Showing papers in "IEEE Transactions on Biomedical Engineering in 2021"

Domain Adaptation for Medical Image Analysis: A Survey

Abstract: Machine learning techniques used in computer-aided medical image analysis usually suffer from the domain shift problem caused by different distributions between source/reference data and target data. As a promising solution, domain adaptation has attracted considerable attention in recent years. The aim of this paper is to survey the recent advances of domain adaptation methods in medical image analysis. We first present the motivation of introducing domain adaptation techniques to tackle domain heterogeneity issues for medical image analysis. Then we provide a review of recent domain adaptation models in various medical image analysis tasks. We categorize the existing methods into shallow and deep models, and each of them is further divided into supervised, semi-supervised and unsupervised methods. We also provide a brief summary of the benchmark medical image datasets that support current domain adaptation research. This survey will enable researchers to gain a better understanding of the current status, challenges and future directions of this energetic research field.

Towards More Accurate Automatic Sleep Staging via Deep Transfer Learning

Abstract: Background: Despite recent significant progress in the development of automatic sleep staging methods, building a good model still remains a big challenge for sleep studies with a small cohort due to the data-variability and data-inefficiency issues. This work presents a deep transfer learning approach to overcome these issues and enable transferring knowledge from a large dataset to a small cohort for automatic sleep staging. Methods: We start from a generic end-to-end deep learning framework for sequence-to-sequence sleep staging and derive two networks as the means for transfer learning. The networks are first trained in the source domain (i.e. the large database). The pretrained networks are then finetuned in the target domain (i.e. the small cohort) to complete knowledge transfer. We employ the Montreal Archive of Sleep Studies (MASS) database consisting of 200 subjects as the source domain and study deep transfer learning on three different target domains: the Sleep Cassette subset and the Sleep Telemetry subset of the Sleep-EDF Expanded database, and the Surrey-cEEGrid database. The target domains are purposely adopted to cover different degrees of data mismatch to the source domains. Results: Our experimental results show significant performance improvement on automatic sleep staging on the target domains achieved with the proposed deep transfer learning approach. Conclusions: These results suggest the efficacy of the proposed approach in addressing the above-mentioned data-variability and data-inefficiency issues. Significance: As a consequence, it would enable one to improve the quality of automatic sleep staging models when the amount of data is relatively small. 1 1 The source code and the pretrained models are published at https://github.com/pquochuy/sleep_transfer_learning .

Home Use of a Percutaneous Wireless Intracortical Brain-Computer Interface by Individuals With Tetraplegia

Abstract: Objective. Individuals with neurological disease or injury such as amyotrophic lateral sclerosis, spinal cord injury or stroke may become tetraplegic, unable to speak or even locked-in. For people with these conditions, current assistive technologies are often ineffective. Brain-computer interfaces are being developed to enhance independence and restore communication in the absence of physical movement. Over the past decade, individuals with tetraplegia have achieved rapid on-screen typing and point-and-click control of tablet apps using intracortical brain-computer interfaces (iBCIs) that decode intended arm and hand movements from neural signals recorded by implanted microelectrode arrays. However, cables used to convey neural signals from the brain tether participants to amplifiers and decoding computers and require expert oversight, severely limiting when and where iBCIs could be available for use. Here, we demonstrate the first human use of a wireless broadband iBCI. Methods. Based on a prototype system previously used in pre-clinical research, we replaced the external cables of a 192-electrode iBCI with wireless transmitters and achieved high-resolution recording and decoding of broadband field potentials and spiking activity from people with paralysis. Two participants in an ongoing pilot clinical trial completed on-screen item selection tasks to assess iBCI-enabled cursor control. Results: Communication bitrates were equivalent between cabled and wireless configurations. Participants also used the wireless iBCI to control a standard commercial tablet computer to browse the web and use several mobile applications. Within-day comparison of cabled and wireless interfaces evaluated bit error rate, packet loss, and the recovery of spike rates and spike waveforms from the recorded neural signals. In a representative use case, the wireless system recorded intracortical signals from two arrays in one participant continuously through a 24-hour period at home. Significance. Wireless multi-electrode recording of broadband neural signals over extended periods introduces a valuable tool for human neuroscience research and is an important step toward practical deployment of iBCI technology for independent use by individuals with paralysis. On-demand access to high-performance iBCI technology in the home promises to enhance independence and restore communication and mobility for individuals with severe motor impairment.

Deep Learning Regression for Prostate Cancer Detection and Grading in Bi-Parametric MRI

Abstract: One of the most common types of cancer in men is prostate cancer (PCa). Biopsies guided by bi-parametric magnetic resonance imaging (MRI) can aid PCa diagnosis. Previous works have mostly focused on either detection or classification of PCa from MRI. In this work, however, we present a neural network that simultaneously detects and grades cancer tissue in an end-to-end fashion. This is more clinically relevant than the classification goal of the ProstateX-2 challenge. We used the dataset of this challenge for training and testing. We use a 2D U-Net with MRI slices as input and lesion segmentation maps that encode the Gleason Grade Group (GGG), a measure for cancer aggressiveness, as output. We propose a method for encoding the GGG in the model target that takes advantage of the fact that the classes are ordinal. Furthermore, we evaluate methods for incorporating prostate zone segmentations as prior information, and ensembling techniques. The model scored a voxel-wise weighted kappa of $0.446 \pm 0.082$ and a Dice similarity coefficient for segmenting clinically significant cancer of $0.370 \pm 0.046$ , obtained using 5-fold cross-validation. The lesion-wise weighted kappa on the ProstateX-2 challenge test set was $0.13 \pm 0.27$ . We show that our proposed model target outperforms standard multiclass classification and multi-label ordinal regression. Additionally, we present a comparison of methods for further improvement of the model performance.

Optimizing Survival Analysis of XGBoost for Ties to Predict Disease Progression of Breast Cancer

Abstract: Objective: Some excellent prognostic models based on survival analysis methods for breast cancer have been proposed and extensively validated, which provide an essential means for clinical diagnosis and treatment to improve patient survival. To analyze clinical and follow-up data of 12119 breast cancer patients, derived from the Clinical Research Center for Breast (CRCB) in West China Hospital of Sichuan University, we developed a gradient boosting algorithm, called EXSA, by optimizing survival analysis of XGBoost framework for ties to predict the disease progression of breast cancer. Methods: EXSA is based on the XGBoost framework in machine learning and the Cox proportional hazards model in survival analysis. By taking Efron approximation of partial likelihood function as a learning objective for ties, EXSA derives gradient formulas of a more precise approximation. It optimizes and enhances the ability of XGBoost for survival data with ties. After retaining 4575 patients (3202 cases for training, 1373 cases for test), we exploit the developed EXSA method to build an excellent prognostic model to estimate disease progress. Risk score of disease progress is evaluated by the model, and the risk grouping and continuous functions between risk scores and disease progress rate at 5- and 10-year are also demonstrated. Results: Experimental results on test set show that the EXSA method achieves competitive performance with concordance index of 0.83454, 5-year and 10-year AUC of 0.83851 and 0.78155, respectively. Conclusion: The proposed EXSA method can be utilized as an effective method for survival analysis. Significance: The proposed method in this paper can provide an important means for follow-up data of breast cancer or other disease research.

Multidimensional Ground Reaction Forces and Moments From Wearable Sensor Accelerations via Deep Learning

Abstract: Objective: Monitoring athlete internal workload exposure, including prevention of catastrophic non-contact knee injuries, relies on the existence of a custom early-warning detection system. This system must be able to estimate accurate, reliable, and valid musculoskeletal joint loads, for sporting maneuvers in near real-time and during match play. However, current methods are constrained to laboratory instrumentation, are labor and cost intensive, and require highly trained specialist knowledge, thereby limiting their ecological validity and wider deployment. An informative next step towards this goal would be a new method to obtain ground kinetics in the field. Methods: Here we show that kinematic data obtained from wearable sensor accelerometers, in lieu of embedded force platforms, can leverage recent supervised learning techniques to predict near real-time multidimensional ground reaction forces and moments (GRF/M). Competing convolutional neural network (CNN) deep learning models were trained using laboratory-derived stance phase GRF/M data and simulated sensor accelerations for running and sidestepping maneuvers derived from nearly half a million legacy motion trials. Then, predictions were made from each model driven by five sensor accelerations recorded during independent inter-laboratory data capture sessions. Results: The proposed deep learning workbench achieved correlations to ground truth, by maximum discrete GRF component, of vertical $F_z$ 0.97, anterior $F_y$ 0.96 (both running), and lateral $F_x$ 0.87 (sidestepping), with the strongest mean recorded across GRF components 0.89, and for GRM 0.65 (both sidestepping). Conclusion: These best-case correlations indicate the plausibility of the approach although the range of results was disappointing. The goal to accurately estimate near real-time on-field GRF/M will be improved by the lessons learned in this study. Significance: Coaching, medical, and allied health staff could ultimately use this technology to monitor a range of joint loading indicators during game play, with the aim to minimize the occurrence of non-contact injuries in elite and community-level sports.

Gesture Recognition in Robotic Surgery: A Review

Abstract: Objective: Surgical activity recognition is a fundamental step in computer-assisted interventions. This paper reviews the state-of-the-art in methods for automatic recognition of fine-grained gestures in robotic surgery focusing on recent data-driven approaches and outlines the open questions and future research directions. Methods: An article search was performed on 5 bibliographic databases with the following search terms: robotic, robot-assisted, JIGSAWS, surgery, surgical, gesture, fine-grained, surgeme, action, trajectory, segmentation, recognition, parsing. Selected articles were classified based on the level of supervision required for training and divided into different groups representing major frameworks for time series analysis and data modelling. Results: A total of 52 articles were reviewed. The research field is showing rapid expansion, with the majority of articles published in the last 4 years. Deep-learning-based temporal models with discriminative feature extraction and multi-modal data integration have demonstrated promising results on small surgical datasets. Currently, unsupervised methods perform significantly less well than the supervised approaches. Conclusion: The development of large and diverse open-source datasets of annotated demonstrations is essential for development and validation of robust solutions for surgical gesture recognition. While new strategies for discriminative feature extraction and knowledge transfer, or unsupervised and semi-supervised approaches, can mitigate the need for data and labels, they have not yet been demonstrated to achieve comparable performance. Important future research directions include detection and forecast of gesture-specific errors and anomalies. Significance: This paper is a comprehensive and structured analysis of surgical gesture recognition methods aiming to summarize the status of this rapidly evolving field.

A Magnetically Navigated Microcannula for Subretinal Injections

Abstract: Retinal disorders, including age-related macular degeneration, are leading causes of vision loss worldwide. New treatments, such as gene therapies and stem cell regeneration, require therapeutics to be introduced to the subretinal space due to poor diffusion to the active component of the retina. Subretinal injections are a difficult and risky surgical procedure and have been suggested as a candidate for robot-assisted surgery. We propose a different actuation paradigm to existing robotic approaches using remote magnetic navigation to control a flexible microcannula. A flexible cannula allows for high dexterity and considerable safety advantages over rigid tools, while maintaining the benefits of micrometer precision, hand tremor removal, and telemanipulation. The position of the cannula is tracked in real-time using near-infrared tip illumination, allowing for semi-automatic placement of the cannula and an intuitive user interface. Using this tool, we successfully performed several subretinal injections in ex-vivo porcine eyes under both microscope and optical coherence tomography visualization.

Multispectral Video Fusion for Non-Contact Monitoring of Respiratory Rate and Apnea

Abstract: Continuous monitoring of respiratory activity is desirable in many clinical applications to detect respiratory events. Non-contact monitoring of respiration can be achieved with near- and far-infrared spectrum cameras. However, current technologies are not sufficiently robust to be used in clinical applications. For example, they fail to estimate an accurate respiratory rate (RR) during apnea. We present a novel algorithm based on multispectral data fusion that aims at estimating RR also during apnea. The algorithm independently addresses the RR estimation and apnea detection tasks. Respiratory information is extracted from multiple sources and fed into an RR estimator and an apnea detector whose results are fused into a final respiratory activity estimation. We evaluated the system retrospectively using data from 30 healthy adults who performed diverse controlled breathing tasks while lying supine in a dark room and reproduced central and obstructive apneic events. Combining multiple respiratory information from multispectral cameras improved the root mean square error (RMSE) accuracy of the RR estimation from up to 4.64 monospectral data down to 1.60 breaths/min. The median F1 scores for classifying obstructive (0.75 to 0.86) and central apnea (0.75 to 0.93) also improved. Furthermore, the independent consideration of apnea detection led to a more robust system (RMSE of 4.44 vs. 7.96 breaths/min). Our findings may represent a step towards the use of cameras for vital sign monitoring in medical applications.

Deriving Polarimetry Feature Parameters to Characterize Microstructural Features in Histological Sections of Breast Tissues

Abstract: Objective: Mueller matrix polarimetry technique has been regarded as a powerful tool for probing the microstructural information of tissues. The multiplying of cells and remodeling of collagen fibers in breast carcinoma tissues have been reported to be related to patient survival and prognosis, and they give rise to observable patterns in hematoxylin and eosin (H&E) sections of typical breast tissues (TBTs) that the pathologist can label as three distinctive pathological features (DPFs)-cell nuclei, aligned collagen, and disorganized collagen. The aim of this paper is to propose a pixel-based extraction approach of polarimetry feature parameters (PFPs) using a linear discriminant analysis (LDA) classifier. These parameters provide quantitative characterization of the three DPFs in four types of TBTs. Methods: The LDA-based training method learns to find the most simplified linear combination from polarimetry basis parameters (PBPs) constrained under the accuracy remains constant to characterize the specific microstructural feature quantitatively in TBTs. Results: We present results from a cohort of 32 clinical patients with analysis of 224 regions-of-interest. The characterization accuracy for PFPs ranges from 0.82 to 0.91. Conclusion: This work demonstrates the ability of PFPs to quantitatively characterize the DPFs in the H&E pathological sections of TBTs. Significance: This technique paves the way for automatic and quantitative evaluation of specific microstructural features in histopathological digitalization and computer-aided diagnosis.

Energy-Efficient Neural Network for Epileptic Seizure Prediction.

Abstract: Seizure prediction for drug-refractory epilepsy patients can improve their quality of life, reduce their anxiety, and help them take the necessary precautions. Nowadays, numerous deep learning algorithms have been proposed to predict seizure onset and obtain better performance than traditional machine learning methods. However, these methods require a large set of parameters and large hardware resources; they also have high energy consumption. Therefore, these methods cannot be implemented on compact, low-power wearable, or implantable medical devices. The devices should operate on a real-time basis to continually inform the epileptic patients. In this paper, we describe energy-efficient and hardware-friendly methods to predict the epileptic seizures. A model of only 45 kB was obtained by the neural architecture search and was evaluated across three datasets. The overall accuracy, sensitivity, false prediction rate, and area under receiver operating characteristic curve were 99.53%, 99.81%, 0.005/h, 1 and 93.60%, 93.48%, 0.063/h, 0.977 and 86.86%, 85.19%, 0.116/h, 0.933, respectively, for the CHB-MIT scalp, the AES and Melbourne University intracranial electroencephalography (EEG) datasets. This model was further reduced with network pruning, quantization, and compact neural networks. The performances for the model sizes less than 50 kB for scalp EEG data and less than 10 kB for intracranial EEG data outperformed all the other models of similar model sizes. In particular, the energy consumption estimation was less than 10 mJ per inference for scalp EEG signal and less than 0.5 mJ per inference for intracranial EEG signal, which meet the requirements for low-power wearable and implantable devices, respectively.

An open-source and wearable system for measuring 3D human motion in real-time.

Abstract: Objective: Analyzing human motion is essential for diagnosing movement disorders and guiding rehabilitation for conditions like osteoarthritis, stroke, and Parkinson's disease. Optical motion capture systems are the standard for estimating kinematics but require expensive equipment located in a predefined space. While wearable sensor systems can estimate kinematics in any environment, existing systems are generally less accurate than optical motion capture. Many wearable sensor systems require a computer in close proximity and use proprietary software, limiting experimental reproducibility. Methods: Here, we present OpenSenseRT, an open-source and wearable system that estimates upper and lower extremity kinematics in real time by using inertial measurement units and a portable microcontroller. Results: We compared the OpenSenseRT system to optical motion capture and found an average RMSE of 4.4 degrees across 5 lower-limb joint angles during three minutes of walking and an average RMSE of 5.6 degrees across 8 upper extremity joint angles during a Fugl-Meyer task. The open-source software and hardware are scalable, tracking 1 to 14 body segments, with one sensor per segment. A musculoskeletal model and inverse kinematics solver estimate Kinematics in real-time. The computation frequency depends on the number of tracked segments, but is sufficient for real-time measurement for many tasks of interest; for example, the system can track 7 segments at 30 Hz in real-time. The system uses off-the-shelf parts costing approximately $100 USD plus $ 20 for each tracked segment. Significance: The OpenSenseRT system is validated against optical motion capture, low-cost, and simple to replicate, enabling movement analysis in clinics, homes, and free-living settings.

Progress in Nanorobotics for Advancing Biomedicine

Abstract: Nanorobotics, which has long been a fantasy in the realm of science fiction, is now a reality due to the considerable developments in diverse fields including chemistry, materials, physics, information and nanotechnology in the past decades. Not only different prototypes of nanorobots whose sizes are nanoscale are invented for various biomedical applications, but also robotic nanomanipulators which are able to handle nano-objects obtain substantial achievements for applications in biomedicine. The outstanding achievements in nanorobotics have significantly expanded the field of medical robotics and yielded novel insights into the underlying mechanisms guiding life activities, remarkably showing an emerging and promising way for advancing the diagnosis & treatment level in the coming era of personalized precision medicine. In this review, the recent advances in nanorobotics (nanorobots, nanorobotic manipulations) for biomedical applications are summarized from several facets (including molecular machines, nanomotors, DNA nanorobotics, and robotic nanomanipulators), and the future perspectives are also presented.

Magnetic Control of a Flexible Needle in Neurosurgery

Abstract: Minimally invasive neurosurgery does not require large incisions and openings in the skull to access the desired brain region, which often results in a faster recovery with fewer complications than traditional open neurosurgery. For disorders treated by the implantation of neurostimulators and thermocoagulation probes, current procedures incorporate a straight rigid needle, which restricts surgical trajectories and limits the number of possible targets and degrees of freedom at the respective target. A steerable needle with a flexible body could overcome these limitations. In this paper, we present a flexible needle steering system with magnetic and fluoroscopic guidance for neurosurgical procedures. A permanent magnet at the proximal end of a flexible needle is steered by an external magnetic field, and the resultant tip-deflection angle bends the flexible body like a bevel-tip needle. We implemented a kinematic model for the magnetic needle derived from a nonholonomic bicycle model and a closed-loop control strategy with feed-forward and feed-back components using a chained-form transformation. The proposed needle steering method was investigated through in vitro and ex vivo experiments.

Deep Learning for Robust Decomposition of High-Density Surface EMG Signals

Abstract: Blind source separation (BSS) algorithms, such as gradient convolution kernel compensation (gCKC), can efficiently and accurately decompose high-density surface electromyography (HD-sEMG) signals into constituent motor unit (MU) action potential trains. Once the separation matrix is blindly estimated on a signal interval, it is also possible to apply the same matrix to subsequent signal segments. Nonetheless, the trained separation matrices are sub-optimal in noisy conditions and require that incoming data undergo computationally expensive whitening. One unexplored alternative is to instead use the paired HD-sEMG signal and BSS output to train a model to predict MU activations within a supervised learning framework. A gated recurrent unit (GRU) network was trained to decompose both simulated and experimental unwhitened HD-sEMG signal using the output of the gCKC algorithm. The results on the experimental data were validated by comparison with the decomposition of concurrently recorded intramuscular EMG signals. The GRU network outperformed gCKC at low signal-to-noise ratios, proving superior performance in generalising to new data. Using 12 seconds of experimental data per recording, the GRU performed similarly to gCKC, at rates of agreement of 92.5% (84.5%–97.5%) and 94.9% (88.8%–100.0%) respectively for GRU and gCKC against matched intramuscular sources.

Tongue Tumor Detection in Hyperspectral Images Using Deep Learning Semantic Segmentation

Abstract: Objective: The utilization of hyperspectral imaging (HSI) in real-time tumor segmentation during a surgery have recently received much attention, but it remains a very challenging task. Methods: In this work, we propose semantic segmentation methods, and compare them with other relevant deep learning algorithms for tongue tumor segmentation. To the best of our knowledge, this is the first work using deep learning semantic segmentation for tumor detection in HSI data using channel selection, and accounting for more spatial tissue context, and global comparison between the prediction map, and the annotation per sample. Results, and Conclusion: On a clinical data set with tongue squamous cell carcinoma, our best method obtains very strong results of average dice coefficient, and area under the ROC-curve of $\text{0.891}\pm \text{0.053}$ , and $\text{0.924}\pm \text{0.036}$ , respectively on the original spatial image size. The results show that a very good performance can be achieved even with a limited amount of data. We demonstrate that important information regarding tumor decision is encoded in various channels, but some channel selection, and filtering is beneficial over the full spectra. Moreover, we use both visual (VIS), and near-infrared (NIR) spectrum, rather than commonly used only VIS spectrum; although VIS spectrum is generally of higher significance, we demonstrate NIR spectrum is crucial for tumor capturing in some cases. Significance: The HSI technology augmented with accurate deep learning algorithms has a huge potential to be a promising alternative to digital pathology or a doctors’ supportive tool in real-time surgeries.

Intraoperative Margin Assessment in Oral and Oropharyngeal Cancer Using Label-Free Fluorescence Lifetime Imaging and Machine Learning

Abstract: Objective: To demonstrate the diagnostic ability of label-free, point-scanning, fiber-based Fluorescence Lifetime Imaging (FLIm) as a means of intraoperative guidance during oral and oropharyngeal cancer removal surgery. Methods: FLIm point-measurements acquired from 53 patients (n = 67893 pre-resection in vivo , n = 89695 post-resection ex vivo ) undergoing oral or oropharyngeal cancer removal surgery were used for analysis. Discrimination of healthy tissue and cancer was investigated using various FLIm-derived parameter sets and classifiers (Support Vector Machine, Random Forests, CNN). Classifier output for the acquired set of point-measurements was visualized through an interpolation-based approach to generate a probabilistic heatmap of cancer within the surgical field. Classifier output for dysplasia at the resection margins was also investigated. Results: Statistically significant change (P $ 0.01) between healthy and cancer was observed in vivo for the acquired FLIm signal parameters (e.g., average lifetime) linked with metabolic activity. Superior classification was achieved at the tissue region level using the Random Forests method (ROC-AUC: 0.88). Classifier output for dysplasia (% probability of cancer) was observed to lie between that of cancer and healthy tissue, highlighting FLIm's ability to distinguish various conditions. Conclusion: The developed approach demonstrates the potential of FLIm for fast, reliable intraoperative margin assessment without the need for contrast agents. Significance: Fiber-based FLIm has the potential to be used as a diagnostic tool during cancer resection surgery, including Transoral Robotic Surgery (TORS), helping ensure complete resections and improve the survival rate of oral and oropharyngeal cancer patients.

Novel Density Poincaré Plot Based Machine Learning Method to Detect Atrial Fibrillation From Premature Atrial/Ventricular Contractions

Abstract: Objective: Detection of Atrial fibrillation (AF) from premature atrial contraction (PAC) and premature ventricular contraction (PVC) is difficult as frequent occurrences of these ectopic beats can mimic the typical irregular patterns of AF In this paper, we present a novel density Poincare plot-based machine learning method to detect AF from PAC/PVCs using electrocardiogram (ECG) recordings Methods: First, we propose the generation of this new density Poincare plot which is derived from the difference of the heart rate (DHR) and provides the overlapping phase-space trajectory information of the DHR Next, from this density Poincare plot, several image processing domain-based approaches including statistical central moments, template correlation, Zernike moment, discrete wavelet transform and Hough transform features are used to extract suitable features Subsequently, the infinite latent feature selection algorithm is implemented to rank the features Finally, classification of AF vs PAC/PVC is performed using K-Nearest Neighbor, Support vector machine (SVM) and Random Forest (RF) classifiers Our method is developed and validated using a subset of Medical Information Mart for Intensive Care (MIMIC) III database containing 10 AF and 10 PAC/PVC subjects Results- During the segment-wise 10-fold cross-validation, SVM achieved the best performance with 9899% sensitivity, 9518% specificity and 9745% accuracy with the extracted features In subject-wise scenario, RF achieved the highest accuracy of 9193% Moreover, we further validated the proposed method using two other databases: wearable armband ECG data and the Physionet AFPDB 100% PAC detection accuracy was obtained for both databases without any further training Conclusion: Our proposed density Poincare plot-based method showed superior performance when compared with four existing algorithms; thus showing the efficacy of the extracted image domain-based features Significance: From intensive care unit's ECG to wearable armband ECGs, the proposed method is shown to discriminate PAC/PVCs from AF with high accuracy

Enabling Robust Radar-Based Localization and Vital Signs Monitoring in Multipath Propagation Environments

Abstract: Objective: Over the last two decades, radar-based contactless monitoring of vital signs (heartbeat and respiration rate) has raised increasing interest as an emerging and added value to health care. However, until now, the flaws caused by indoor multipath propagation formed a fundamental hurdle for the adoption of such technology in practical healthcare applications where reliability and robustness are crucial. Multipath reflections, originated from one person, combine with the direct signals and multipaths of other people and stationary objects, thus jeopardizing individual vital signs extraction and localization. This work focuses on tackling indoor multipath propagation. Methods: We describe a methodology, based on accurate models of the indoor multipaths and of the radar signals, that enables separating the undesired multipaths from desired signals of multiple individuals, removing a key obstacle to real-world contactless vital signs monitoring and localization. Results: We also demonstrated it by accurately measure individual heart rates, respiration rates, and absolute distances (range information) of paired volunteers in a challenging real-world office setting. Conclusion: The approach, validated using a frequency-modulated continuous wave (FMCW) radar, was shown to function in an indoor environment where radar signals are severely affected by multipath reflections. Significance: Practical applications arise for health care, assisted living, geriatric and quarantine medicine, rescue and security purposes.

Do CNNs Solve the CT Inverse Problem

Abstract: Objective: This work examines the claim made in the literature that the inverse problem associated with image reconstruction in sparse-view computed tomography (CT) can be solved with a convolutional neural network (CNN). Methods: Training, and testing image/data pairs are generated in a dedicated breast CT simulation for sparse-view sampling, using two different object models. The trained CNN is tested to see if images can be accurately recovered from their corresponding sparse-view data. For reference, the same sparse-view CT data is reconstructed by the use of constrained total-variation (TV) minimization (TVmin), which exploits sparsity in the gradient magnitude image (GMI). Results: There is a significant discrepancy between the image obtained with the CNN and the image that generated the data. TVmin is able to accurately reconstruct the test images. Conclusion: We find that the sparse-view CT inverse problem cannot be solved for the particular published CNN-based methodology that we chose, and the particular object model that we tested. Significance: The inability of the CNN to solve the inverse problem associated with sparse-view CT, for the specific conditions of the presented simulation, draws into question similar unsupported claims being made for the use of CNNs and deep-learning to solve inverse problems in medical imaging.

A Preclinical System Prototype for Focused Microwave Breast Hyperthermia Guided by Compressive Thermoacoustic Tomography

Abstract: Objective: As a newly developed technique, focused microwave breast hyperthermia (FMBH) can provide accurate and cost-effective treatment of breast tumors with low side effect. A clinically feasible FMBH system requires a guidance technique to monitor the microwave power distribution in the breast. Compressive thermoacoustic tomography (CTT) is a suitable guidance approach for FMBH, which is more cost-effective than MRI. However, no experimental validation based on a realized FMBH-CTT system has been reported, which greatly hinders the further advancement of this novel approach. Methods: We developed a preclinical system prototype for the FMBH-CTT technique, containing a microwave phased antenna array, a microwave source, an ultrasound transducer array and associated data acquisition module. Results: Experimental results employing homogeneous and inhomogeneous breast-mimicking phantoms demonstrate that the CTT technique can offer reliable guidance for the entire process of the FMBH. In addition, small phase noises do not deteriorate the overall performance of the system prototype. Conclusion: The realized preclinical FMBH-CTT system prototype is capable for noninvasive, accurate and low-side-effect breast tumor treatment with effective guidance. Significance: The experimentally validated FMBH-CTT system prototype provides a feasible paradigm for CTT guided FMBH, establishes a practical platform for future improvement of this technique, and paves the way for potential clinical translation.

Functional Connectivity-Based Prediction of Autism on Site Harmonized ABIDE Dataset

Abstract: Objective: The larger sample sizes available from multi-site publicly available neuroimaging data repositories makes machine-learning based diagnostic classification of mental disorders more feasible by alleviating the curse of dimensionality. However, since multi-site data are aggregated post-hoc, i.e. they were acquired from different scanners with different acquisition parameters, non-neural inter-site variability may mask inter-group differences that are at least in part neural in origin. Hence, the advantages gained by the larger sample size in the context of machine-learning based diagnostic classification may not be realized. Methods: We address this issue using harmonization of multi-site neuroimaging data using the ComBat technique, which is based on an empirical Bayes formulation to remove inter-site differences in data distributions, to improve diagnostic classification accuracy. Specifically, we demonstrate this using ABIDE (Autism Brain Imaging Data Exchange) multi-site data for classifying individuals with Autism from healthy controls using resting state fMRI-based functional connectivity data. Results: Our results show that higher classification accuracies across multiple classification models can be obtained (especially for models based on artificial neural networks) from multi-site data post harmonization with the ComBat technique as compared to without harmonization, outperforming earlier results from existing studies using ABIDE. Furthermore, our network ablation analysis facilitated important insights into autism spectrum disorder pathology and the connectivity in networks shown to be important for classification covaried with verbal communication impairments in Autism. Conclusion: Multi-site data harmonization using ComBat improves neuroimaging-based diagnostic classification of mental disorders. Significance: ComBat has the potential to make AI-based clinical decision-support systems more feasible in psychiatry.

Autonomic Robotic Ultrasound Imaging System Based on Reinforcement Learning

Abstract: O bjective: In this paper, we introduce an autonomous robotic ultrasound (US) imaging system based on reinforcement learning (RL). The proposed system and framework are committed to controlling the US probe to perform fully autonomous imaging of a soft, moving and marker-less target based only on single RGB images of the scene. Methods: We propose several different approaches and methods to achieve the following objectives: real-time US probe controlling, soft surface constant force tracking and automatic imaging. First, to express the state of the robotic US imaging task, we proposed a state representation model to reduce the dimensionality of the imaging state and encode the force and US information into the scene image space. Then, an RL agent is trained by a policy gradient theorem based RL model with the single RGB image as the only observation. To achieve adaptable constant force tracking between the US probe and the soft moving target, we propose a force-to-displacement control method based on an admittance controller. Results: In the simulation experiment, we verified the feasibility of the integrated method. Furthermore, we evaluated the proposed force-to-displacement method to demonstrate the safety and effectiveness of adaptable constant force tracking. Finally, we conducted phantom and volunteer experiments to verify the feasibility of the method on a real system. Conclusion: The experiments indicated that our approaches were stable and feasible in the autonomic and accurate control of the US probe. Significance: The proposed system has potential application value in the image-guided surgery and robotic surgery.

Supervised Descent Learning for Thoracic Electrical Impedance Tomography

Abstract: Objective: The absolute image reconstruction problem of electrical impedance tomography (EIT) is ill-posed. Traditional methods usually solve a nonlinear least squares problem with some kind of regularization. These methods suffer from low accuracy, poor anti-noise performance, and long computation time. Besides, the integration of a priori information is not very flexible. This work tries to solve EIT inverse problem using a machine learning algorithm for the application of thorax imaging. Methods: We developed the supervised descent learning EIT (SDL-EIT) inversion algorithm based on the idea of supervised descent method (SDM). The algorithm approximates the mapping from measured data to the conductivity image by a series of descent directions learned from training samples. We designed a training data set in which the thorax contour, and some general structure of lungs, and heart are embedded. The algorithm is implemented in both two-, and three-dimensional cases, and is evaluated using synthetic, and measured thoracic data. Results, and conclusion: For synthetic data, SDL-EIT shows better accuracy, and anti-noise performance compared with traditional Gauss–Newton inversion (GNI) method. For measured data, the result of SDL-EIT is reasonable compared with computed tomography (CT) scan image. Significance: Using SDL-EIT, prior information can be easily integrated through the specifically designed training data set, and the image reconstruction process can be accelerated. The algorithm is effective in inverting measured thoracic data. It is a potential algorithm for human thorax imaging.

Detection and Classification of Sleep Apnea and Hypopnea Using PPG and SpO $_2$ Signals

Abstract: In this work, a detection and classification method for sleep apnea and hypopnea, using photopletysmography (PPG) and peripheral oxygen saturation (SpO $_2$ ) signals, is proposed. The detector consists of two parts: one that detects reductions in amplitude fluctuation of PPG (DAP)and one that detects oxygen desaturations. To further differentiate among sleep disordered breathing events (SDBE), the pulse rate variability (PRV) was extracted from the PPG signal, and then used to extract features that enhance the sympatho-vagal arousals during apneas and hypopneas. A classification was performed to discriminate between central and obstructive events, apneas and hypopneas. The algorithms were tested on 96 overnight signals recorded at the UZ Leuven hospital, annotated by clinical experts, and from patients without any kind of co-morbidity. An accuracy of 75.1% for the detection of apneas and hypopneas, in one-minute segments,was reached. The classification of the detected events showed 92.6% accuracy in separating central from obstructive apnea, 83.7% for central apnea and central hypopnea and 82.7% for obstructive apnea and obstructive hypopnea. The low implementation cost showed a potential for the proposed method of being used as screening device, in ambulatory scenarios.

Align and pool for EEG headset domain adaptation (ALPHA) to facilitate dry electrode based SSVEP-BCI

Abstract: Objective: The steady-state visual evoked potential based brain-computer interface (SSVEP-BCI) implemented in dry electrodes is a promising paradigm for alternative and augmentative communication in real-world applications. To improve its performance and reduce the calibration effort for dry-electrode systems, we utilize cross-device transfer learning by exploiting auxiliary individual wet-electrode electroencephalogram (EEG). Methods: We proposed a novel transfer learning framework named ALign and Pool for EEG Headset domain Adaptation (ALPHA), which aligns the spatial pattern and the covariance for domain adaptation. To evaluate its efficacy, 75 subjects performed an experiment of 2 sessions involving a 12-target SSVEP-BCI task. Results: ALPHA significantly outperformed a baseline approach (canonical correlation analysis, CCA) and two competing transfer learning approaches (transfer template CCA, ttCCA and least square transformation, LST) in two transferring directions. When transferring from wet to dry EEG headsets, ALPHA significantly outperformed the fully calibrated approach of task-related component analysis (TRCA). Conclusion: ALPHA advances the frontier of recalibration-free cross-device transfer learning for SSVEP-BCIs and boosts the performance of dry electrode based systems. Significance: ALPHA has methodological and practical implications and pushes the boundary of dry electrode based SSVEP-BCI toward real-world applications.

Reconstruction of Organ Boundaries With Deep Learning in the D-Bar Method for Electrical Impedance Tomography

Abstract: Objective: Medical electrical impedance tomography is a non-ionizing imaging modality in which low-amplitude, low-frequency currents are applied on electrodes on the body, the resulting voltages are measured, and an inverse problem is solved to determine the conductivity distribution in the region of interest. Due the ill-posedness of the inverse problem, the boundaries of internal organs are typically blurred in the reconstructed image. Methods: A deep learning approach is introduced in the D-bar method for reconstructing a 2-D slice of the thorax to recover the boundaries of organs. This is accomplished by training a deep neural network on labeled pairs of scattering transforms and the boundaries of the organs in the data from which the transforms were computed. This allows the network to “learn” the nonlinear mapping between them by minimizing the error between the output of the network and known actual boundaries. Further, a “sparse” reconstruction is computed by fusing the results of the standard D-bar reconstruction with reconstructed organ boundaries from the neural network. Results: Results are shown on simulated and experimental data collected on a saline-filled tank with agar targets simulating the conductivity of the heart and lungs. Conclusions and Significance: The results demonstrate that deep neural networks can successfully learn the mapping between scattering transforms and the internal boundaries of structures.

Evaluation of Non-Invasive Ankle Joint Effort Prediction Methods for Use in Neurorehabilitation Using Electromyography and Ultrasound Imaging

Abstract: Objective: Reliable measurement of voluntary human effort is essential for effective and safe interaction between the wearer and an assistive robot. Existing voluntary effort prediction methods that use surface electromyography (sEMG) are susceptible to prediction inaccuracies due to non-selectivity in measuring muscle responses. This technical challenge motivates an investigation into alternative non-invasive effort prediction methods that directly visualize the muscle response and improve effort prediction accuracy. The paper is a comparative study of ultrasound imaging (US)-derived neuromuscular signals and sEMG signals for their use in predicting isometric ankle dorsiflexion moment. Furthermore, the study evaluates the prediction accuracy of model-based and model-free voluntary effort prediction approaches that use these signals. Methods: The study evaluates sEMG signals and three US imaging-derived signals: pennation angle, muscle fascicle length, and echogenicity and three voluntary effort prediction methods: linear regression (LR), feedforward neural network (FFNN), and Hill-type neuromuscular model (HNM). Results: In all the prediction methods, pennation angle and fascicle length significantly improve the prediction accuracy of dorsiflexion moment, when compared to echogenicity. Also, compared to LR, both FFNN and HNM improve dorsiflexion moment prediction accuracy. Conclusion: The findings indicate FFNN or HNM approach and using pennation angle or fascicle length predict human ankle movement intent with higher accuracy. Significance: The accurate ankle effort prediction will pave the path to safe and reliable robotic assistance in patients with drop foot.

Clinically Verified Hybrid Deep Learning System for Retinal Ganglion Cells Aware Grading of Glaucomatous Progression

Abstract: Objective: Glaucoma is the second leading cause of blindness worldwide. Glaucomatous progression can be easily monitored by analyzing the degeneration of retinal ganglion cells (RGCs). Many researchers have screened glaucoma by measuring cup-to-disc ratios from fundus and optical coherence tomography scans. However, this paper presents a novel strategy that pays attention to the RGC atrophy for screening glaucomatous pathologies and grading their severity. Methods: The proposed framework encompasses a hybrid convolutional network that extracts the retinal nerve fiber layer, ganglion cell with the inner plexiform layer and ganglion cell complex regions, allowing thus a quantitative screening of glaucomatous subjects. Furthermore, the severity of glaucoma in screened cases is objectively graded by analyzing the thickness of these regions. Results: The proposed framework is rigorously tested on publicly available Armed Forces Institute of Ophthalmology (AFIO) dataset, where it achieved the $F_1$ score of 0.9577 for diagnosing glaucoma, a mean dice coefficient score of 0.8697 for extracting the RGC regions and an accuracy of 0.9117 for grading glaucomatous progression. Furthermore, the performance of the proposed framework is clinically verified with the markings of four expert ophthalmologists, achieving a statistically significant Pearson correlation coefficient of 0.9236. Conclusion: An automated assessment of RGC degeneration yields better glaucomatous screening and grading as compared to the state-of-the-art solutions. Significance: An RGC-aware system not only screens glaucoma but can also grade its severity and here we present an end-to-end solution that is thoroughly evaluated on a standardized dataset and is clinically validated for analyzing glaucomatous pathologies.

Real-Time Forecasting of sEMG Features for Trunk Muscle Fatigue Using Machine Learning

Abstract: Objective: Several features of the surface electromyography (sEMG) signal are related to muscle activity and fatigue. However, the time-evolution of these features are non-stationary and vary between subjects. The aim of this study is to investigate the use of adaptive algorithms to forecast sEMG feature of the trunk muscles. Methods: Shallow models and a deep convolutional neural network (CNN) were used to simultaneously learn and forecast 5 common sEMG features in real-time to provide tailored predictions. This was investigated for: up to a 25 second horizon; for 14 different muscles in the trunk; across 13 healthy subjects; while they were performing various exercises. Results: The CNN was able to forecast 25 seconds ahead of time, with 6.88% mean absolute percentage error and 3.72% standard deviation of absolute percentage error, across all the features. Moreover, the CNN outperforms the best shallow model in terms of a figure of merit combining accuracy and precision by at least 30% for all the 5 features. Conclusion: Even though the sEMG features are non-stationary and vary between subjects, adaptive learning and forecasting, especially using CNNs, can provide accurate and precise forecasts across a range of physical activities. Significance: The proposed models provide the groundwork for a wearable device which can forecast muscle fatigue in the trunk, so as to potentially prevent low back pain. Additionally, the explicit real-time forecasting of sEMG features provides a general model which can be applied to many applications of muscle activity monitoring, which helps practitioners and physiotherapists improve therapy.