Search

Two wheeled Self balancing robots have become more popular in recent times and used in many applications, commonly used in transportation as these balancing robots are faster and apparently stable. Balancing is achieved by using controllers tuned for the system. Inverted pendulum principle is used in these robots. This paper proposes the use of two control methods, namely Proportional-Integral-Derivative (PID) and Extended PID (EPID), for maintaining the balance of two-wheeled and self-balancing robots, even when slightly tilted, thus preventing them from falling. PID control theory generally depends up on transfer functions, this paper extends the theory using EPID. EPID provides a new perspective on the working principle of PID control and its relationship with other control theories, such as tracking control and disturbance observer. This approach is illustrated using the fundamental principle of inverted pendulum technology. The key advantage of these robots over other mobile robots is their increased stability due to zero turning radius, which allows for smoother rotations during changes in direction.

Rehabilitation level evaluation is an important part of the automatic rehabilitation training system. As a general rule, this process is manually performed by rehabilitation doctors using chart-based ordinal scales which can be both subjective and inefficient. In this paper, a novel approach based on ensemble learning is proposed which automatically evaluates stroke patients' rehabilitation level using multi-channel sEMG signals to this problem. The correlation between rehabilitation levels and rehabilitation training actions is investigated and actions suitable for rehabilitation assessment are selected. Then, features are extracted from the selected actions. Finally, the features are used to train the stacking classification model. Experiments using sEMG data collected from 24 stroke patients have been carried out to examine the validity and feasibility of the proposed method. The experiment results show that the algorithm proposed in this paper can improve the classification accuracy of 6 Brunnstrom stages to 94.36%, which can promote the application of home-based rehabilitation training in practice.

Clinical assessment of abnormal neuromechanics is typically performed by manipulation of the affected limbs; a process with low inter- and intra-rater reliability. This paper aims at formalizing a framework that closes the loop between a clinician's expertise and computational algorithms, to enhance the clinician's diagnostic capabilities during physical manipulation. The framework's premise is that the dynamics that can be measured by manipulation of a limb are distinct between movement disorders. An a priori database contains measurements encoded in a space called the information map. Based on this map, a computational algorithm identifies which probing motions are more likely to yield distinguishing information about a patient's movement disorder. The clinician executes this movement and the resulting dynamics, combined with clinician input, is used by the algorithm to estimate which of the movement disorders in the database are most probable. This is recursively repeated until a diagnosis can be confidently made. The main contributions of this paper are the formalization of the framework and the addition of the information map to select informative movements. The establishment of the framework provides a foundation for a standardized assessment of movement disorders and future work will aim at testing the framework's efficacy.

This paper presents the development of a new active assistive eating device, which aims to stabilize the movements of people living with movement disorders, such as spasticity and ataxia. Many people living with upper-body incapacities are unable to eat on their own, due to movement disorders (ex. tremors, spastic motions, lack of muscular tone), resulting from various ailments like Cerebral palsy, Parkinson's disease, Dystonia, Multiple sclerosis, strokes, and Muscular dystrophy). Our past work focused on the development of a purely mechanical device, which involved damping of the system via passive mechanical dampers. This paper extends said work by using active stabilization of user movements. The active assistance enables the design of intelligent algorithms that can assist human movements more efficiently. This active version has the benefits of being easily adjustable; the level of damping can be adjusted in real-time, depending on the user movement; different control modes are offered, and the guiding of user movements is also allowed. Firstly, the mechanical design of the device is presented, followed by the damping arrangement, the electronic design, the control algorithms and finally, the preliminary experiments are mentioned.

Alzheimer's disease (AD) accounts for 60%-70% of all dementia cases, and clinical diagnosis at its early stage is extremely difficult. As several new drugs aiming to modify disease progression or alleviate symptoms are being developed, to assess their efficacy, novel robust biomarkers of brain function are urgently required. This paper aims to explore a routine to gain such biomarkers using the quantitative analysis of electroencephalography (QEEG). This paper proposes a supervised classification framework that uses EEG signals to classify healthy controls (HC) and AD participants. The framework consists of data augmentation, feature extraction, K-nearest neighbor (KNN) classification, quantitative evaluation, and topographic visualization. Considering the human brain either as a stationary or a dynamical system, both the frequency-based and time-frequency-based features were tested in 40 participants. The results show that: 1) the proposed method can achieve up to a 99% classification accuracy on short (4s) eyes open EEG epochs, with the KNN algorithm that has best performance when compared with alternative machine learning approaches 2) the features extracted using the wavelet transform produced better classification performance in comparison to the features based on FFT and 3) in the spatial domain, the temporal and parietal areas offer the best distinction between healthy controls and AD. The proposed framework can effectively classify HC and AD participants with high accuracy, meanwhile offering identification and the localization of significant QEEG features. These important findings and the proposed classification framework could be used for the development of a biomarker for the diagnosis and monitoring of disease progression in AD. 27 5 826 835

Statistical analysis using large-scale brain magnetic resonance (MR) image databases has examined that brain tissues have age-related morphological changes. The age of a subject can be estimated from the brain MR image by evaluating morphological changes with healthy aging. This paper proposes an age estimation method using local features of T1-weighted MR images. The brain local features are defined by volumes of brain tissues parcellated into 1,024 local regions defined by the automated anatomical labeling atlas. This paper also proposes the effective local feature selection method to improve the accuracy of age estimation. We evaluate the accuracy of the proposed method using 1,099 T1-weighted images from a Japanese MR image database. We also analyze effectiveness of each local region for age estimation and discuss its medical implication.

This paper considers the auditory attention detection (AAD) paradigm, where the goal is to determine which of two simultaneous speakers a person is attending to. The paradigm relies on recordings of the listener's brain activity, e.g., from electroencephalography (EEG). To perform AAD, decoded EEG signals are typically correlated with the temporal envelopes of the speech signals of the separate speakers. In this paper, we study how the inclusion of various degrees of auditory modelling in this speech envelope extraction process affects the AAD performance, where the best performance is found for an auditory-inspired linear filter bank followed by power law compression. These two modelling stages are computationally cheap, which is important for implementation in wearable devices, such as future neuro-steered auditory prostheses. We also introduce a more natural way to combine recordings (over trials and subjects) to train the decoder, which reduces the dependence of the algorithm on regularization parameters. Finally, we investigate the simultaneous design of the EEG decoder and the audio subband envelope recombination weights vector using either a norm-constrained least squares or a canonical correlation analysis, but conclude that this increases computational complexity without improving AAD performance. ispartof: IEEE Transactions on Neural Systems and Rehabilitation Engineering vol:25 issue:2 pages:1-11 ispartof: location:United States status: published

The design of communication systems based on the transmission of information through neurons is envisioned as a key technology for the pervasive interconnection of future wearable and implantable devices. While previous literature has mainly focused on modeling propagation of electrochemical spikes carrying natural information through the nervous system, in recent work the authors of this paper proposed the so-called subthreshold electrical stimulation as a viable technique to propagate artificial information through neurons. This technique promises to limit the interference with natural communication processes, and it can be successfully approximated with linear models. In this paper, a novel model is proposed to account for the subthreshold stimuli propagation from the dendritic tree to the soma of a neuron. A computational approach is detailed to obtain this model for a given realistic 3D dendritic tree with an arbitrary morphology. Numerical results from the model are obtained over a stimulation signal bandwidth of 1KHz, and compared with the results of a simulation through the NEURON software.

Perimetry is essential for identifying visual field defects due to disorders of the eye and brain. However, young children are often unable to reliably perform the preferred method of visual field assessment known as automated static perimetry (ASP). This paper introduces a novel method of ASP specifically developed for children called Saccadic Vector Optokinetic Perimetry (SVOP). SVOP uses eye tracking to detect the natural saccadic eye response of gaze orientation towards visual field stimuli if they are seen. In this paper, the direction and magnitude of a sample of subject gaze responses to visual field stimuli is used to construct a software decision algorithm for use in SVOP. SVOP was clinically evaluated in a group of 24 subjects, comprising children and adults, with and without visual field defects, by comparison with an equivalent test on the Humphrey Field Analyser (HFA). SVOP provides promising visual field test results when compared with the reference HFA test, and has proven extremely useful in detecting visual field defects in children unable to perform traditional ASP.

In this paper, particle swarm optimization (PSO) algorithm is used to estimate the permittivities of the tissue layers at microwave frequency band. According to the literature, microwave radiometry (MWR) is potentially a promising cancer detection technique. In addition, breast cancer is an appropriate candidate of MWR due to the breast's exclusive physiology. Several algorithms have been evaluated for analyzing the measurement data and solving the inverse scattering problem in MWR, and different levels of accuracy have been reported. In this paper, the potential of PSO in solving this problem is demonstrated at 1-2.25 GHz. Two distinct algorithms are developed for the two considered scenarios. In the first scenario, we assume no a priori knowledge of the tissue under the test, whereas, in the second scenario, a priori knowledge is assumed. It is noteworthy that, there are only a few research articles studying PSO for permittivity estimation. However, since these studies underestimate the loss encountered by the test samples, the methods are not valid for body tissue case. Here, measurement-based loss coefficients, reported in the existing literature, are included in the calculations. It is shown that the algorithm converges relatively fast, and, distinguishes between different tissues with an acceptable accuracy.

This paper presents a time-varying cortical connectivity analysis for to a source localization approach within the inner layers of the newborn brain. (Additional details can be found in the comprehensive book on Time-Frequency Signal Analysis and Processing (see http://www.elsevier.com/locate/isbn/0080443354). In addition, the most recent upgrade of the original software package that calculates Time-Frequency Distributions and Instantaneous Frequency estimators can be downloaded from the web site: www.time-frequency.net. This was the first software developed in the field, and it was first released publicly in 1987 at the 1st ISSPA conference held in Brisbane, Australia, and then continuously updated). Multivariate Granger causality in the timefrequency domain as a representation of time-varying cortical connectivity in the brain has been investigated for the adult case. This is, however, not the case in newborns as the nature of the transient changes in the newborn EEG is different from that of adults. This paper aims to evaluate the performance of the time-varying versions of the two popular Granger causality measures, namely Partial Directed Coherence (PDC) and direct Directed Transfer Function (dDTF). The parameters of the time-varying AR, that models the inter-channel interactions, are estimated using Dual Extended Kalman Filter (DEKF) as it accounts for both non-stationarity and non-linearity behaviors of the EEG. Using simulated data, we show that fast changing cortical connectivity between channels can be measured more accurately using the time-varying PDC. The performance of the time-varying PDC is also tested on a neonatal EEG exhibiting seizure.

The ballistocardiogram (BCG) signal represents the movements of the body in response to cardiac ejection of blood. Recently, many groups have developed low-cost instrumentation for facilitating BCG measurement in the home. The standard method used in the literature for estimating the BCG pulse response has generally been ensemble averaging over several beats. Unfortunately, since the BCG pulse response is likely longer than a typical heartbeat interval, this standard approach does not yield a full-length estimate of the response. This paper describes a simple, novel algorithm for estimating the full-length BCG pulse response using the R-wave timing of a simultaneously acquired electrocardiogram (ECG). With this pulse response, the full signal can be reconstructed, enabling the analysis of slow transient effects in the BCG signal, and of the measurement noise. Additionally, while this paper focuses only on the BCG signal, the same algorithm could be applied to other biomedical signals such as the phonocardiogram or impedance cardiogram, particularly when the heartbeat interval is shorter than the duration of the cpulse response.

Biological ion channels are water filled pores in the cell membrane. They regulate the flow of ions in and out of the cell. Modeling the dynamics of these channels and relating their structure to functionality is crucial in understanding the mechanisms by which they conduct. This paper proposes a novel Finite Element Method (FEM) based simulation framework for modeling of ion channels that does not assume channel symmetry. This is the first framework that allows the use of multiple dielectric constants inside such channels without assuming geometrical symmetry thus providing a more realistic model of the channel. Due to the run-time complexity of the problem, lookup tables must be constructed in memory to store pre-calculated electric potential information. The large number of elements involved in FEM and channel resolution requirements can potentially result in very large lookup tables leading to a performance "bottleneck". This paper answers the following question: Does the accuracy introduced by using an asymmetric model outweigh the inaccuracy caused by having to reduce the size and resolution of electric-field look-up tables? This paper compares the memory footprint of an ion channel simulator that assumes a symmetric channel model versus an asymmetric model. We show that currently available personal computers are sufficient for attaining reasonable levels of accuracy for both. Our results show diminishing returns in accuracy with tables sized greater than 8.5 GB for the asymmetric model.

Dynamic infrared imaging is a promising technique in breast oncology. In this paper, a quantum well infrared photodetector infrared camera is used to acquire a sequence of consecutive thermal images of the patientpsilas breast for 10 s. Information on the local blood perfusion is obtained from the spectral analysis of the time series at each image pixel. Due to respiratory and motion artifacts, the direct comparison of the temperature values that a pixel assumes along the sequence becomes difficult. In fact, the small temperature changes due to blood perfusion, of the order of 10-50 mK, which constitute the signal of interest in the time domain, are superimposed onto large temperature fluctuations due to the subjectpsilas motion, which represent noise. To improve the time series S/N, and as a consequence, enhance the specificity and sensitivity of the dynamic infrared examination, it is important to realign the thermal images of the acquisition sequence, thus reducing motion artifacts. In a previous study, we demonstrated that a registration algorithm based on fiducial points is suitable to both clinical applications and research, when associated with a proper set of skin markers. In this paper, we quantitatively evaluate the performance of different marker sets by means of a model that allows for estimating the S/N increment due to registration, and we conclude that a 12-marker set is a good compromise between motion artifact reduction and the time required to prepare the patient.

Closed Angle Glaucoma causes an increase in fluid pressure in the eye due to decreased fluid flow between the iris and the cornea. To diagnose Glaucoma, ultrasound images of the eye are taken and a technician studies the images and takes measurements. This process is time consuming, and results vary among technicians. In a previous paper, we developed an image processing algorithm to extract several key landmark features in the iris. In this paper, we extend this algorithm to calculate the complete set of diagnostic parameters from ultrasound biomicroscopy. A robust algorithm is described to calculate the parameters which describe the thickness of iris; ID1, ID2, and ID3. The method is tested on 89 clinical images and compared to the values identified by the technician. Results indicate good agreement in 97% of cases.

The number of people that now go on to develop Alzheimer's disease (AD) and other types of dementia is rapidly rising. For maximum benefits from new treatments, the disease should be diagnosed as early as possible, but this is difficult with current clinical criteria. Potentially, the EEG can serve as an objective, first line of decision support tool to improve diagnosis. It is non-invasive, widely available, low-cost and could be carried out rapidly in the high-risk age group that will develop AD. Changes in the EEG due to the dementing process could be quantified as an index or marker. In this paper, we investigate two information theoretic methods (Tsallis entropy and universal compression algorithm) as a way to generate potentially robust markers from the EEG. The hypothesis is that the information theoretic makers for AD are significantly different to those of normal subjects. An attraction of the information theoretic approach is that, unlike most existing methods, there may be a natural link between the underlying ideas of information theoretic methods, the physiology of AD and its impact on brain functions. Data compression has not been investigated as a means of generating EEG markers before and is attractive because it does not require a priori knowledge of the source model. In this paper, we focus on the LZW algorithm because of its sound theoretical foundation. We used the LZW algorithm and Tsallis model to compute the markers (compression ratios and normalized entropies, respectively) from two EEG datasets. The results show that the information theoretic methods can be used to compute EEG markers for AD.

An electrocardiogram (ECG) is an important and commonly used diagnostic aid in cardiovascular disease diagnosis. Physicians routinely perform diagnosis by a simple visual examination of ECG waveform shapes. In this paper, we address the problem of shape-based retrieval of ECG recordings, both digital and scanned from paper, to infer similarity in diagnosed diseases. Specifically, we use the knowledge of ECG recording structure to segment and extract curves representing various recording channels from ECG images. We then present a method of capturing the perceptual shape similarity of ECG waveforms by combining shape matching with dynamic time warping. The shape similarity of each recording channel is combined to develop an overall shape similarity measure between ECG recordings. Results are presented that demonstrate the method on shape-based matching of various cardiovascular diseases.

In electrical impedance tomography, currents are applied to the body through electrodes that are attached to the surface and the corresponding surface voltages are measured. Based on these boundary measurements, the internal admittivity distribution of the body can be reconstructed. In order to improve the image quality it is necessary and useful to apply physiologically meaningful prior information into the image reconstruction. Such prior information usually can be obtained from other sources. For example, information on the object's boundary shape and internal structure can be obtained from computed tomography and magnetic resonance imaging scan. However, this type of prior information may change from time to time and from person to person. As these changes are limited anatomically and physiologically, the prior information including the possible changes can be presented in a number of variational forms. The aim of this paper is to find which form of prior information is more compatible for a specific imaged object at the time of imaging. This paper proposes a new method for selecting the most appropriate form of prior information, through the procedure of iterative image reconstruction by using the information obtained from boundary measurements. The method is based on the principle that incompatible prior information causes errors which are able to affect the image reconstruction's convergence behavior. In this method, according to the various forms of prior information available, several image reconstruction configurations are designed. Then, through monitoring the convergence behavior in an iterative image reconstruction, the configuration with compatible prior information can be found among those different configurations. As an example, the prior information regarding the imaged object's boundary shape and internal structure was studied by computer simulation. Results were shown and discussed.

Magnetic resonance fingerprinting is a recent quantitative MRI technique that simultaneously acquires multiple tissue parameter maps (e.g., T(1), T(2), and spin density) in a single imaging experiment. In our early work, we demonstrated that the low-rank/subspace reconstruction significantly improves the accuracy of tissue parameter maps over the conventional MR fingerprinting reconstruction that utilizes simple pattern matching. In this paper, we generalize the low-rank/subspace reconstruction by introducing a multilinear low-dimensional image model (i.e., a low-rank tensor model). With this model, we further estimate the subspace associated with magnetization evolutions to simplify the image reconstruction problem. The proposed formulation results in a nonconvex optimization problem which we solve by an alternating minimization algorithm. We evaluate the performance of the proposed method with numerical experiments, and demonstrate that the proposed method improves the conventional reconstruction method and the state-of-the-art low-rank reconstruction method.

Laparoscopic cholecystectomy surgery is a minimally invasive surgery to remove the gallbladder, where surgical instruments are inserted through small incisions in the abdomen with the help of a laparoscope. Identification of tool presence and precise segmentation of tools from the video is very important in understanding the quality of the surgery and training budding surgeons. Precise segmentation of tools is required to track the tools during real-time surgeries. In this paper, a new pixel-wise instance segmentation algorithm is proposed, which segments and localizes the surgical tool using spatio-temporal deep network. The performance of the proposed has been compared with the state-of-the-art image-based instance segmentation method using the Cholec80 dataset. It is also compared with methods in the literature using frame-level presence detection and spatial detection with good results.

The phase-amplitude coupling in EEG signal of different frequencies is considered as a useful biomarker in delineating epileptogenic tissues, but some physiological processes can also generate phase-amplitude coupling pattern, such as memory process. Current analysis on cross-frequency coupling (CFC) feature is mostly based on extracting the strength of coupling but not coupling patterns in frequency-frequency domain. In this paper, we proposed a method for identifying epileptogenic tissue using convolutional neural networks (CNN) based on CFC pattern. Stereo-electroencephalograph (SEEG) from six patients with intractable epilepsy were used in this analysis. First, modulation indexes (MIs) were calculated using a moving window for each channel across seizures. Then those MIs were marked as inside epileptogenic zone (EZ) or outside EZ based on the surgical resection area. CNN was trained by those two-dimensional coupling patterns and tested by leave-one-out method. The receiver operating characteristics (ROC) curve was further generated. The results showed that average area-under-curve (AUC) performance reached 0.88. The sensitivity was 0.81, and the specificity was 0.79. Those results suggest that the CFC pattern can be used to identify SEEG channels in the epileptogenic region using the CNN.Clinical Relevance- This method has the potential to be used as an analytical tool for neurologists to identify epileptogenic brain tissues.

A large amount of calibration data is typically needed to train an electroencephalogram (EEG)-based brain-computer interfaces (BCI) due to the non-stationary nature of EEG data. This paper proposes a novel weighted transfer learning algorithm using a dynamic time warping (DTW) based alignment method to alleviate this need by using data from other subjects. DTW-based alignment is first applied to reduce the temporal variations between a specific subject data and the transfer learning data from other subjects. Next, similarity is measured using Kullback Leibler divergence (KL) between the DTW aligned data and the specific subject data. The other subjects' data are then weighted based on their KL similarity to each trials of the specific subject data. This weighted data from other subjects are then used to train the BCI model of the specific subject. An experiment was performed on publicly available BCI Competition IV dataset 2a. The proposed algorithm yielded an average improvement of 9% compared to a subject-specific BCI model trained with 4 trials, and the results yielded an average improvement of 10% compared to naive transfer learning. Overall, the proposed DTW-aligned KL weighted transfer learning algorithm show promise to alleviate the need of large amount of calibration data by using only a short calibration data.

This paper proposes a new method for automatic detection of glaucoma from stereo pair of fundus images. The basis for detecting glaucoma is using the optic cup-to-disc area ratio, where the surface area of the optic cup is segmented from the disparity map estimated from the stereo fundus image pair. More specifically, we first estimate the disparity map from the stereo image pair. Then, the optic disc is segmented from one of the stereo image. Based upon the location of the optic disc, we perform an active contour segmentation on the disparity map to segment the optic cup. Thereafter, we can compute the optic cup-to-disc area ratio by dividing the area (i.e. the total number of pixels) of the segmented optic cup region to that of the segmented optic disc region. Our experimental results using the available test dataset shows the efficacy of our proposed approach.

Early diagnosis of mild traumatic brain injury (mTBI) is challenging, yet significantly important in order to grant the patients with timely treatment and mitigating the risks of possible long-term psychiatric and neurological disorders. To tackle this problem, in this paper, we develop an mTBI detection framework based on graph embedding features combined with convolutional neural networks (CNN). Cortical activity in transgenic calcium reporter mice expressing Thy1-GCaMP6s is recorded in two sessions, prior to and after inducing injury. Functional networks are then constructed for recordings obtained in each session. The Node2vec algorithm is employed to represent nodes of these networks in the node embedding space. Node embedding feature vectors are then aligned, compressed, and represented as three-channel images. A CNN model is used for the classification of brain networks into two categories of normal and mTBI. A maximum classification accuracy of 95.4% is achieved. Our results suggest that functional networks as biomarkers along with the proposed method can effectively be used for detecting mTBI.

Recent years have seen a growing interest in the development of non-invasive devices capable of detecting seizures which can be worn in everyday life. Such devices must be lightweight and unobtrusive which severely limit their on-board computing power and battery life. In this paper, we propose a novel technique based on hyperdimensional (HD) computing to detect epileptic seizures from 2-channel surface EEG recordings. The proposed technique eliminates the need for complicated feature extraction techniques required in conventional ML algorithms. The HD algorithm is also simple to implement and does not require expert knowledge for architectural optimizations needed for approaches based on neural networks. In addition, our proposed technique is light-weight and meets the computation and memory constraints of ultra-small devices. Experimental results on a publicly available dataset indicates our approach improves the accuracy compared to state-of-the-art techniques while consuming smaller or comparable power.

The scope and relevance of wearable robotics spans across a number of research fields with a variety of applications. A challenge across these research areas is improving user-interface control. One established approach is using neural control interfaces derived from surface electromyography (sEMG). Although there has been some success with sEMG controlled prosthetics, the coarse nature of traditional sEMG processing has limited the development of fully functional prosthetics and wearable robotics. To solve this problem, blind source separation (BSS) techniques have been implemented to extract the user's movement intent from high-density sEMG (HDsEMG) measurements; however, current methods have only been well validated during static, low-level muscle contractions, and it is unclear how they will perform during movement. In this paper we present a neural drive based method for predicting output torque during a constant force, concentric contraction. This was achieved by modifying an existing HDsEMG decomposition algorithm to decompose 1 sec. overlapping windows. The neural drive profile was computed using both rate coding and kernel smoothing. Neither rate coding nor kernel smoothing performed as well as HDsEMG amplitude estimation, indicating that there are still significant limitations in adapting current methods to decompose dynamic contractions, and that sEMG amplitude estimation methods still remain highly reliable estimators.

Electrocardiograph (ECG) is one of the most critical physiological signals for arrhythmia diagnosis in clinical practice. In recent years, various algorithms based on deep learning have been proposed to solve the heartbeat classification problem and achieved saturated accuracy in intrapatient paradigm, but encountered performance degradation in inter-patient paradigm due to the drastic variation of ECG signals among different individuals. In this paper, we propose a novel unsupervised domain adaptation scheme to address this problem. Specifically, we first propose a robust baseline model called Multi-path Atrous Convolutional Network (MACN) to tackle ECG heartbeat classification. Further, we introduce Cluster-aligning loss and Cluster-separating loss to align the distributions of training and test data and increase the discriminability, respectively. The proposed method requires no expert annotations but a short period of unlabelled data in new records. Experimental results on the MIT-BIH database demonstrate that our scheme effectively intensifies the baseline model and achieves competitive performance with other state-of-the-arts.

One crucial key of developing an automatic sleep stage scoring method is to extract discriminative features. In this paper, we present a novel technique, termed common frequency pattern (CFP), to extract the variance features from a single-channel electroencephalogram (EEG) signal for sleep stage classification. The learning task is formulated by finding significant frequency patterns that maximize variance for one class and that at the same time, minimize variance for the other class. The proposed methodology for automated sleep scoring is tested on the benchmark Sleep-EDF database and finally achieves 97.9%, 94.22%, and 90.16% accuracy for two-state, three-state, and five-state classification of sleep stages. Experimental results demonstrate that the proposed method identifies discriminative characteristics of sleep stages robustly and achieves better performance as compared to the state-of-the-art sleep staging algorithms. Apart from the enhanced classification, the frequency patterns that are determined by the CFP algorithm is able to find the most significant bands of frequency for classification and could be helpful for a better understanding of the mechanisms of sleep stages.

Cerebral Microbleeds (CMBs) are small chronic brain hemorrhages, which have been considered as diagnostic indicators for different cerebrovascular diseases including stroke, dysfunction, dementia, and cognitive impairment. In this paper, we propose a fully automated two-stage integrated deep learning approach for efficient CMBs detection, which combines a regional-based You Only Look Once (YOLO) stage for potential CMBs candidate detection and three-dimensional convolutional neural networks (3D-CNN) stage for false positives reduction. Both stages are conducted using the 3D contextual information of microbleeds from the MR susceptibility-weighted imaging (SWI) and phase images. However, we average the adjacent slices of SWI and complement the phase images independently and utilize them as a two- channel input for the regional-based YOLO method. The results in the first stage show that the proposed regional-based YOLO efficiently detected the CMBs with an overall sensitivity of 93.62% and an average number of false positives per subject (FP avg ) of 52.18 throughout the five-folds cross-validation. The 3D-CNN based second stage further improved the detection performance by reducing the FP avg to 1.42. The outcomes of this work might provide useful guidelines towards applying deep learning algorithms for automatic CMBs detection.

Ultrasound elastography is used to estimate the mechanical properties of the tissue by monitoring its response to an internal or external force. Different levels of deformation are obtained from different tissue types depending on their mechanical properties, where stiffer tissues deform less. Given two radio frequency (RF) frames collected before and after some deformation, we estimate displacement and strain images by comparing the RF frames. The quality of the strain image is dependent on the type of motion that occurs during deformation. In-plane axial motion results in high-quality strain images, whereas out-of-plane motion results in low-quality strain images. In this paper, we introduce a new method using a convolutional neural network (CNN) to determine the suitability of a pair of RF frames for elastography in only 5.4 ms. Our method could also be used to automatically choose the best pair of RF frames, yielding a high-quality strain image. The CNN was trained on 3,818 pairs of RF frames, while testing was done on 986 new unseen pairs, achieving an accuracy of more than 91%. The RF frames were collected from both phantom and in vivo data., Comment: 2020 42nd Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC)

This paper presents a new asynchronous hybrid brain-computer interface (BCI) system that integrates a speller, a web browser, an e-mail client, and a file explorer using electroencephalographic (EEG) and electrooculography (EOG) signals. More specifically, an EOG-based button selection method, which requires the user to blink his/her eyes synchronously with the target button's flashes during button selection, is first presented. Next, we propose a mouse control method by combining EEG and EOG signals, in which the left-/right-hand motor imagery (MI)-related EEG is used to control the horizontal movement of the mouse and the blink-related EOG is used to control the vertical movement of the mouse and to select/reject a target. These two methods are further combined to develop the integrated hybrid BCI system. With the hybrid BCI, users can input text, access the internet, communicate with others via e-mail, and manage files in their computer using only EEG and EOG without any body movements. Ten healthy subjects participated in a comprehensive online experiment, and superior performance was achieved compared with our previously developed P300- and MI-based BCI and some other asynchronous BCIs, therefore demonstrating the system's effectiveness.

When generating automatic sleep reports with mobile sleep monitoring devices, it is crucial to have a good grasp of the reliability of the result. In this paper, we feed features derived from the output of a sleep scoring algorithm to a 'regression ensemble' to estimate the quality of the automatic sleep scoring. We compare this estimate to the actual quality, calculated using a manual scoring of a concurrent polysomnography recording. We find that it is generally possible to estimate the quality of a sleep scoring, but with some uncertainty ('root mean squared error' between estimated and true Cohen's kappa is 0.078). We expect that this method could be useful in situations with many scored nights from the same subject, where an overall picture of scoring quality is needed, but where uncertainty on single nights is less of an issue.

Retinopathy screening is a non-invasive method to collect retinal images and neovascularization detection from retinal images plays a significant role on the identification and classification of diabetes retinopathy. In this paper, an automatic parallel detection framework for neovascularization with color retinal images using ensemble of extreme learning machine is proposed. The framework employs two Map-Reduce Jobs to extract features and trains Extreme Learning Machine models. Ensemble methods such as bagging, subspace partitioning and cross validating are used to increase the accuracy. The framework is evaluated with retinal images from MESSIDOR database. Experimental results show the framework can improve the detection accuracy, as well as speedup the processing time to 22 times on average.

In this paper we propose a sensor embedded knee brace to monitor knee flexion and extension and other lower limb joint kinematics after anterior cruciate ligament (ACL) injury. The system can be easily attached to a standard post-surgical brace and uses a novel sensor fusion algorithm that does not require calibration. The wearable system and the sensor fusion algorithm were validated for various physical therapy exercises against a validated motion capture system. The proposed sensor fusion algorithm demonstrated significantly lower root-mean-square error (RMSE) than the benchmark Kalman filtering algorithm and excellent correlation coefficients (CCC and ICC). The demonstrated error for most exercises was lower than other devices in the literature. The quantitative measures obtained by this system can be used to obtain longitudinal range-of-motion and functional biomarkers. These biomarkers can be used to improve patient outcomes through the early detection of at-risk patients, tracking patient function outside of the clinic, and the identification of relationships between patient presentation, intervention, and outcomes.

Despite the severe consequences of dysfunctional swallowing, there is no simple method of monitoring swallowing outside of clinical settings. People who cannot swallow cannot eat safely, resulting in profound changes in quality of life and risk of death from aspiration pneumonia. A non-invasive swallowing detector may have widespread impact in both clinical care and research. Detection of swallowing from laryngeal sounds could become an ideal assessment tool because sounds are simple to record, quantifiable, and amenable to software analysis. The focus of this paper is to achieve high accuracy binary swallowing detection from sound recordings. A dataset with 2500 swallow sound samples and 1700 mixed laryngeal noise samples from 15 healthy adults was used to train and test three supervised machine learning algorithms. A decision tree, support vector machine (SVM), and neural network trained with the scaled conjugate gradient (SCG) method had areas under the receiver operating characteristic (ROC) curve of 0.970, 0.961, and 0.971 and average accuracies of 93.2 percent, 86.2 percent, and 93.7 percent respectively. While further work needs to be done to further optimize these algorithms and validate their efficacy, these initial results suggest machine learning strategies may be helpful to improve accuracy of swallowing detection.

Dynamic positron emission tomography (dPET) is known for its ability to extract spatiotemporal information of a radio tracer in living tissue. In this paper, a novel direct reconstruction framework is presented, which include concurrent clustering as a potential aid in addressing high levels of noise typical of voxel-wise kinetic modeling. Core assumption is that the imaged volume is formed by a finite number of different functional regions, and that voxel-wise time courses are determined by the functional cluster they belong to. Probabilistic Graphical Modeling (PGM) theory is used to describe the problem, and to derive the inference strategy. The proposed iterative estimation scheme provides concurrent estimate of kinetic parameter maps, activity images, and segmented clusters. Simulation studies and exploratory application to real data are performed to validate the proposal.

Cough is a common symptom that manifests in numerous respiratory diseases. In chronic respiratory diseases, such as asthma and COPD, monitoring of cough is an integral part in managing the disease. This paper presents an algorithm for automatic detection of cough events from acoustic signals. The algorithm uses only three spectral features with a logistic regression model to separate sound segments into cough and non-cough events. The spectral features were derived using simple calculation from two frequency bands of the sound spectrum. The frequency bands of interest were chosen based on its characteristics in the spectrum. The algorithm achieved high sensitivity of 90.31%, specificity of 98.14%, and F1-score of 88.70%. Its low-complexity and high detection performance demonstrate its potential for use in remote patient monitoring systems for real-time, automatic cough detection.

The irreversible damage and eventual heart failure caused by untreated aortic stenosis (AS) can be prevented by early detection and timely intervention. Prior work in the field of phonocardiogram (PCG) signal analysis has provided proof of concept for using heart-sound data in AS diagnosis. However, such systems either require operation by trained technicians, fail to address a diverse subject set, or involve unwieldy configuration procedures that challenge real-world application. This paper presents an end-to-end, fully-automated system that uses noise-subtraction, heartbeat-segmentation and quality-assurance algorithms to extract physiologically-motivated features from PCG signals to diagnose AS. When tested on n=96 patients showing a diverse set of cardiac and non-cardiac conditions, the system was able to diagnose AS with 92% sensitivity and 95% specificity.

Ambulatory Photoplethysmogram (PPG) is a more user-friendly choice for continuous cardiac monitoring as compared to Electrocardiogram (ECG). However, wearable PPG is often prone to motion artefacts. In this paper, we propose a novel pipeline for motion-resistant beat-to-beat interval extraction from noisy PPG. Firstly, the effects of motion artefacts are minimized by using Adaptive Recursive-Least-Square (RLS) Filtering and Singular Spectrum Analysis (SSA). Next, the signal peaks are identified and their locations are corrected by weighted local interpolation. Finally, outlier peak-to-peak intervals are marked as incorrigible. Experimental validation on the training dataset of IEEE Signal Processing Cup 2015 reveals that the proposed method achieves 1.68% mean peak detection error rate and 11.32 milliseconds mean absolute error of detected beat-to-beat intervals. The metric values outperform those obtained by the state-of-the-art techniques by at least 12.58 and 5.74 times respectively.

The most common brain-computer interface (BCI) devices use electroencephalography (EEG). EEG signals are noisy owing to the presence of many artifacts, namely head movement, and facial movements like eye blinks or jaw movements. Removal of these artifacts from EEG signals is essential for the success of any downstream BCI application. These artifacts influence different sensors of the EEG. In this paper, we devise algorithms for detection and classification of artifacts. Classification of artifacts into head nod, jaw movement and eye-blink is performed using two different varieties of time warping; namely, linear time warping, and dynamic time warping. The average accuracy of 85% and 90% is obtained using the former, and the later, respectively.

White matter hyperintensity (WMH) is associated with various aging and neurodegenerative diseases. In this paper, we proposed and validated a fully automatic system which integrated classical image processing and deep neural network for segmenting WMH from fluid attenuation inversion recovery (FLAIR) and T1-weighed magnetic resonance (MR) images. A novel skip connection U-net (SC U-net) was proposed and compared with the classical U-net. Experiments were performed on a dataset of 60 images, acquired from three different scanners. Validation analysis and cross-scanner testing were conducted. Compared with U-net, the proposed SC U-net had a faster convergence and higher segmentation accuracy. The software environment and models of the proposed system were made publicly accessible at Dockerhub.

This paper presents the design of a machine learning-based classifier for the differentiation between Schizophrenia patients and healthy controls using features extracted from electroencephalograph(EEG) signals based on event related potential(ERP). A number of features are extracted from an online EEG dataset with 81 subjects, including 32 healthy controls and 49 Schizophrenia patients. The EEG signals are preprocessed and since the dataset is relatively small, the random forest machine learning algorithm is chosen to be applied on different combinations of feature sets for classification. It is found that the classification accuracy can be improved by adding certain features extracted from EEG signals.

Digital pathology can be thought of as a model composed of 3 main elements; classification algorithm, Graphical User Interface (GUI) and the pathologists. Currently there is only a one way interaction from the classification algorithm to the pathologist. This paper, proposes an additional backward path which is a new feedback-based method, aimed to improve the performance of the classification algorithms by utilizing the feedback of the pathologists. The GUI developed for this purpose, is aimed to be simple and adaptive to different classification algorithms. The method showed significant improvement in the classification performance of the applied Convolutional Neural Network (CNN) algorithm. The 25% quantile of the probability score of the predictions increased from 0.48 to 0.89 and the median of the data increased from 0.95 to 0.99.

3D Ultrasound imaging has become an important means of scoliosis assessment as it is a real-time, cost-effective and radiation-free imaging technique. However, the coronal images from different depths of a 3D ultrasound image have different imaging definitions. So there is a need to select the coronal image that would give the best image definition. Also, manual selection of coronal images is time-consuming and limited to the discretion and capability of the assessor. Therefore, in this paper, we have developed a convolution learning-to-rank algorithm to select the best ultrasound images automatically using raw ultrasound images. The ranking is done based on the curve angle of the spinal cord. Firstly, we approached the image selection problem as a ranking problem; ranked based on probability of an image to be a good image. Here, we use the RankNet, a pairwise learning-to-rank method, to rank the images automatically. Secondly, we replaced the backbone of the RankNet, which is the traditional artificial neural network (ANN), with convolution neural network (CNN) to improve the feature extracting ability for the successive iterations. The experimental result shows that the proposed convolutional RankNet achieves the perfect accuracy of 100% while conventional DenseNet achieved 35% only. This proves that the convolutional RankNet is more suitable to highlight the best quality of ultrasound image from multiple mediocre ones.

In this paper, we extracted hand crafted features from the ECG signals and evaluated the performance of different combination of features for sleep apnoea detection. We calculated the ECG derived respiratory (EDR) signal using three methods (QRS area, amplitude demodulation and fast PCA methods) and then calculated the cardiopulmonary coupling (CPC) spectrum using each EDR method. We then extracted features from the CPC spectrums and the time and frequency representations of the heart rate variability (HRV) and EDR signals Then, we compared the performance results of different combinations of the features used for automated sleep apnoea detection. We also applied a temporal optimisation method by averaging the features of every three adjacent epochs. Two classifiers were used to detect sleep apnoea: the extreme learning machine (ELM), and linear discriminant analysis. The features were evaluated on the MIT PhysioNet Apnea-ECG database. Apnoea detection was evaluated with leave-one-record-out cross-validation. The PCA CPC features obtained the highest accuracy of 86.5% and AUC of 0.94 using LDA classifier. The performance results of the combined features (of PCA method) obtained the same results. We conclude that for this study, the CPC features using fast PCA method are our best feature set for sleep apnoea detection.

Independent component analysis (ICA), as a data driven method, has shown to be a powerful tool for functional magnetic resonance imaging (fMRI) data analysis. One drawback of this multivariate approach is, that it is naturally not convenient for analysis of group studies. Therefore various techniques have been proposed in order to overcome this limitation of ICA. In this paper a novel ICA based work-flow for extracting resting state networks from fMRI group studies is proposed. An empirical mode decomposition (EMD) is used to generate reference signals in a data driven manner, which can be incorporated into a constrained version of ICA (cICA), what helps to overcome the inherent ambiguities. The results of the proposed workflow are then compared to those obtained by a widely used group ICA approach. It is demonstrated that intrinsic modes, extracted by EMD, are suitable to serve as references for cICA to obtain typical resting state patterns, which are consistent over subjects. This novel processing pipeline makes it transparent for the user, how comparable activity patterns across subjects emerge, and also the trade-off between similarity across subjects and preserving individual features can be well adjusted and adapted for different requirements in the new work-flow.

Optical coherence tomographic (OCT) images provide valuable information for understanding the changes occurring in the retina due to glaucoma, specifically, related to the retinal nerve fiber layer and the optic nerve head. In this paper, we propose a deep learning approach using Capsule network for glaucoma classification, which directly operates on 3D OCT volumes. The network is trained only on labelled volumes and does not attempt any region/structure segmentation. The proposed network was assessed on 50 volumes and found to achieve 0.97 for the area under the ROC curve (AUC). This is considerably higher than the existing approaches which are majorly based on machine learning or rely on segmentation of the required structures from OCT. Our network also outperforms 3D convolutional neural networks despite the fewer network parameters and fewer epochs needed for training.

Steady-State Visual Evoked Potentials (SSVEP) based Brain-Computer Interface (BCI) has achieved very high information transmission rate (ITR), but its portability and fundamental interactions with the surrounding environment were limited. The combination of Augmented Reality (AR) and BCI is expected to solve these problems. In this paper, we combined AR with the SSVEP-BCI to build a more portable and natural BCI system in Microsoft HoloLens. We designed the AR-BCI system and studied the influence of different algorithms on the system performance. The analysis of SSVEP signals collected in AR environment shows that the extended filter bank canonical correlation analysis was better than task-related component analysis. The average recognition accuracy and ITR obtained by using Electroencephalography (EEG) data of 1s, 1.5s, and 2s length were 87.7%、95.4%、 97.6% and 64.6 bit/min 、 62.9 bit/min 、 55.6 bit/min, respectively. Compared with the existing AR-BCI studies, the ITR has been greatly improved in this study.

This paper presents a novel approach to monitor office workers' behavioral patterns and heart rate variability. We integrated an EMFi sensor into a chair to measure the pressure changes caused by a user's body movements and heartbeat. Then, we employed machine learning methods to develop a classification model through which different work behaviors (body moving, typing, talking and browsing) could be recognized from the sensor data. Subsequently, we developed a BCG processing method to process the data recognized as 'browsing' and further calculate heart rate variability. The results show that the developed model achieved classification accuracies of up to 91% and the HRV could be calculated effectively with an average error of 5.77ms. By combining these behavioral and physiological measures, the proposed approach portrays work-related stress in a more comprehensive manner and could contribute an unobtrusive early stress detection system for future smart offices.

Image reconstruction for positron emission tomography (PET) can be challenging and the resulting image typically has high noise. The kernel-based reconstruction method [1], incorporates prior anatomic information in the reconstruction algorithm to reduce noise while preserving resolution. Prior information is incorporated in the reconstruction algorithm by means of spatial kernels originally used in machine learning. In this paper, the kernel-based method is used to reconstruct PET images of sympathetic innervation in the heart. The resulting images are compared with standard Ordered Subset Expectation Maximization (OSEM) reconstructed images qualitatively and quantitatively using data from 6 human subjects. The kernel-based method demonstrated superior SNR with preserved contrast and accuracy compared to OSEM.