A Self-learning and Adaptive Control Scheme for Phantom Prosthesis Control Using Combined Neuromuscular and Brain-Wave Bio-Signals

The control scheme in a myoelectric prosthesis includes a pattern recognition section whose task is to decode an input signal, produce a respective actuation signal and drive the motors in the prosthesis limb towards the completion of the user’s intended gesture motion. The pattern recognition architecture works with a classifier which is typically trained and calibrated offline with a supervised learning framework. This method involves the training of classifiers which form part of the pattern recognition scheme, but also induces additional and often undesired lead time in the prosthesis design phase. In this study, a three-phase identification framework is formulated to design a control architecture capable of self-learning patterns from bio-signal inputs from electromyography (neuromuscular) and electroencephalography (brain wave) biosensors, for a transhumeral amputee case study. The results show that the designed self-learning framework can help reduce lead time in prosthesis control interface customisation, and can also be extended as an adaptive control scheme to minimise the performance degradation of the prosthesis controller.

-Intent decoders/Classifiers are trained via the 'Supervised Learning' frameworkthus, expert in loop required & lag time induced from training process -Classifier degradation due to uncertainties i.e. electrode shift, physiological changes in stump etc  Proposed Solution -Design of Self Learning and Adaptive Controllers with 'Unsupervised Learning' framework which can help further enhance intuitiveness of prosthesis control and increase overall autonomy 5 https://medium.com/the-21st-century/machinelearning-a-strategy-to-learn-and-understand-chapter-3-9daaad4afc55  Electromyography (EMG) Represent superimposed electrical manifestations of action potentials from motor neurons, and can be mathematically modelled using dipole theory as a continuous extracellular action potential from a multiple source as seen in equation 1: Where is the time varying extracellular potential, is the conductivity of the extracellular medium, is the intracellular conductivity, is the radius of the fiber, is time, is the distance of the source excitation to the recording sensor, is a point in space within the fiber element, − is the length of the anatomical fiber and is the dipole strength at a point along the fiber axis.

 EMG Sensors
The EMG instrumentation used for data acquisition by Li et al [1] was the Refa 128 high-density electrodes by TMS International BV, Netherlands, with 32 electrodes [2]. The acquisition electronics comprised of a bandpass filter in the 10-500Hz frequency range, 24bit resolution and a sample rate of 1024Hz.
 Electroencephalography (EEG) EEG signals occur from the synchronous neuronal firing of billions of pyramid-like cells within the skull of a human being. Using a combination of dipole theory, and assuming the forward EEG problem, a measured potential of an EEG signal can be formulated as follows : Where s is the dipole source located within proximity of sphere of radius r s of moment q, boundary sphere r L , anisotropic conductivity within boundary sub-domain of L, is the EEG measurement for nth element in the infinite set, ∝ is the angle between the point S and measurement point x, is the angle between two planar vectors pairs of S & q and S & x, and 1 represent the Legendre polynomial coefficient of the series.

 EEG Sensors
The 64 sensors EEG channel EasyCap, Herrsching, Germany, with the Al-AgCl electrodes and Neuroscan system version 4.3 was used. The EEG signals were band passed filters in the region of 0.05-100Hz with a sample rate of 1024Hz.

7
(2) Data Collection  Simultaneous acquisition of EMG and EEG signals  The Hand Open and Hand Close Gestures were used for the work done as part of this paper and represent key hand gestures in prosthesis control 8 Proposed Self-Learning Architecture  Assuming the acquisition of a bio-signal, the Self-Learning architecture comprising of an electrode selection process followed by a 3-phase self learning process as seen below: -Classifier Re-calibration to adapt to dynamic changes in the signal acquisition chain, which ultimately causes classifier degradation i.e. electrode shifts and physiological changes in stump -The Self-learning process for classifier recalibration -thus a form of Adaptive Control, can be initiated in either of two ways: *As an interrupt following a series of misclassified motion intents *As an interval based re-calibration prompt 16 https://www.embs.org/tbme/articles/limbposition-tolerant-pattern-recognitionmyoelectric-prosthesis-control-adaptivesparse-representations-extreme-learning/

Conclusion
-A 3-phase Self Learning Controller framework has been proposed to help reduce lag-time in the prosthesis controller customization -The Self Learning Control scheme consists of Feature Extraction Stage, Dimensionality Reduction and Unsupervised Iterative Clustering -The control architecture can also be extended towards an adaptive framework to minimize classifier degradation due to drifts and uncertainties