Constantinos Gavriel

Wireless kinematic body sensor network for low-cost neurotechnology applications “in-the-wild”

We present an ultra-portable and low-cost body sensor network (BSN), which enables wireless recording of human motor movement kinematics and neurological signals in unconstrained, daily-life environments. This is crucial as activities of daily living (ADL) and thus metrics of everyday movement enable us to diagnose motor and neurological disorders in the patients context, and not artificial laboratory settings. Moreover, ADL kinematics inform us how to control neuroprosthetics and brain-machine interfaces in a natural manner. Our system uses a network of battery-powered embedded micro-controllers, to capture data from motion sensors placed all over the human body and wireless connectivity to stream process data in real time at 100 Hz. Our prototype compares well against two gold-standard measures, a ground-truth motion tracking system and high-end motion capture suit as reference. At 2.5% of the cost, performance in capturing natural joint kinematics are accurate R2 = 0.89 and precise RMSE = 1.19°. The systems low-cost (approximately

A Comparison of Day-Long Recording Stability and Muscle Force Prediction between BSN-Based Mechanomyography and Electromyography

100 per unit), wireless capability, low weight and millimetre-scale size allow subjects to be unconstrained in their actions while having the sensors attached to everyday clothing. These features establish our systems usefulness in clinical studies, risk-group monitoring, neuroscience and neuroprosthetics.

EthoHand: A dexterous robotic hand with ball-joint thumb enables complex in-hand object manipulation

Day-long continuous monitoring requires stable sensors that can minimise the effects of drift and maintain high accuracy and precision over time. We have recently shown that our inertial motion tracking system can capture stable kinematic data, calibrated against ground-truth over a long period of time. However, for many clinical and daily life activities, it is also essential to monitor the muscle-activity. In this study, we evaluate the long-term recording stability of our prototype mechanomyography (MMG) sensors as an extension to our existing ETHO1 body sensor network platform. We attached the MMG sensors along with commercial high-accuracy EMG electrodes on the arm muscles of 5 subjects throughout a working day of 9 hours. The subjects followed their daily routine but they had to perform a multi-level force-matching task through flexion and extension of their arm during four short sessions of the day, as measures of practical signal quality. We designed a force predictor that used either EMG or MMG signals to predict the forces generated by subjects. Our prototype low-cost MMG channels have shown comparable results (RMSE: 23N and R2: 0.91) in predicting the force levels applied when compared against the commercial high-accuracy EMG sensor (RMSE: 19N and R2: 0.95).

Data-efficient hand motor imagery decoding in EEG-BCI by using Morlet wavelets & Common Spatial Pattern algorithms

Our dexterous hand is a fundmanetal human feature that distinguishes us from other animals by enabling us to go beyond grasping to support sophisticated in-hand object manipulation. Our aim was the design of a dexterous anthropomorphic robotic hand that matches the human hands 24 degrees of freedom, under-actuated by seven motors. With the ability to replicate human hand movements in a naturalistic manner including in-hand object manipulation. Therefore, we focused on the development of a novel thumb and palm articulation that would facilitate in-hand object manipulation while avoiding mechanical design complexity. Our key innovation is the use of a tendon-driven ball joint as a basis for an articulated thumb. The design innovation enables our under-actuated hand to perform complex in-hand object manipulation such as passing a ball between the fingers or even writing text messages on a smartphone with the thumbs end-point while holding the phone in the palm of the same hand. We then proceed to compare the dexterity of our novel robotic hand design to other designs in prosthetics, robotics and humans using simulated and physical kinematic data to demonstrate the enhanced dexterity of our novel articulation exceeding previous designs by a factor of two. Our innovative approach achieves naturalistic movement of the human hand, without requiring translation in the hand joints, and enables teleoperation of complex tasks, such as single (robot) handed messaging on a smartphone without the need for haptic feedback. Our simple, under-actuated design outperforms current state-of-the-art prostheses or robotic and prosthetic hands regarding abilities that encompass from grasps to activities of daily living which involve complex in-hand object manipulation.

Smartphone as an ultra-low cost medical tricorder for real-time cardiological measurements via ballistocardiography

EEG-based Brain Computer Interfaces (BCIs) are quite noisy brain signals recorded from the scalp (electroencephalography, EEG) to translate the users intent into action. This is usually achieved by looking at the pattern of brain activity across many trials while the subject is imagining the performance of an instructed action - the process known as motor imagery. Nevertheless, existing motor imagery classification algorithms do not always achieve good performances because of the noisy and non-stationary nature of the EEG signal and inter-subject variability. Thus, current EEG BCI takes a considerable upfront toll on patients, who have to submit to lengthy training sessions before even being able to use the BCI. In this study, we developed a data-efficient classifier for left/right hand motor imagery by combining in our pattern recognition both the oscillation frequency range and the scalp location. We achieve this by using a combination of Morlet wavelet and Common Spatial Pattern theory to deal with nonstationarity and noise. The system achieves an average accuracy of 88% across subjects and was trained by about a dozen training (10-15) examples per class reducing the size of the training pool by up to a 100-fold, making it very data-efficient way for EEG BCI.

Dynamic forward prediction for prosthetic hand control by integration of EMG, MMG and kinematic signals

In this preliminary study, we investigate the potential use of smartphones as portable heart-monitoring devices that can capture and analyse heart activity in real time. We have developed a smartphone application called “Medical Tricorder” that can exploit smartphone;s inertial sensors and when placed on a subject;s chest, it can efficiently capture the motion patterns caused by the mechanical activity of the heart. Using the measured ballistocardiograph signal (BCG), the application can efficiently extract the heart rate in real time while matching the performance of clinical-grade electrocardiographs (ECG). Although the BCG signal can provide much richer information regarding the mechanical aspects of the human heart, we have developed a method of mapping the chest BCG signal into an ECG signal, which can be made directly available to clinicians for diagnostics. Comparing the estimated ECG signal to empirical data from cardiovascular diseases, may allow detection of heart abnormalities at a very early stage without any medical staff involvement. Our method opens up the potential of turning smartphones into portable healthcare systems which can provide patients and general public an easy access to continuous healthcare monitoring. Additionally, given that our solution is mainly software based, it can be deployed on smartphones around the world with minimal costs.

Towards a brain-derived neurofeedback framework for unsupervised personalisation of Brain-Computer Interfaces

We propose a new framework for extracting information from extrinsic muscles in the forearm that will allow a continuous, natural and intuitive control of a neuroprosthetic devices and robotic hands. This is achieved through a continuous mapping between muscle activity and joint angles rather than prior discretisation of hand gestures. We instructed 6 able-bodied subjects, to perform everyday object manipulation tasks. We recorded the Electromyographic (EMG) and Mechanomyographic (MMG) activities of 5 extrinsic muscles of the hand in their forearm, while simultaneously monitoring 11 joints of hand and fingers using a sensorised glove. We used these signals to train a Gaussian Process (GP) and a Vector AutoRegressive Moving Average model with Exogenous inputs (VARMAX) to learn the mapping from current muscle activity and current joint state to predict future hand configurations. We investigated the performances of both models across tasks, subjects and different joints for varying time-lags, finding that both models have good generalisation properties and high correlation even for time-lags in the order of hundreds of milliseconds. Our results suggest that regression is a very appealing tool for natural, intuitive and continuous control of robotic devices, with particular focus on prosthetic replacements where high dexterity is required for complex movements.

Gaussian Process Regression for accurate prediction of prosthetic limb movements from the natural kinematics of intact limbs

Modern Brain Computer Interfaces (BCIs) use EEG signals recorded from the scalp to transduce a users intent into action. However, achieving an optimal control requires a physically and mentally demanding series of long-lasting training sessions based on the use of common neurofeedback. In this study we propose a framework that bypasses the training phase (unsupervised personalisation), where the BCI is automatically detecting whether it is acting according to the users intention or not. We used mismatch negativity (MMN), a brain response elicited every time someone is exposed to an unexpected event. However, rather than the classical auditory mismatch negativity, we found another signature of the brain, which is elicited when the brain is subjected to an action that breaks the regularity (and so an expectation) of another action previously happening - Action Mismatch. We investigated the presence of Action Mismatch Signature (AMS) in an oddball paradigm where instead of a sound we replaced this by video sequences of a hand catching or missing a ball. Performing this experiment on 8 people, our classifier achieves 67% average detection accuracy both across and within subjects. Our AMS signature may provide a powerful tool to automatically monitor and adapt Brain-Robot-Interface and Neuroprosthetic performance.

Robust, ultra low-cost MMG system with brain-machine-interface applications

We propose a Gaussian Process-based regression framework for continuous prediction of the state of missing limbs by exclusively decoding missing limb movements from intact limbs - we achieve this as we have measured the correlation structure and synergies of natural limb kinematics in daily life. Using the example of hand neuroprosthetic, we demonstrate how our model can use non-linear regression to infer the velocity of the flexion/extension joints of missing fingers by observing the intact joints using a data glove. We based our framework on hand joint velocity data, that we recorded with a sensorised glove from 7 able-bodied subjects performing everyday hand movements. We then simulate missing fingers by making our regressors predict the motion that a neuroprosthetic finger should execute based on the previously observed movements of intact fingers. Perhaps surprisingly, we achieve and R2 = 0.89 and an RMSE = 0.20°/s across all missing joints. Moreover, by performing one-subject-out cross validation, we can show that the prediction accuracy and precision has negligible significant loss of performance when tested on new subjects. This suggests that kinematic correlations in daily life can provide a powerful channel refining, if not driving, multi-source neuroprosthetic and Brain Computer Interface approaches.