) for the decoding of motion intention. Results reported better noise performances and higher MPR accuracy for the ADS1299 and similar SNR values for both devices.