Category: IEEE Transactions on Biomedical Engineering

Background: Despite the tremendous prog- ress recently made towards automatic sleep staging in adults, it is currently unknown if the most advanced algorithms generalize to the pediatric population, which displays distinctive characteristics in overnight polysomnography (PSG). Methods: To answer the question, in this work, we conduct a large-scale comparative study on the state-of-the-art deep learning methods for pediatric automatic sleep staging. Six different deep neural networks with diverging features are adopted to evaluate a sample of more than 1,200 children across a wide spectrum of obstructive sleep apnea (OSA) severity. Results: Our experimental results show that the individual performance of automated pediatric sleep stagers when evaluated on new subjects is equivalent to the expert-level one reported on adults. Combining the six stagers into ensemble models further boosts the staging accuracy, reaching an overall accuracy of 88.8%, a Cohen’s kappa of 0.852, and a macro F1-score of 85.8%. At the same time, the ensemble models lead to reduced predictive uncertainty. The results also show that the studied algorithms and their ensembles are robust to concept drift when the training and test data were recorded seven months apart and after clinical intervention. Conclusion: However, we show that the improvements in the staging performance are not necessarily clinically significant although the ensemble models lead to more favorable clinical measures than the six standalone models. Significance: Detailed analyses further demonstrate “almost perfect” agreement between the automatic stagers to one another and their similar patterns on the staging errors, suggesting little room for improvement.

First order cutaneous neurons allow object recognition, texture discrimination, and sensorimotor feedback. Their function is well-investigated under passive stimulation while their role during active touch or sensorimotor control is understudied. To understand how human perception and sensorimotor controlling strategy depend on cutaneous neural signals under active tactile exploration, the finite element (FE) hand and Izhikevich neural dynamic model were combined to predict the cutaneous neural dynamics and the resulting perception during a discrimination test. Using in-vivo microneurography generated single afferent recordings, 75% of the data was applied for the model optimization and another 25% was used for validation. By using this integrated numerical model, the predicted tactile neural signals of the single afferent fibers agreed well with the microneurography test results, achieving the out-of-sample values of 0.94 and 0.82 for slowly adapting type I (SAI) and fast adapting type I unit (FAI) respectively. Similar discriminating capability with the human subject was achieved based on this computational model. Comparable performance with the published numerical model on predicting the cutaneous neural response under passive stimuli was also presented, ensuring the potential applicability of this multi-level numerical model in studying the human tactile sensing mechanisms during active touch. The predicted population-level 1st order afferent neural signals under active touch suggest that different coding strategies might be applied to the afferent neural signals elicited from different cutaneous neurons simultaneously.

Objective: The Auditory brainstem response (ABR) can provide valuable information on the function of the auditory pathway. However, the ABR signal has a very small amplitude, and it is easily submerged in different background noises with large amplitude. Conventional ABR extraction methods such as time-domain averaging (TDA) and Kalman filter (KF) were greatly affected by noise intensity, and the result relies on the empirical settings of parameters. ABR extraction method that can automatically adjust parameters to adapt different background noises was needed. Methods: An adaptive Kalman filtering (AKF) based ABR signal extraction method was proposed, in which two recursive rules were introduced to constantly update the parameters according to the real-time noise properties. It was used for ABR extraction from recordings in noises with different orders of larger magnitude. Results: The AKF method demonstrated the best performance in obtaining reliable ABR waveform morphologies in the presence of large EMG noises compared with traditional methods of TDA or KF. It could extract satisfactory ABR signal with fewer trials of acoustic stimulus repetition, even from noise 10000 times larger than ABR signal. The AKF results also showed smaller absolute errors and higher correlation coefficients with the target ABR signal when different types (gum chewing, mouth opening and milk drinking) or levels of noises were introduced. Conclusion: The proposed AKF method is a great candidate to increase the robustness of current ABR measurements. Significance: It could provide reduced testing time and relaxed recording conditions for ABR and other evoked potentials extraction.

Objective: Preterm birth is the leading cause of morbidity and mortality involving over 10% of infants. Tools for timely diagnosis of preterm birth are lacking and the underlying physiological mechanisms are unclear. The aim of the present study is to improve early assessment of pregnancy progression by combining and optimizing a large number of electrohysterography (EHG) features with a dedicated machine learning framework. Methods: A set of reported EHG features are extracted. In addition, novel cross and multichannel entropy and mutual information are employed. The optimal feature set is selected using a wrapper method according to the accuracy of the leave-one-out cross validation. An annotated database of 74 EHG recordings in women with preterm contractions was employed to test the ability of the proposed method to recognize the onset of labor and the risk of preterm birth. Difference between using the contractile segments only and the whole EHG signal was compared. Results: The proposed method produces an accuracy of 96.4% and 90.5% for labor and preterm prediction, respectively, much higher than that reported in previous studies. The best labor prediction was observed with the contraction segments and the best preterm prediction achieved with the whole EHG signal. Entropy features, particularly the newly-employed cross entropy contribute significantly to the optimal feature set for both labor and preterm prediction. Significance: Our results suggest that changes in the EHG, particularly the regularity, might manifest early in pregnancy. Single-channel and cross entropy may therefore provide relevant prognostic opportunities for pregnancy monitoring.

Objective: To explain the 0.2–2Hz oscillation in human balance. Motivation: Oscillation (0.2–2 Hz) in the control signal (ankle moment) is sustained independently of external disturbances and exaggerated in Parkinson's disease. Does resonance or limit cycles in the neurophysiological feedback loop cause this oscillation? We investigate two linear (non-predictive, predictive) and one non-linear (intermittent-predictive) control model (NPC, PC, IPC). Methods: Fourteen healthy participants, strapped to an actuated single segment robot with dynamics of upright standing, used natural haptic-visual feedback and myoelectric control signals from lower leg muscles to maintain balance. An input disturbance applied stepwise changes in external force. A linear time invariant model (ARX) extracted the delayed component of the control signal related linearly to the disturbance, leaving the remaining, larger, oscillatory non-linear component. We optimized model parameters and noise (observation, motor) to replicate concurrently (i) estimated-delay, (ii) time-series of the linear component, and (iii) magnitude-frequency spectrum and transient magnitude response of the non-linear component. Results (mean±S.D., p<0.05): NPC produced estimated delays (0.116±0.03s) significantly lower than experiment (0.145±0.04s). Overall fit (i)–(iii) was (79±7%, 83±7%, 84±6% for NPC, PC, IPC). IPC required little or no noise. Mean frequency of experimental oscillation (0.99±0.16 Hz) correlated trial by trial with closed loop resonant frequency (f_res), not limit cycles, nor sampling rate. NPC (0.36±0.08Hz) an- PC (0.86±0.4Hz) showed f_res significantly lower than IPC (0.98±0.2Hz). Conclusion: Human balance control requires short-term prediction. Significance: IPC mechanisms (prediction error, threshold related sampling, sequential re-initialization of open-loop predictive control) explain resonant gain without uncontrolled oscillation for healthy balance.

Objective: This study demonstrates intravascular micro-agent visualization by utilizing robotic ultrasound-based tracking and visual servoing in clinically-relevant scenarios. Methods: Visual servoing path is planned intraoperatively using a body surface point cloud acquired with a 3D camera and the vessel reconstructed from ultrasound (US) images, where both the camera and the US probe are attached to the robot end-effector. Developed machine vision algorithms are used for detection of micro-agents from minimal size of 250$boldsymbol{mu }$m inside the vessel contour and tracking with error recovery. Finally, real-time positions of the micro-agents are used for servoing of the robot with the attached US probe. Constant contact between the US probe and the surface of the body is accomplished by means of impedance control. Results: Breathing motion is compensated to keep constant contact between the US probe and the body surface, with minimal measured force of 2.02 N. Anthropomorphic phantom vessels are segmented with an Intersection-Over-Union (IOU) score of 0.93 $pm$ 0.05, while micro-agent tracking is performed with up to 99.8% success rate at 28–36 frames per second. Path planning, tracking and visual servoing are realized over 80 mm and 120 mm long surface paths. Conclusion: Experiments performed using anthropomorphic surfaces, biological tissue, simulation of physiological movement and simulation of fluid flow through the vessels indicate that robust visualization and tracking of micro-agents involving human patients is an achievable goal.

Objective: Brain-machine interfaces (BMIs) aim to provide direct brain control of devices such as prostheses and computer cursors, which have demonstrated great potential for motor restoration. One major limitation of current BMIs lies in the unstable performance due to the variability of neural signals, especially in online control, which seriously hinders the clinical availability of BMIs. Method: We propose a dynamic ensemble Bayesian filter (DyEnsemble) to deal with the neural variability in online BMI control. Unlike most existing approaches using fixed models, DyEnsemble learns a pool of models that contains diverse abilities in describing the neural functions. In each time slot, it dynamically weights and assembles the models according to the neural signals in a Bayesian framework. In this way, DyEnsemble copes with variability in signals and improves the robustness of online control. Results: Online BMI experiments with a human participant demonstrate that, compared with the velocity Kalman filter, DyEnsemble significantly improves the control accuracy (increases the success rate by 13.9% in the random target pursuit task) and robustness (performs more stably over different experiment days). Conclusion: Experimental results demonstrate the superiority of DyEnsemble in online BMI control. Significance: DyEnsemble frames a novel and flexible dynamic decoding framework for robust BMIs, beneficial to various neural decoding applications.

Dynamic positron emission tomography (dPET) is currently a widely used medical imaging technique for the clinical diagnosis, staging and therapy guidance of all kinds of human cancers. Higher temporal imaging resolution for the early stage of radiotracer metabolism is desired; however, in this case, the reconstructed images with short frame durations always suffer from a limited image signal-to-noise ratio (SNR) and unsatisfactory image spatial resolution. The appearance of uEXPLORER (United Imaging Healthcare, Inc.) with higher PET imaging sensitivity and resolution may help solving this problem. In this work, based on dynamic PET data acquired by uEXPLORER, we proposed a dPET processing method that denoises images with short frame durations via pixel-level time-activity curve (TAC) correction based on third-order Hermite interpolation (Pitch-In). The proposed method was validated and compared to several state-of-the-art methods to demonstrate its superior performance in terms of high temporal resolution dPET image noise reduction and imaging accuracy. Higher stability and feasibility of the proposed Pitch-In method for future clinical application with high temporal resolution (HTR) dPET imaging can be expected.

Fluorescence lifetime imaging is a valuable technique for probing characteristics of wide ranging samples and sensing of the molecular environment. However, the desire to measure faster and reduce effects such as photo bleaching in optical photon-count measurements for lifetime estimation lead to inevitable effects of convolution with the instrument response functions and noise, causing a degradation of the lifetime accuracy and precision. To tackle the problem, this paper presents a robust and computationally efficient framework for recovering fluorophore sample decay from the histogram of photon-count arrivals modelled as a decaying single-exponential function. In the proposed approach, the temporal histogram data is first decomposed into multiple bins via an adaptive multi-bin signal representation. Then, at each level of the multi-resolution temporal space, decay information including both the amplitude and the lifetime of a single-exponential function is rapidly decoded based on a novel statistical estimator. Ultimately, a game-theoretic model consisting of two players in an “amplitude-lifetime” game is constructed to be able to robustly recover optimal fluorescence decay signal from a set of fused multi-bin estimates. In addition to theoretical demonstrations, the efficiency of the proposed framework is experimentally shown on both synthesised and real data in different imaging circumstances. On a challenging low photon-count regime, our approach achieves about 28% improvement in bias than the best competing method. On real images, the proposed method processes data on average around 63 times faster than the gold standard least squares fit. Implementation codes are available to researchers.