10 results found
Clarke AK, Atashzar SF, Vecchio AD, et al., 2021, Deep learning for robust decomposition of high-density surface EMG signals, IEEE Transactions on Biomedical Engineering, Vol: 68, Pages: 526-534, ISSN: 0018-9294
Blind source separation (BSS) algorithms, such as gradient convolution kernel compensation (gCKC), can efficiently and accurately decompose high-density surface electromyography (HD-sEMG) signals into constituent motor unit (MU) action potential trains. Once the separation matrix is blindly estimated on a signal interval, it is also possible to apply the same matrix to subsequent signal segments. Nonetheless, the trained separation matrices are sub-optimal in noisy conditions and require that incoming data undergo computationally expensive whitening. One unexplored alternative is to instead use the paired HD-sEMG signal and BSS output to train a model to predict MU activations within a supervised learning framework. A gated recurrent unit (GRU) network was trained to decompose both simulated and experimental unwhitened HD-sEMG signal using the output of the gCKC algorithm. The results on the experimental data were validated by comparison with the decomposition of concurrently recorded intramuscular EMG signals. The GRU network outperformed gCKC at low signal-to-noise ratios, proving superior performance in generalising to new data. Using 12 seconds of experimental data per recording, the GRU performed similarly to gCKC, at rates of agreement of 92.5% (84.5%-97.5%) and 94.9% (88.8%-100.0%) respectively for GRU and gCKC against matched intramuscular sources.
Braecklein M, Ibanez J, Barsakcioglu DY, et al., 2021, Towards human motor augmentation by voluntary decoupling beta activity in the neural drive to muscle and force production, JOURNAL OF NEURAL ENGINEERING, Vol: 18, ISSN: 1741-2560
Barsakcioglu DY, Bracklein M, Holobar A, et al., 2020, Control of spinal motoneurons by feedback from a non-invasive real-time interface, IEEE Transactions on Biomedical Engineering, Vol: 68, Pages: 926-935, ISSN: 0018-9294
Interfacing with human neural cells during natural tasks provides the means for investigating the working principles of the central nervous system and for developing human-machine interaction technologies. Here we present a computationally efficient non-invasive, real-time interface based on the decoding of the activity of spinal motoneurons from wearable high-density electromyogram (EMG) sensors. We validate this interface by comparing its decoding results with those obtained with invasive EMG sensors and offline decoding, as reference. Moreover, we test the interface in a series of studies involving real-time feedback on the behavior of a relatively large number of decoded motoneurons. The results on accuracy, intuitiveness, and stability of control demonstrate the possibility of establishing a direct non-invasive interface with the human spinal cord without the need for extensive training. Moreover, in a control task, we show that the accuracy in control of the proposed neural interface may approach that of the natural control of force. These results are the first that demonstrate the feasibility and validity of a non-invasive direct neural interface with the spinal cord, with wearable systems and matching the neural information flow of natural movements.
Barsakcioglu DY, Farina D, 2018, A real-time surface EMG decomposition system for non-invasive human-machine interfaces, IEEE Biomedical Circuits and Systems Conference (BioCAS), Publisher: IEEE, ISSN: 2163-4025
Real-time surface EMG decomposition, to extract neural activity of spinal motor neurons, provides a non-invasive solution for establishing direct interfaces with the central nervous system. In this paper, we present a real-time EMG decomposition system, validate it through both synthetic and experimental high-density surface EMG (HD-sEMG) data, and demonstrate the system in an upper-limb prosthetic control scenario. The proposed system achieves (in real-time) median decomposition accuracy comparable to offline methods (within 0.5 %) with minimal utilisation of computational resources (x20 faster compared to the literature).
Dávila-Montero S, Barsakcioglu DY, Jackson A, et al., 2017, Real-time clustering algorithm that adapts to dynamic changes in neural recordings, IEEE International Symposium on Circuits and Systems (ISCAS), Publisher: IEEE, Pages: 690-693
This work presents a computationally efficient real-time adaptive clustering algorithm that recognizes and adapts to dynamic changes observed in neural recordings. The algorithm consists of an off-line training phase that determines initial cluster positions, and an on-line operation phase that continuously tracks drifts in clusters and periodically verifies acute changes in cluster composition. Analysis of chronic recordings from non-human primates shows that adaptive clustering achieves an improvement of 14% in classification accuracy and demonstrates an ability to recognize acute changes with 78% accuracy, with up to 29% computational efficiency compared to the state-of-the-art. The presented algorithm is suitable for long-term chronic monitoring of neural activity in various applications such as neuroscience research and control of neural prosthetics and assistive devices.
Barsakcioglu DY, Constandinou TG, 2016, A 32-Channel MCU-Based Feature Extraction and Classification for Scalable on-Node Spike Sorting, IEEE International Symposium on Circuits and Systems (ISCAS), Publisher: IEEE, Pages: 1310-1313
This paper describes a new hardware-efficientmethod and implementation for neural spike sorting basedon selection of a channel-specific near-optimal subset of fea-tures given a larger predefined set. For each channel, real-time classification is achieved using a simple decision matrixthat considers the features that provide the highest separabilitydetermined through off-line training. A 32-channel system for on-line feature extraction and classification has been implementedin an ARM Cortex-M0+ processor. Measured results of thehardware platform consumes 268 W per channel during spikesorting (includes detection). The proposed method provides atleast x10 reduction in computational requirements compared toliterature, while achieving an average classification error of lessthan 10% across wide range of datasets and noise levels.
Barsakcioglu D, Liu Y, Bhunjun P, et al., 2014, An Analogue Front-End Model for Developing Neural Spike Sorting Systems, IEEE Transactions on Biomedical Circuits and Systems, Vol: 8, Pages: 216-227
Navajas J, Barsakcioglu D, Eftekhar A, et al., 2014, Minimum Requirements for Accurate and Efficient Real-Time On-Chip Spike Sorting, Journal of Neuroscience Methods, Pages: 51-64
Barsakcioglu DY, Eftekhar A, Constandinou TG, 2013, Design Optimisation of Front-End Neural Interfaces for Spike Sorting Systems, IEEE International Symposium on Circuits and Systems (ISCAS)
This work investigates the impact of the analoguefront-end design (pre-amplifier, filter and converter) on spike sorting performance in neural interfaces. By examining key design parameters including the signal-to-noise ratio, bandwidth,filter type/order, data converter resolution and sampling rate, their sensitivity to spike sorting accuracy is assessed. This is applied to commonly used spike sorting methods such as template matching, 2nd derivative-features, and principle component analysis. The results reveal a near optimum set of parameters to increase performance given the hardware-constraints. Finally, the relative costs of these design parameters on resource efficiency (silicon area and power requirements) are quantified through reviewing the state-of-the-art.
Paraskevopoulou SE, Barsakcioglu D, Saberi M, et al., 2013, Feature Extraction using First and Second Derivative Extrema (FSDE), for Real-time and Hardware-Efficient Spike Sorting, Journal of Neuroscience Methods, Vol: 215, Pages: 29-37, ISSN: 0165-0270
Next generation neural interfaces aspire to achieve real-time multi-channel systems by integrating spike sorting on chip to overcome limitations in communication channel capacity. The feasibility of this approach relies on developing highly-efficient algorithms for feature extraction and clustering with the potential of low-power hardware implementation. We are proposing a feature extraction method, not requiring any calibration, based on first and second derivative features of the spike waveform. The accuracy and computational complexity of the proposed method are quantified and compared against commonly used feature extraction methods, through simulation across four datasets (with different single units) at multiple noise levels (ranging from 5 to 20% of the signal amplitude). The average classification error is shown to be below 7% with a computational complexity of 2N-3, where N is the number of sample points of each spike. Overall, this method presents a good trade-off between accuracy and computational complexity and is thus particularly well-suited for hardware-efficient implementation.
This data is extracted from the Web of Science and reproduced under a licence from Thomson Reuters. You may not copy or re-distribute this data in whole or in part without the written consent of the Science business of Thomson Reuters.