No description available.

This work seeks to more fully characterize how the representation of English vowels changes with increasing sound level in the frequency following response (FFR) of normal-hearing adult subjects. It further seeks to help inform the design of brain-computer interfaces (BCI) that exploit the FFR for hearing aid (HA) applications. The results of three studies are presented, followed by a theoretical examination of the potential BCI space as it relates to HA design. The first study examines how the representation of a long vowel changes with level in normal hearing subjects. The second study examines how the representation of four short vowels changes with level in normal hearing subjects. The third study utilizes machine learning techniques to automatically classify the FFRs captured in the second study. Based in-part on the findings from these three studies, potential avenues to pursue with respect to the utilization of the FFR in the automated fitting of HAs are proposed. The results of the first two studies suggest that the FFR to vowel stimuli presented at levels in the typical speech range provide robust and differentiable representations of both envelope and temporal fine structure cues present in the stimuli in both the time and frequency domains. The envelope FFR at the fundamental frequency (F0) was generally not monotonic-increasing with level increases. The growth of the harmonics of F0 in the envelope FFR were consistent indicators of level-related effects, and the harmonics related to the first and second formants were also consistent indicators of level effects. The third study indicates that common machine-learning classification algorithms are able to exploit features extracted from the FFR, both in the time and frequency domains, in order to accurately predict both vowel and level classes among responses. This has positive implications for future work regarding BCI-based approaches to HA fitting, where controlling for clarity and loudness are important considerations.

This work details CMOS, bio-inspired, bio-compatible circuits which were used as synapses between an artificial neuron and a living neuron and between two living neurons. An intracellular signal from a living neuron was amplified, an integrate-and-fire neuron was used as a simple processing element to detect the spikes, and an artificial synapse was used to send outputs to another living neuron. The key structure is an electronic synapse which is based around a floating-gate pFET. The charge on the floating-gate is analogous to the synaptic weight and can be modified. This modification can be viewed as similar to long-term potentiation and long-term depression. The modification can either be programmed (supervised learning) or can adapt to the inputs (unsupervised learning). Since the technology to change the floating-gate weight has greatly improved, these weights can be set quickly and accurately. Intrinsic floating-gate learning rules were explored and the ability to change the synaptic weight was shown.

Neuro-engineering is an emerging multi-disciplinary domain which investigates the electrophysiological activities of the nervous system. It provides procedures and techniques to explore, analyze and characterize the functions of the different components comprising the nervous system. Neuro-engineering is not limited to research applications; it is employed in developing unconventional therapeutic techniques for treating different neurological disorders and restoring lost sensory or motor functions. Microelectrodes are principal elements in functional electric stimulation (FES) systems used in electrophysiological procedures. They are used in establishing an interface with the individual neurons or in clusters to record activities and communications, as well as modulate neuron behaviour through stimulation. Microelectrode technologies progressed through several modifications and innovations to improve their functionality and usability. However, conventional electrode technologies are open to further development, and advancement in microelectrodes technology will progressively meliorate the neuro-interfacing and electrotherapeutic techniques. This research introduced design methodology and fabrication processes for intra-cortical microelectrodes capable of befitting a wide range of design requirements and applications. The design process was employed in developing and implementing an ensemble of intra-cortical microelectrodes customized for different neuro-interfacing applications. The proposed designs presented several innovations and novelties. The research addressed practical considerations including assembly and interconnection to external circuitry. The research was concluded by exhibiting the Waterloo Array which is a high channel count flexible 3-D neuro-interfacing array. Finally, the dissertation was concluded by demonstrating the characterization, in vitro and acute in vivo testing results of the Waterloo Array. The implemented electrodes were tested and benchmarked against commercial equivalents and the results manifested improvement in the electrode performance compared to conventional electrodes. Electrode testing and evaluation were conducted in the Krembil Neuroscience Centre Research Lab (Toronto Western Hospital), and the Neurosciences & Mental Health Research Institute (the Sick Kids hospital). The research results and outcomes are currently being employed in developing chronic intra-cortical and electrocorticography (ECoG) electrode arrays for the epilepsy research and rodents nervous system investigations. The introduced electrode technologies will be used to develop customized designs for the clinical research labs collaborating with CIRFE Lab.

abstract: Optical Fibers coupled to laser light sources, and Light Emitting Diodes are the two classes of technologies used for optogenetic experiments. Arizona State University's Flexible Display Center fabricates novel flexible Organic Light Emitting Diodes(OLEDs). These OLEDs have the capability of being monolithically fabricated over flexible, transparent plastic substrates and having power efficient ways of addressing high density arrays of LEDs. This thesis critically evaluates the technology by identifying the key advantages, current limitations and experimentally assessing the technology in in-vivo and in-vitro animal models. For in-vivo testing, the emitted light from a flat OLED panel was directly used to stimulate the neo-cortex in the M1 region of transgenic mice expressing ChR2 (B6.Cg-Tg (Thy1-ChR2/EYFP) 9Gfng/J). An alternative stimulation paradigm using a collimating optical system coupled with an optical fiber was used for stimulating neurons in layer 5 of the motor cortex in the same transgenic mice. EMG activity was recorded from the contralateral vastus lateralis muscles. In vitro testing of the OLEDs was done in primary cortical neurons in culture transfected with blue light sensitive ChR2. The neurons were cultured on a microelectrode array for taking neuronal recordings. / Dissertation/Thesis / ICMS response in front and hind limb / Optogenetic response using iLEDs and OLEDs / iLED vs iLED coupled to optical fiber response / Masters Thesis Bioengineering 2015

Controlling an exoskeleton glove with a highly efficient human-machine interface (HMI), while accurately applying force to each joint remains a hot topic. This paper proposes a fast, secure, accurate, and portable solution to control an exoskeleton glove. This state of the art solution includes both hardware and software components. The exoskeleton glove uses a modified serial elastic actuator (SEA) to achieve accurate force sensing. A portable electronic system is designed based on the SEA to allow force measurement, force application, slip detection, cloud computing, and a power supply to provide over 2 hours of continuous usage. A voice-control-based HMI referred to as the integrated trigger-word configurable voice activation and speaker verification system (CVASV), is integrated into a robotic exoskeleton glove to perform high-level control. The CVASV HMI is designed for embedded systems with limited computing power to perform voice-activation and voice-verification simultaneously. The system uses MobileNet as the feature extractor to reduce computational cost. The HMI is tuned to allow better performance in grasping daily objects. This study focuses on applying the CVASV HMI to the exoskeleton glove to perform a stable grasp with force-control and slip-detection using SEA based exoskeleton glove. This research found that using MobileNet as the speaker verification neural network can increase the speed of processing while maintaining similar verification accuracy. / Master of Science / The robotic exoskeleton glove used in this research is designed to help patients with hand disabilities. This thesis proposes a voice-activated grasping system to control the exoskeleton glove. Here, the user can use a self-defined keyword to activate the exoskeleton and use voice to control the exoskeleton. The voice command system can distinguish between different users' voices, thereby improving the safety of the glove control. A smartphone is used to process the voice commands and send them to an onboard computer on the exoskeleton glove. The exoskeleton glove then accurately applies force to each fingertip using a force feedback actuator.This study focused on designing a state of the art human machine interface to control an exoskeleton glove and perform an accurate and stable grasp.

The Hemispherical Resonating Gyroscope (HRG) is an inertial sensor, flagship of Safran’s industry. When exiting the assembly line, it has its own physical flaws. In order to identify and correct them, operators perform several tests on the sensor: this process corresponds to the calibration step of the sensors. The latter is done by a Matlab calibration bench, which allows to calculate compensation polynomial functions, which are then included in the algorithms of the sensor’s implementation card. However, the link between the calculated functions and the sensor’s flaws is not obvious and therefore, it is impossible to check their truthfulness without further verification. In this document, an interfacing method between a calibration bench and a virtual HRG, modeled in Simulink, has been described. After presenting the sensor’s capabilities, several interfacing methods are discussed, before keeping the more dynamical one, based on object oriented programming and the implementation of a time continuity between Simulink data recordings. Such interfacing allows for the simulation of the behavior of a gyroscope during calibration, and the comparison of these results to the ones obtained on real sensors. This comparison highlighted a certain consistency between the results and also several flaws caused by the interfacing. Particularly, the fact that the signal discretization has a significant impact on the errors. Moreover, one can notice that the simulation time is significantly longer than the calibration time and suggests that the interfacing time may require optimization of its efficiency. / HRG är en tröghetssensor som är flaggskeppet inom Safrans industri. När sensorn lämnar monteringslinjen är den inte felfri. För att identifiera och kompensera för dessa fel utför operatörerna flera tester på sensorn, i flera olika kalibreringssteg. De senare görs med hjälp av en Matlab-kalibreringsbänk, som gör det möjligt att beräkna kompensationspolynomfunktioner, som sedan implementeras i algoritmerna på sensorns implementeringskort. Kopplingen mellan de beräknade funktionerna och sensorns fel är dock inte uppenbar och därför är det omöjligt att kontrollera deras noggrannhet utan ytterligare kontroller. I detta dokument beskrivs en gränssnittsmetod mellan en kalibreringsbänk och en virtuell HRG, modellerad i Simulink. Efter att ha presenterat sensorns funktion har flera gränssnittsmetoder studerats, innan man valde den mer dynamiska, baserad på objektorienterad programmering och implementeringen av en tidskontinuitet mellan Simulinkdatainspelningar. Ett sådant gränssnitt gjorde det möjligt att få vissa resultat som simulerar gyroskopets beteende under en kalibrering och att jämföra dessa resultat med dem som erhållits på verkliga sensorer. Jämförelsen visade på en viss överensstämmelse mellan resultaten, men också på flera brister som orsakats av gränssnittet. I synnerhet det faktum att signaldiskretiseringen har en betydande inverkan på felen. Dessutom kan man notera att simuleringstiden är mycket längre än kalibreringstiden och leder till tanken att det finns sätt att förbättra gränssnittet för att göra det mer tidseffektivt.

Sequences are ubiquitous in the domain of biology. One of the current best machine learning techniques for analysing sequences is the long short-term memory (LSTM) network. Owing to significant barriers to adoption in biology, focussed efforts are required to realize the use of LSTMs in practice. Thus, the aim of this work is to improve the state of LSTMs for biology, and we focus on biological tasks pertaining to physiological signals, peripheral neural signals, and molecules. This goal drives the three subplots in this thesis: biological applications, visualizations, and extensions. We start by demonstrating the utility of LSTMs for biological applications. On two new physiological-signal datasets, LSTMs were found to outperform hidden Markov models. LSTM-based models, implemented by other researchers, also constituted the majority of the best performing approaches on publicly available medical datasets. However, even if these models achieve the best performance on such datasets, their adoption will be limited if they fail to indicate when they are likely mistaken. Thus, we demonstrate on medical data that it is straightforward to use LSTMs in a Bayesian framework via dropout, providing model predictions with corresponding uncertainty estimates. Another dataset used to show the utility of LSTMs is a novel collection of peripheral neural signals. Manual labelling of this dataset is prohibitively expensive, and as a remedy, we propose a sequence-to-sequence model regularized by Wasserstein adversarial networks. The results indicate that the proposed model is able to infer which actions a subject performed based on its peripheral neural signals with reasonable accuracy. As these LSTMs achieve state-of-the-art performance on many biological datasets, one of the main concerns for their practical adoption is their interpretability. We explore various visualization techniques for LSTMs applied to continuous-valued medical time series and find that learning a mask to optimally delete information in the input provides useful interpretations. Furthermore, we find that the input features looked for by the LSTM align well with medical theory. For many applications, extensions of the LSTM can provide enhanced suitability. One such application is drug discovery -- another important aspect of biology. Deep learning can aid drug discovery by means of generative models, but they often produce invalid molecules due to their complex discrete structures. As a solution, we propose a version of active learning that leverages the sequential nature of the LSTM along with its Bayesian capabilities. This approach enables efficient learning of the grammar that governs the generation of discrete-valued sequences such as molecules. Efficiency is achieved by reducing the search space from one over sequences to one over the set of possible elements at each time step -- a much smaller space. Having demonstrated the suitability of LSTMs for biological applications, we seek a hardware efficient implementation. Given the success of the gated recurrent unit (GRU), which has two gates, a natural question is whether any of the LSTM gates are redundant. Research has shown that the forget gate is one of the most important gates in the LSTM. Hence, we propose a forget-gate-only version of the LSTM -- the JANET -- which outperforms both the LSTM and some of the best contemporary models on benchmark datasets, while also reducing computational cost.

The emergence of organic electronics - a technology that relies on carbon-based semiconductors to deliver devices with unique properties - represents one of the most dramatic developments of the past two decades. A rapidly emerging new direction in the field involves the interface with biology. The "soft" nature of organics offers better mechanical compatibility with tissue than traditional electronic materials, while their natural compatibility with mechanically flexible substrates suits the non-planar form factors often required for implants. More importantly, their ability to conduct ions in addition to electrons and holes opens up a new communication channel with biology. The coupling of electronics with living tissue holds the key to a variety of important life-enhancing technologies. One example is bioelectronic implants that record neural signals and/or electrically stimulate neurons. These devices offer unique opportunities to understand and treat conditions such as hearing and vision loss, epilepsy, brain degenerative diseases, and spinal cord injury.The engineering aspect of the work includes the development of a photolithographic process to integrate the conducting polymer poly(3,4-ethylenedioxythiophene: poly(styrene sulfonate) (PEDOT:PSS) with parylene C supports to make an active device. The technology is used to fabricate electrocorticography (ECoG) probes, high-speed transistors and wearable biosensors. The experimental work explores the fundamentals of communication at the interface between conducting polymers and the brain. It is shown that conducting polymers outperform conventional metallic electrodes for brain signals recording.Organic electrochemical transistors (OECTs) represent a step beyond conducting polymer electrodes. They consist of a conducting polymer channel in contact with an electrolyte. When a gate electrode excites an ionic current in the electrolyte, ions enter the polymer film and change its conductivity. Since a small amount of ions can effectively "block" the transistor channel, these devices offer significant amplification in ion-to-electron transduction. Using the developed technology a high-speed and high-density OECTs array is presented. The dense architecture of the array improves the resolution of the recording from neural networks and the transistors temporal response are 100 μs, significantly faster than the action potential. The experimental transistor responses are fit and modeled in order to optimize the gain of the transistor. Using the model, an OECT with two orders of magnitude higher normalized transconductance per channel width is fabricated as compared to Silicon-based field effect transistors. Furthermore, the OECTs are integrated to a highly conformable ECoG probe. This is the first time that a transistor is used to record brain activities in vivo. It shows a far superior signal-to-noise-ratio (SNR) compare to electrodes. The high SNR of the OECT recordings enables the observation of activities from the surface of the brain that only a perpetrating probe can record. Finally, the application of OECTs for biosensing is explored. The bulk of the currently available biosensors often require complex liquid handling, and thus suffer from problems associated with leakage and contamination. The use of an organic electrochemical transistor for detection of lactate by integration of a room temperature ionic liquid in a gel-format, as a solid-state electrolyte is demonstrated.

Brain-computer interfacing (BCI) is a new method of human-machine interaction. It involves the extraction of information from the electroencephalogram (EEG) through signal processing and pattern recognition. The technology has far reaching implications for those with severe physical disabilities and has the potential to enhance machine interaction for the rest of the population. In this work we investigate time-frequency analysis in motor-imagery BCI. We consider two methods for signal analysis: adaptive autoregressive models (AAR) and wavelet transform (WAV). There are three major contributions of this research to single-trial analysis in motor-imagery BCI. First, we improve classification of AAR features over a conventional method by applying a temporal evidence accumulation (TEA) framework. Second, we compare the performance of AAR and WAV under the TEA framework for three subjects and find that WAV outperforms AAR for two subjects. The subject for whom AAR outperforms WAV has the lowest overall signal-to-noise ratio in their BCI output, an indication that the AAR model is more robust than WAV for noisier signals. Lastly, we find empirical evidence of complimentary information between AAR and WAV and propose a fusion scheme that increases the mutual information between the BCI output and classes.