Dysphagia (swallowing disorder) is a common health problem that degrades the quality of life of many people. The videofluoroscopic swallowing study (VFSS) is the current gold standard in dysphagia assessment but is associated with high cost, long wait times, and a lack of portability. As a result, there is a pining need for an alternative technique that can serve day-to-day monitoring of dysphagia as well as screening for VFSS referral. The primary objective of this thesis was to investigate three non-invasive signal modalities, namely dual-axis accelerometry, submental mechanomyography (MMG), and nasal airflow, for their potential as alternatives to VFSS. To this end, signals were acquired from 17 healthy individuals and 24 patients with dysphagia, with various stimuli. In a characterization study, the anterior-posterior (A-P) and superior-inferior (S-I) axes in dual-axis accelerometry were found to contain non-overlapping information about swallowing, justifying the extension of single-axis (A-P only) to dual-axis (A-P and S-I) accelerometry. Also, several dual-axis accelerometry signal features were found to be stimulus dependent, and the observed stimulus effects were linked to slower swallowing function with increasing bolus viscosity. Age and stimulus effects on submental MMG were scrutinized, as an analogy to previous electromyography (EMG) studies of similar design. Similarities to EMG confirmed the validity of MMG as a muscle activity measurement tool in swallowing research. Automatic swallow segmentation, which is a crucial precursory step to swallow diagnosis, was investigated with artificial neural networks. Segmentation performance was shown to improve as more signal modalities were included, verifying the value of multi-sensor fusion. When all signal modalities were utilized, an adjusted accuracy of 89.6% was achieved. Automatic discrimination between healthy and abnormal swallows was investigated in two studies. Using previously collected pediatric data, a radial basis classifier based only on A-P accelerometry resulted in an adjusted accuracy of 81.3% in aspiration detection. In an adult study, linear discriminant classifiers resulted in adjusted accuracies of 74.7%, 83.7%, and 84.2% for aspiration, valleculae residue, and pyriform sinus residue detection, respectively. It was concluded that the three signal modalities analyzed in this thesis possess promising potential for abnormal swallow detection.

Dysphagia (swallowing disorder) is a common health problem that degrades the quality of life of many people. The videofluoroscopic swallowing study (VFSS) is the current gold standard in dysphagia assessment but is associated with high cost, long wait times, and a lack of portability. As a result, there is a pining need for an alternative technique that can serve day-to-day monitoring of dysphagia as well as screening for VFSS referral. The primary objective of this thesis was to investigate three non-invasive signal modalities, namely dual-axis accelerometry, submental mechanomyography (MMG), and nasal airflow, for their potential as alternatives to VFSS. To this end, signals were acquired from 17 healthy individuals and 24 patients with dysphagia, with various stimuli. In a characterization study, the anterior-posterior (A-P) and superior-inferior (S-I) axes in dual-axis accelerometry were found to contain non-overlapping information about swallowing, justifying the extension of single-axis (A-P only) to dual-axis (A-P and S-I) accelerometry. Also, several dual-axis accelerometry signal features were found to be stimulus dependent, and the observed stimulus effects were linked to slower swallowing function with increasing bolus viscosity. Age and stimulus effects on submental MMG were scrutinized, as an analogy to previous electromyography (EMG) studies of similar design. Similarities to EMG confirmed the validity of MMG as a muscle activity measurement tool in swallowing research. Automatic swallow segmentation, which is a crucial precursory step to swallow diagnosis, was investigated with artificial neural networks. Segmentation performance was shown to improve as more signal modalities were included, verifying the value of multi-sensor fusion. When all signal modalities were utilized, an adjusted accuracy of 89.6% was achieved. Automatic discrimination between healthy and abnormal swallows was investigated in two studies. Using previously collected pediatric data, a radial basis classifier based only on A-P accelerometry resulted in an adjusted accuracy of 81.3% in aspiration detection. In an adult study, linear discriminant classifiers resulted in adjusted accuracies of 74.7%, 83.7%, and 84.2% for aspiration, valleculae residue, and pyriform sinus residue detection, respectively. It was concluded that the three signal modalities analyzed in this thesis possess promising potential for abnormal swallow detection.

This thesis proposes nonlinear system identification techniques for the mathematical modeling of the human smooth pursuit system (SPS) with application to motor symptom quantification in Parkinson's disease (PD). The SPS refers to the complex neuromuscular system in humans that governs the smooth pursuit eye movements (SPEM). Insight into the SPS and its operation is of importance in a wide and steadily expanding array of application areas and research fields. The ultimate purpose of the work in this thesis is to attain a deeper understanding and quantification of the SPS dynamics and thus facilitate the continued development of novel commercial products and medical devices. The main contribution of this thesis is in the derivation and evaluation of several techniques for SPS characterization. While attempts to mathematically model the SPS have been made in the literature before, several key aspects of the problem have been previously overlooked.This work is the first one to devise dynamical models intended for extended-time experiments and also to consider systematic visual stimuli design in the context of SPS modeling. The result is a handful of parametric mathematical models outperforming current State-of-the-Art models in terms of prediction accuracy for rich input signals. As a complement to the parametric dynamical models, a non-parametric technique involving the construction of individual statistical models pertaining to specific gaze trajectories is suggested. Both the parametric and non-parametric models are demonstrated to successfully distinguish between individuals or groups of individuals based on eye movements.Furthermore, a novel approach to Wiener system identification using Volterra series is proposed and analyzed. It is exploited to confirm that the SPS in healthy individuals is indeed nonlinear, but that the nonlinearity of the system is significantly stronger in PD subjects. The nonlinearity in healthy individuals appears to be well-modeled by a static output function, whereas the nonlinear behavior introduced to the SPS by PD is dynamical.

The main goal of this thesis is to test the feasibility of human identification using the Electrocardiogram (ECG). Such biomedical signal has several key advantages including its intrinsic nature and liveness indicator which makes it more secure compared to some of the existing conventional and traditional biometric modalities. In compliance with the terms and regulations of McMaster University, this work has been assembled into a sandwich thesis format which consist of three journal papers. The main idea of this work is to identify individuals using distance measurement techniques and ECG feature matching. In addition, we gradually developed the content of the three papers. In the first paper, we started with the general criteria for developing ECG based biometric systems. To explain, we proposed both fiducial and non-fiducial approaches to extract the ECG features followed by providing comparative study on the performance of both approaches. Next, we applied non-overlapped data windows to extract the ECG morphological and spectral features. The former set of features include the amplitude and slope differences between the Q, R and S peaks. The later features include extracting magnitudes of the ECG frequency components using short time Fourier Transform (STFT). In addition, we proposed a methodology for QRS detection and segmentation using STFT and binary classification of ECG fiducial features. In the second paper, we proposed a technique for choosing overlapped data windows to extract the abovementioned features. Namely, the dynamic change in the ECG features from heart beats to heartbeat is utilized for identification purposes. To improve the performance of the proposed techniques we developed Frechet-mean based classifier for this application. These classifiers exploit correlation matrix structure that is not accounted for in classical Euclidean techniques. In addition to considering the center of the cluster, the Frechet-mean based techniques account for the shape of the cluster as well. In the third paper, the thesis is extended to address the variability of ECG features over multiple records. Specifically, we developed a multi-level wavelet-based filtering system which utilizes features for multiple ECGs for human identification purposes. In addition, we proposed a soft decision-making technique to combine information collected from multi-level identification channels to reach a common final class. Lastly, we evaluated the robustness of all our proposed methods over several random experiments by changing the testing data and we achieved excellent results. The results of this thesis show that the ECG is a promising biometric modality. We evaluated the performance of the proposed methods on the public ECG ID database because it was originally recorded for biometric purposes. In addition, to make performance evaluation more realistic we used two recordings of the same person obtained under possibly different conditions. Furthermore, we randomly changed both the training and testing data which are obtained from the full ECG records for performance evaluation purposes. However, it is worth mentioning that in all parts of the thesis, various parameters settings are presented to support the main ideas and it is subject to change according to human activity and application requirements. Finally, the thesis concludes with a comparison between all the proposed methods, and it provides suggestions on few open problems that can be considered for future research as extension to the work that has been done in this thesis. Generally, these problems are associated with the constraints on computational time, data volume and ECG clustering. / Thesis / Doctor of Philosophy (PhD)

Many people around the world suffer from neurological injuries of various sorts that cause serious difficulties in their lives, due to the loss of important sensory and motor functions. Functional electrical stimulation (FES) provides a possible solution to these difficulties by means of a feedback connection allowing the target organ (or organs) to be controlled by electrical stimulation. The control signals can be provided using recorded data extracted from the nerves (electroneurogram, ENG). The most common and safe approaches for interfacing with nerves is called cuff electrodes which deliver the required feedback path for the implantable system with minimum risk. The amount of recorded information can be improved by increasing the number of electrodes within a single cuff known as multi-electrode cuffs (MECs) configuration. This strategy can increase the signal to noise ratio for the recorded signals which have typically very low amplitude (less than 5μV). Consequently multiple high gain amplifiers are used in order to amplify the signals and supply a multi-channel recorded data stream for signal processing or monitoring applications. The signal processing unit within the implantable system or outside the body is employed for classification and sorting the action potential signals (APs) depending on their conduction velocities. This method is called velocity selective recording (VSR). Basically, the idea of this approach is that the conduction velocity of AP can be determined by timing the appearance of the signal at two or more points along the nerve and then dividing the distance between the points by the delay. The purpose of this thesis to investigate an alternative approach using artificial network for APs detection and extraction in neural recording applications to increase the velocity selectivity based on VSR using MECs. The prototype systems impose four major requirements which are high velocity selectivity, small size, low power consumption and high reliability. The proposed method has been developed for applications which require online AP classification. A novel time delay neural network (TDNN) approach is used to decompose the recorded data into several matched velocity bands to allow for individual velocity selectivity at each band to be increased. Increasing the velocity selectivity leads to more accurate recording from the target fibre (or fibres) within the nerve bundle which can be used for applications that require AP classification such as bladder control and the adjustment of foot drop. The TDNN method was developed to obtain more information from an individual cuff without increasing the number of electrodes or the sampling rate. Moreover, the optimization of the hardware implementation for the proposed signal processing method permits savings in power consumption and silicon area. Finally, a nerve signal synthesiser and noise generator for the evaluation of the VSRmethod is described. This system generates multiple artificial AP signals with a time offset between the channels with additive white Gaussian noise (AWGN) to simulate the MEC and hence reduce the cost and the number of the animals required for experimental tests.

A convolutional deep neural network is employed to assess renal autoregulation using time series of arterial blood pressure and blood flow rate measurements in conscious rats. The network is trained using representative data samples from rats with intact autoregulation and rats whose autoregulation is impaired by the calcium channel blocker amlodipine. Network performance is evaluated using test data of the types used for training, but also with data from other models for autoregulatory impairment, including different calcium channel blockers and also renal mass reduction. The network is shown to provide effective classification for impairments from calcium channel blockers. However, the assessment of autoregulation when impaired by renal mass reduction was not as clear, evidencing a different signature in the hemodynamic data for that impairment model. When calcium channel blockers were given to those animals, however, the classification again was effective.

Na área de processamento de sinais biomédicos a extração de informação, baseada em um conjunto de medidas adquiridas no tempo, é considerada de suma importância. A qualidade desta informação extraída permite avaliar o funcionamento dos diversos órgãos. Objetivos: (1) propor o método de análise de componentes dependentes para auxiliar a extração de componentes de interesse, a partir de medidas multivariadas; (2) caraterizar as componentes extraídas através de representações em termos de tempo e freqüência, e espectro de potência; e, (3) aplicar o método e avaliar as componentes de interesse extraídas no contexto real MCGf, MGG e fMRI. A proposta para a extração fundamenta-se no método chamado de Análise de Componentes Dependentes ACD. As medidas a serem processadas são multivariadas a partir de sensores distribuídos, espacialmente, no corpo humano dando origem a um conjunto de dados correlacionados no tempo e/ou no espaço. Observa-se que os sinais de interesse raramente são registrados de forma isolada, e sim misturados com outros sinais superpostos, ruído e artefatos fisiológicos ou ambientais, onde a relação sinal-ruído é geralmente baixa. Nesse contexto, a estratégia a ser utilizada baseia-se na ACD, que permitirá extrair um pequeno número de fontes, de potencial interesse, com informações úteis. A estratégia ACD para extração de informação é aplicada em três importantes problemas, na área de processamento de sinais biomédicos: (1) detecção do sinal do feto em magnetocardiografia fetal (MCGf); (2) detecção da atividade de resposta elétrica do estômago em magnetogastrografia (MGG); e, (3) detecção de regiões ativas do cérebro em experimentos de imagens por ressonância magnética funcional (Functional Magnetic Resonance Imaging, fMRI). Os resultados, nos três casos estudados, mostraram que o método utilizado, como estratégia, é efetivo e computacionalmente eficiente para extração de sinais de interesse. Concluímos, baseados nas aplicações, que o método proposto é eficaz, mostrando seu potencial para futuras pesquisas clínicas. / An important goal in biomedical signal processing is the extraction of information based on a set of physiological measurements made along time. Generally, biomedical signals are electromagnetic measurements. Those measurements (usually made with multichannel equipment) are registered using spatially distributed sensors around some areas of the human body, originating a set of time and/or space correlated data. The signals of interest are rarely registered alone, being usually observed as a mixture of other spurious, noisy signals (sometimes superimposed) and environmental or physiological artifacts. More over, the signal-to-noise ratio is generally low. In many applications, a big number of sensors are available, but just a few sources are of interest and the remainder can be considered noise. For such kind of applications, it is necessary to develop trustful, robust and effective learning algorithms that allow the extraction of only a few sources potentially of interest and that hold useful information. The strategy used here for extraction of sources is applied in three important problems in biomedical signal processing: (1) detection of the fetal magnetocardiogram signal (fMCG); (2) detection of the electrical activity of the stomach in magnetogastrograms (MGG); and (3) detection of active regions of the brain in experiments in functional Magnetic Resonance Imaging (fMRI). The results, within the three cases of study, showed that the DCA method used as strategy is effective and computationally efficient on extraction of desired signals.

Na área de processamento de sinais biomédicos a extração de informação, baseada em um conjunto de medidas adquiridas no tempo, é considerada de suma importância. A qualidade desta informação extraída permite avaliar o funcionamento dos diversos órgãos. Objetivos: (1) propor o método de análise de componentes dependentes para auxiliar a extração de componentes de interesse, a partir de medidas multivariadas; (2) caraterizar as componentes extraídas através de representações em termos de tempo e freqüência, e espectro de potência; e, (3) aplicar o método e avaliar as componentes de interesse extraídas no contexto real MCGf, MGG e fMRI. A proposta para a extração fundamenta-se no método chamado de Análise de Componentes Dependentes ACD. As medidas a serem processadas são multivariadas a partir de sensores distribuídos, espacialmente, no corpo humano dando origem a um conjunto de dados correlacionados no tempo e/ou no espaço. Observa-se que os sinais de interesse raramente são registrados de forma isolada, e sim misturados com outros sinais superpostos, ruído e artefatos fisiológicos ou ambientais, onde a relação sinal-ruído é geralmente baixa. Nesse contexto, a estratégia a ser utilizada baseia-se na ACD, que permitirá extrair um pequeno número de fontes, de potencial interesse, com informações úteis. A estratégia ACD para extração de informação é aplicada em três importantes problemas, na área de processamento de sinais biomédicos: (1) detecção do sinal do feto em magnetocardiografia fetal (MCGf); (2) detecção da atividade de resposta elétrica do estômago em magnetogastrografia (MGG); e, (3) detecção de regiões ativas do cérebro em experimentos de imagens por ressonância magnética funcional (Functional Magnetic Resonance Imaging, fMRI). Os resultados, nos três casos estudados, mostraram que o método utilizado, como estratégia, é efetivo e computacionalmente eficiente para extração de sinais de interesse. Concluímos, baseados nas aplicações, que o método proposto é eficaz, mostrando seu potencial para futuras pesquisas clínicas. / An important goal in biomedical signal processing is the extraction of information based on a set of physiological measurements made along time. Generally, biomedical signals are electromagnetic measurements. Those measurements (usually made with multichannel equipment) are registered using spatially distributed sensors around some areas of the human body, originating a set of time and/or space correlated data. The signals of interest are rarely registered alone, being usually observed as a mixture of other spurious, noisy signals (sometimes superimposed) and environmental or physiological artifacts. More over, the signal-to-noise ratio is generally low. In many applications, a big number of sensors are available, but just a few sources are of interest and the remainder can be considered noise. For such kind of applications, it is necessary to develop trustful, robust and effective learning algorithms that allow the extraction of only a few sources potentially of interest and that hold useful information. The strategy used here for extraction of sources is applied in three important problems in biomedical signal processing: (1) detection of the fetal magnetocardiogram signal (fMCG); (2) detection of the electrical activity of the stomach in magnetogastrograms (MGG); and (3) detection of active regions of the brain in experiments in functional Magnetic Resonance Imaging (fMRI). The results, within the three cases of study, showed that the DCA method used as strategy is effective and computationally efficient on extraction of desired signals.

abstract: Research on developing new algorithms to improve information on brain functionality and structure is ongoing. Studying neural activity through dipole source localization with electroencephalography (EEG) and magnetoencephalography (MEG) sensor measurements can lead to diagnosis and treatment of a brain disorder and can also identify the area of the brain from where the disorder has originated. Designing advanced localization algorithms that can adapt to environmental changes is considered a significant shift from manual diagnosis which is based on the knowledge and observation of the doctor, to an adaptive and improved brain disorder diagnosis as these algorithms can track activities that might not be noticed by the human eye. An important consideration of these localization algorithms, however, is to try and minimize the overall power consumption in order to improve the study and treatment of brain disorders. This thesis considers the problem of estimating dynamic parameters of neural dipole sources while minimizing the system's overall power consumption; this is achieved by minimizing the number of EEG/MEG measurements sensors without a loss in estimation performance accuracy. As the EEG/MEG measurements models are related non-linearity to the dipole source locations and moments, these dynamic parameters can be estimated using sequential Monte Carlo methods such as particle filtering. Due to the large number of sensors required to record EEG/MEG Measurements for use in the particle filter, over long period recordings, a large amounts of power is required for storage and transmission. In order to reduce the overall power consumption, two methods are proposed. The first method used the predicted mean square estimation error as the performance metric under the constraint of a maximum power consumption. The performance metric of the second method uses the distance between the location of the sensors and the location estimate of the dipole source at the previous time step; this sensor scheduling scheme results in maximizing the overall signal-to-noise ratio. The performance of both methods is demonstrated using simulated data, and both methods show that they can provide good estimation results with significant reduction in the number of activated sensors at each time step. / Dissertation/Thesis / M.S. Electrical Engineering 2012

Orientador: Prof. Dr. Pai Chi Nan / Dissertação (mestrado) - Universidade Federal do ABC, Programa de Pós-Graduação em Engenharia Biomédica, 2014. / A alta demanda de transplantes de coração não é suprida pela quantidade de doadores do órgão. Diferentes alternativas aos transplantes são pesquisadas, sendo uma delas a utilização de bombas de sangue centrífugas (BSC) com mancal magnético (MM), apresentando maior durabilidade do que demais bombas. Porém acredita-se que o fluxo contínuo produzido por essas bombas possa causar danos no organismo em longo prazo, havendo a necessidade da utilização de fluxo sanguíneo pulsátil. O objetivo do trabalho foi produzir fluxo sanguíneo pulsátil sincronizado com o coração do paciente com uma BSC com MM. Para a produção do fluxo pulsátil foi feita a alteração da velocidade de rotação da BSC de acordo com os sinais eletrocardiográficos (ECG) do paciente. Foram utilizadas três fontes de sinal de ECG para os testes: gerador de funções, simulador de paciente e voluntário. O algoritmo implementado no Simulink® inferiu os instantes dos batimentos cardíacos seguintes identificando regiões acima de um determinado limiar (fixo e móvel) de amplitude máxima (picos QRS), que correspondem ao início dos batimentos. Calculou-se a média (fixa e móvel) da taxa de batimentos e inferiu-se o instante de tempo aproximado do pico QRS seguinte no sinal de ECG. A velocidade de rotação da BSC foi aumentada nos instantes inferidos para os QRS e após algum tempo foi reduzida. A resposta do motor foi avaliada no ar utilizando sinais degraus e velocidades de rotação definidas pelo algoritmo como entrada. As velocidades reais do motor foram obtidas utilizando a amostra do gerador e do voluntário. A possibilidade de produção de fluxo pulsátil nas condições encontradas foi avaliada. O algoritmo permitiu a identificação dos picos dos complexos QRS, o cálculo da taxa de batimentos e a inferência dos batimentos seguintes. Os valores de erro absolutos e relativos entre os instantes inferidos e reais foram baixos para os sinais das três fontes, tendo valores máximos aproximados de 0,5s (120 a 60bpm) e 51% (60 a 90bpm) para as amostras do gerador e do simulador, e de 0,157s e 15,94% para a amostra do voluntário. A elevação da velocidade de rotação de referência para 1900rpm ocorreu nos instantes dos valores inferidos e após 0,4s a velocidade foi reduzida para 1500rpm. Os momentos de elevação da velocidade de rotação se apresentaram sincronizados com os instantes dos batimentos. O tempo de resposta do motor foi de 0,04s. A estabilização das oscilações da velocidade real ocorreu após 2s, com máximos e mínimos acima e abaixo dos valores de referência, e médias das oscilações aproximadamente de 1700rpm, de acordo com a frequência do ECG. Houve sincronização entre a velocidade de referência e a real. As amplitudes de oscilação elevadas indicaram que os parâmetros do controlador do motor deveriam ser modificados para sua utilização em uma aplicação na água, com velocidade variável de entrada, e para seu ganho variável estar em intervalos de estabilidade. Outro dispositivo é necessário para medir fluxo e pressão do sistema produzindo fluxo pulsátil. / The high demand for heart transplants is not supplied by the number of donors. Alternatives to transplants are being researched, being one of these the use of centrifugal blood pumps (CBP) with magnetic bearing (MB), which have higher durability than other types of pumps. Nevertheless it is believed that the continuous flow generated by these pumps may cause long-term damage to the body, being necessary the use of pulsatile blood flow. The objective of this work is to produce pulsatile blood flow synchronized with the heart of the patient using a CBP with MB. The synchronized pulsatile flow was done by changing the rotational speed of the pump, according to the electrocardiographic (EKG) signals of the patient. Three different sources of EKG signal were used: a function generator, a patient simulator and a volunteer. The algorithm, implemented on Simulink®, inferred the moments of the following heart beats by finding points, above a certain threshold (fixed and moving), of maximum amplitude (QRS complexes) on the EKG signal. The mean of the heart beat rate (fixed or moving) was calculated and the instant of the next QRS peaks were inferred. The rotational speed of the CBP was increased on the moments of the QRS inferred and decreased after some time. The motor response was evaluated on air, using step signals and rotational speeds defined by the algorithm as inputs. The motor rotational speeds were obtained with the generators¿ and the volunteer¿s samples. The possibility of generation of pulsatile flow was evaluated. The algorithm identified the QRS complexes peaks, calculated the mean heart rate and inferred the following heart beats. The absolute and relative errors between the inferred and the real instants of the QRS peaks were low for the signals of the three sources, with maximum approximate values of 0,5s (120 to 60bpm) and 51% (60 to 90bpm) for the function generator and the simulator samples, and of 0,157s and 15,94% for the volunteer sample. The increase of the reference rotational speed to 1900rpm was done on the instants of the inferred values of QRS complexes, with reduction to 1500rpm after 0,4s. The moments in which the rotational speed is on its maximum values are about the same as the QRS complexes of the signal. The motor response was of 0,04s. The stabilization of the oscillating speed occurred after 2s, with maximum and minimum values above and below the reference ones, and mean values around 1700rpm, according to the heart beating rate. Synchronization between the reference and real rotational speeds was observed. The high amplitudes of the oscillations pointed out that the motor driver parameters should be changed to be used with variable rotational speed input and for the variable gain to remain between the driver stability values. Another device is required to measure flow and pressure of the system producing pulsatile flow.