Search

In this paper, we propose an end-to-end approach to addressing QRS complex detection and measurement of Electrocardiograph (ECG) paper using convolutional neural networks (CNNs). Unlike conventional detection solutions that convert images to digital data, our method can directly detect QRS complex in images using Faster-RCNN, then the R-peak can be located and measured through a CNN. Validated by clinical ECG data in the St.-Petersburg Institute of Cardiological Technics 12-lead Arrhythmia Database and real ECG paper from Peking University People's Hospital, the proposed method can achieve the recall of 98.32%, the precision of 99.01% in detecting and 0.012 mv of mean absolute error in measuring. Experimental results demonstrate the superior performance of our method over conventional solutions, which would pave the way to detect and measure ECG paper using CNNs.

Chronic wounds cause a number of unnecessary amputations due to a delay in proper treatment. To expedite timely treatment, this paper presents an algorithm which uses a logistic regression classifier to predict whether the wound will heal or not within a specified time. The prediction is made at three time-points: one month, three months, and six months from the first visit of the patient to the healthcare facility. This prediction is made using a systematically collected chronic wound registry and is based entirely on data collected during patients' first visit. The algorithm achieves an area under the receiver operating characteristic curve (AUC) of 0.75, 0.72, and 0.71 for the prediction at the three time-points, respectively.Clinical relevance- Using the proposed prediction model, the clinicians will have an early estimate of the time taken to heal thereby providing appropriate treatments. We hope this will ensure timely treatments and reduce the number of unnecessary amputations.

This paper proposes a novel algorithm that allows a significant improvement of the resolution of frequency modulated magnetic induction sensors while providing high sampling rates. We have implemented this approach in a frequency modulated magnetic induction sensor and our first measurements demonstrate the improvement of the sensor's signal quality.

This paper explores automated face and facial landmark detection of neonates, which is an important first step in many video-based neonatal health applications, such as vital sign estimation, pain assessment, sleep-wake classification, and jaundice detection. Utilising three publicly available datasets of neonates in the clinical environment, 366 images (258 subjects) and 89 (66 subjects) were annotated for training and testing, respectively. Transfer learning was applied to two YOLO-based models, with input training images augmented with random horizontal flipping, photo-metric colour distortion, translation and scaling during each training epoch. Additionally, the re-orientation of input images and fusion of trained deep learning models was explored. Our proposed model based on YOLOv7Face outperformed existing methods with a mean average precision of 84.8% for face detection, and a normalised mean error of 0.072 for facial landmark detection. Overall, this will assist in the development of fully automated neonatal health assessment algorithms.Clinical relevance- Accurate face and facial landmark detection provides an automated and non-contact option to assist in video-based neonatal health applications.

Neural interfaces that electrically stimulate the peripheral nervous system have been shown to successfully improve symptom management for several conditions, such as epilepsy and depression. A crucial part for closing the loop and improving the efficacy of implantable neuromodulation devices is the efficient extraction of meaningful information from nerve recordings, which can have a low Signal-to-Noise ratio (SNR) and non-stationary noise. In recent years, machine learning (ML) models have shown outstanding performance in regression and classification problems, but it is often unclear how to translate and assess these for novel tasks in biomedical engineering. This paper aims to adapt existing ML algorithms to carry out unsupervised denoising of neural recordings instead. This is achieved by applying bandpass filtering and two novel ML algorithms to in-vivo spontaneous, low-SNR vagus nerve recordings. The performance of each approach is compared using the task of extracting respiratory afferent activity and validated using cross-correlation, MSE, and accuracy in terms of extracting the true respiratory rate. A variational autoencoder (VAE) model in particular produces results that show better correlation with respiratory activity compared to bandpass filtering, highlighting that these models have the potential to preserve relevant features in complex neural recordings.

Gastrointestinal (GI) potential mapping could be useful for evaluating GI motility disorders. Such disorders are found in inflammatory bowel diseases, such as Crohn's disease, or GI functional disorders. GI potential mapping data originate from a mixture of several GI electrophysiological sources (termed ExG) and other noise sources, including the electrocardiogram (ECG) and respiration. Denoising and/or source separation techniques are required, however, with real measurements, no ground truth is available. In this paper we propose a framework for the simulation of body surface GI potential mapping data. The framework is an electrostatic model, based on fecgsyn toolbox, using dipoles as electrical sources for the heart, stomach, small bowel and colon, and an array of surface electrodes. It is shown to generate realistic ExG waveforms, which are then used to compare several ECG and respiration cancellation techniques, based on, fast independent component analysis (FastICA) and pseudo-periodic component analysis (PiCA). The best performance was obtained with PiCA with a median root mean squared error of 0.005.

Cephalometric analysis plays an important role in orthodontic diagnosis and treatment planning. It depends on the detection of multiple landmarks, while the process is time-consuming and tedious. Although some deep learning-based automatic landmark detection algorithms have achieved excellent performance, most of them adopt multi-stage models increasing the complexity and detection time. Meanwhile, few studies focused on the uncertainty of detection results, thereby ignoring its significant clinical value. In this paper, we propose a novel approach based on heatmap regression for landmark detection, which can achieve competitive accuracy and good robustness with only one step. Furthermore, by applying Monte Carlo dropout to a U-shaped convolutional neural network, we can obtain not only the coordinate of each landmark but also the corresponding simple uncertainty, so that doctors can pay more attention to those landmarks with higher uncertainty. The evaluation results showed the mean radial error is 1.39±1.06mm and the average successful detection rate is 79.65%, 97.22% within 2mm, 4mm for the IEEE ISBI2015 Test Dataset 1, the indicators for the IEEE ISBI2015 Test Dataset 2 are 1.33±0.93mm, 80.05% and 97.53%, respectively. Our method has the potential to become an assistant tool in clinical practice. Automatic and accurate detection with uncertainty analysis is expected to help guide the doctor's judgment.

This paper presents a novel technique for cell segmentation, named "Cellsketch," which generates an RGB mask containing simplified representations of cells (including nuclei, whole-cell, and cell boundaries) from microscopic images, and applies the watershed algorithm to produce segmentation masks of cells and nuclei. The RGB mask is generated using a generator model trained with a combination of L1 loss and adversarial loss. The method achieved accurate cell and nuclei segmentation from differential interference contrast (DIC) images using only automatically annotated training data and shows potential for a generalizable algorithm for cell segmentation. The code is freely available at: https://github.com/iranovianti/cellsketchClinical Relevance- This method simultaneously detects individual cells and nuclei from DIC images.

Brain-Machine Interfaces (BMIs) have the potential to allow subjects to brain control (BC) external devices, where their brain signals could be translated to the action of the neuro-prosthesis by reinforcement learning (RL) based decoder. During the BC task, feedback cues are provided to guide subject's learning. Subjects will adapt the neural signals according to the feedback cues. Concurrently, the RL decoding parameters are adjusted when the subject explores the BC task through trial and error, leading to a co-adaptive process between the subject and the decoder. However, when subjects receive the feedback cues and enhance their learning, the decoder does not actively utilize the feedback cues. If the RL decoder could integrate both neural signals and feedback cues, the training efficiency of the BC task would increase. A major challenge is the different temporal scales of neural signals and feedback cues, making it difficult to integrate them into a single decoder. In this paper, we propose a novel kernel RL decoding method as the first attempt to combine two signals with different temporal scales for RL decoding. The neural signals and the feedback cues comprise the decoding input, which is then projected into individual Reproducing Kernel Hilbert Spaces (RKHSs) respectively. These two RKHSs form a joint feature space, where the action of the neuro-prosthesis could be decoded linearly. We evaluate the proposed method on a simulated brain control cursor-reaching task. Our proposed method is compared with the kernel RL that only uses neural signals as the input. The proposed method has a faster learning speed and better decoding accuracy. The results demonstrate that our proposed method has successfully integrated the information of the feedback cue and facilitates the training procedure for the BC task.Clinical Relevance-This paper provides an integrated reinforcement learning decoding framework, which combines the neural signals and the feedback cues to increase the learning speed and the accuracy of the brain control task. Subjects could learn the task more easily with this decoder.

Behaviors are encoded by multi-scale brain signals, from microscopic spike signals to macroscopic extracellular Field Potentials (FPs). Extracting neuronal spike information from FPs is an important, yet challenging problem. Because FPs stem from summed contributions of a large population of neurons. Previous work inferred single-neuron spiking activity from the FPs using a generalized linear model (GLM). However, FPs reflect the states of neural ensembles more than single-neuron spike trains. In this paper, we propose a computational model to decode ensemble spike states from FPs. This framework first extracts transient features in FPs, and then detects typical ensemble spike patterns and assigns state labels accordingly. Finally, we use a neural network to decode the ensemble spike states from the FP neuromodulations. This FP-Spike decoder is tested on the FP and spike data from the M1 area of an SD rat. We show that our model can effectively decode multi-neuron spike states. Compared with the GLM method for single-neuron spike prediction, our model exhibits 37% less ensemble spike pattern decoding error. These preliminary results show that we can decode informative spike states from FPs, indicating that the decode results can further benefit long-term stable brain-machine interfaces.

Identifying the physical locations of neurons based on the spike waveforms captured by multiple recording channels, namely spike localization, can potentially enhance spike sorting accuracy. This study proposes a new method for spike localization, where the problem is first described as a nonconvex optimization problem and then the optimization is attempted heuristically via a numerical Ising solver. The paper first presents a quadratic unconstrained binary optimization (QUBO) formulation of spike localization. Then, a MATLAB solver simulating an Ising machine is written to solve the QUBO. The proposed method is evaluated on a 2D toy problem consisting of two electrodes and a single spike event, where the neuron location search is conducted in three different regions placed at increasing distances from the electrodes. The results indicate that the neuron can be accurately identified when in one of the nearest nodes to the electrodes, whereas the accuracy decreases to 87.5% and 75% as the search region distance increases. The study for the first time formulates the spike localization problem as a QUBO and demonstrates the feasibility of solving the resultant non-convex optimization problem heuristically using an Ising machine.Clinical Relevance- High channel-count implantable neural monitoring systems allow tracking large brain regions at the cost of increased data volumes to transmit and power dissipation. The new spike localization approach presented can potentially decrease the data volume and power consumption by enabling high accuracy spike localization at the implantable system.

There is a growing concern regarding the driving safety of Motorized Mobility Scooters (MMSs) for the elderly and mobility-impaired individuals. Although various research has made progress in sensor-based driving assistance systems to identify environmental hazards, few studies focus on investigating the impact of user behavior on MMS driving. In this paper, a driving status logging (DSL) system is developed to measure the user's behavior while driving. A cross-correlation analysis is implemented to quantify the temporal relationship between the head movement and steering operation in the driving of MMSs. The preliminary results suggest that the head movement can be used as an appropriate index to predict the intended steering operation in the driving of MMSs. Moreover, the quantified head-steering lag time provides the possibility to identify the hazardous driving pattern of MMS users.Clinical Relevance- The investigation of user behavior has the potential to improve the safety of MMSs. In this study, the user behavior in the driving of MMSs was quantitatively measured using the developed DSL system. Consequently, the temporal relationship between head movement and steering operation was first quantified in MMS-related research. These outcomes provide valuable insights into developing behavioral interventions to address the user's risky behavior patterns, thereby promoting the driving safety of MMSs.

3D cell tracking in a living organism has a crucial role in live cell image analysis. Cell tracking in C. elegans has two difficulties. First, cell migration in a consecutive frame is large since they move their head during scanning. Second, cell detection is often inconsistent in consecutive frames due to touching cells and low-contrast images, and these inconsistent detections affect the tracking performance worse. In this paper, we propose a cell tracking method to address these issues, which has two main contributions. First, we introduce cell position heatmap-based non-rigid alignment with test-time fine-tuning, which can warp the detected points to near the positions at the next frame. Second, we propose a pairwise detection method, which uses the information of detection results at the previous frame for detecting cells at the current frame. The experimental results demonstrate the effectiveness of each module, and the proposed method achieved the best performance in comparison.

With the application of high-density neural probes, a neuron can be detected by multiple adjacent probes, and the traditional single-channel spike sorting is no longer suitable. In this paper, we propose a five-channel weighted real-time spike sorting algorithm based on template-matching to process neural signals recorded by high-density probes. This work uses the signals of the center channel and the adjacent four channels to form a five-channel template by weighting, and employs a modified OSort algorithm with unsupervised learning to update the template. We implemented automatic online spike sorting, and tested it with both ground truth recordings and simulated datasets. The experiments show that our algorithm utilizing the information of adjacent channels has a higher sorting accuracy than traditional single-channel spike sorting. The average sorting accuracy reaches 89%, compared to 78% for single-channel.

The two most common evaluators for CT scan denoising are Peak Signal to Noise Ratio (PSNR) and Structural Similarity Index (SSIM). This paper offers an alternative evaluator by utilizing of Natural Image Quality Evaluator (NIQE) assessment to determine the performance of denoising work on noise artefact. The noise artefact was obtained during the cancer screening process and had a particular noise density pattern across the image. NIQE is one of the blind image assessments which rely on the measurable deviation of image patch as a reference; it can determine the improved quality of denoising image. Due to the method of comparison in NIQE, the two parameters: patch size and sharpness threshold, will play an essential part in getting the score compared with the result from the other evaluators (PSNR and SSIM).

Surface electromyogram (EMG) can be employed as an interface signal for various devices and software via pattern recognition. In EMG-based pattern recognition, the classifier should not only be accurate, but also output an appropriate confidence (i.e., probability of correctness) for its prediction. If the confidence accurately reflects the likelihood of true correctness, then it will be useful in various application tasks, such as motion rejection and online adaptation. The aim of this paper is to identify the types of classifiers that provide higher accuracy and better confidence in EMG pattern recognition. We evaluate the performance of various discriminative and generative classifiers on four EMG datasets, both visually and quantitatively. The analysis results show that while a discriminative classifier based on a deep neural network exhibits high accuracy, it outputs a confidence that differs from true probabilities. By contrast, a scale mixture model-based classifier, which is a generative classifier that can account for uncertainty in EMG variance, exhibits superior performance in terms of both accuracy and confidence.

The identification of fetal-head standard planes (FHSPs) from ultrasound (US) images is of fundamental importance to visualize cerebral structures and diagnose neural anomalies during gestation in a standardized way. To support the activity of healthcare operators, deep-learning algorithms have been proposed to classify these planes. To date, the translation of such algorithms in clinical practice is hampered by several factors, including the lack of large annotated datasets to train robust and generalizable algorithms. This paper proposes an approach to generate synthetic FHSP images with conditional generative adversarial network (cGAN), using class activation maps (CAMs) obtained from FHSP classification algorithms as cGAN conditional prior. Using the largest publicly available FHSP dataset, we generated realistic images of the three common FHSPs: trans-cerebellum, trans-thalamic and trans-ventricular. The evaluation through t-SNE shows the potential of the proposed approach to attenuate the problem of limited availability of annotated FHSP images.

Good quality (annotated) data is one of the most important aspects of supervised deep learning. Tasks such as semantic segmentation have a huge data requirement in exchange for only satisfactory performance. Large-scale annotations spread across multiple annotators tends to create inconsistencies, as there are various manual and semi-automated techniques involved. This mandates an external evaluator or expert to check and narrow down the problematic annotations. Studies have shown that even marking a few instances wrong in classification can lead to a significant performance drop in the model (mislabeling only 10% of one class can degrade the total performance of all classes by up to 10%). It has been noticed that fault localization by a medical expert is one of the most expensive and time-consuming processes. In this paper, we propose a novel framework for detecting the inconsistencies in the annotation of every object/anatomy in a specific image. We leverage the power of semi-supervised deep learning models (STCN) to help produce high-quality data for AI segmentation algorithms. Evaluation using this algorithm has been shown to reduce annotation review time by at least 5 hours for just 1000 images, and the quality of ground truth data improved thereby increasing the performance of the model by almost 3%.

Automatic diagnosis of eye diseases from retinal fundus images is quite challenging. Common public datasets include images of subjects with multiple diseases with uneven distribution of labels. Rare diseases are especially challenging due to their under-representation in such datasets. In this paper, we propose a training pipeline for the multi-labeled classification with uneven distribution of the sample size and sample difficulty. First, we guide the training of the initial model by weighing the training loss using an inverse-frequency for each class. This will balance the training on over-represented and under-represented samples. We then adjust the class weights using the aggregated loss for each class, and train for more iterations. In this way, the model at each iteration will focus more on difficult samples and cover the shortcomings of the previous model. Finally, we ensemble together all the models using out proposed Heuristic Stacking algorithm for improving multi-label predictions beyond simple averaging. Our experimental results on the Retinal Image Analysis for Multi-Disease Detection(RIADD)-2021 challenge dataset show that the proposed approach achieves a 88.24 % accuracy score, which is competitive with the top three ranked methods of the competition. Furthermore, we perform ablation study to stress-test our Heuristic Stacking ensemble methods versus classical methods such as bagging n multi-label classification problems.

Cardiac pacing is a clinical therapy widely used for treating irregular heart rhythms. Equivalent techniques for the treatment of gastric functional motility disorders hold great potential. Accurate analysis of pacing studies is often hindered by the stimulus artifacts which are superimposed on the recorded signals. This paper presents a semi-automated artifact detection method using a Hampel filter accompanied by 2 separate artifact reconstruction methods: (i) an auto-regressive model, and (ii) weighted mean to estimate the underlying signal. The developed framework was validated on synthetic and experimental signals containing large periodic pacing artifacts alongside evoked bioelectrical events. The performance of the proposed algorithms was quantified for gastric and cardiac pacing data collected in vivo. A lower mean RMS difference was achieved by the artifact segment reconstructed using the auto-regression ([Formula: see text]), method compared to the weighted mean ([Formula: see text]) method. Therefore, a more accurate artifact reconstruction was provided by the auto-regression approach. Clinical Relevance- The ability to efficiently and accurately isolate evoked bioelectrical events by eliminating large artifacts is a critical advancement for the analysis of paced recordings. The developed framework allows more efficient analysis of preclinical pacing data and thereby contributes to the advancement of pacing as a clinical therapy.

This paper presents a novel approximation unit added to the conventional spike processing chain which provides an appreciable reduction of complexity of the high-hardware cost feature extractors. The use of the Taylor polynomial is proposed and modelled employing its cascaded derivatives to non-uniformly capture the essential samples in each spike for reliable feature extraction and sorting. Inclusion of the approximation unit can provide 3X compression (i.e. from 66 to 22 samples) to the spike waveforms while preserving their shapes. Detailed spike waveform sequences based on in-vivo measurements have been generated using a customized neural simulator for performance assessment of the approximation unit tested on six published feature extractors. For noise levels σ N between 0.05 and 0.3 and groups of 3 spikes in each channel, all the feature extractors provide almost same sorting performance before and after approximation. The overall implementation cost when including the approximation unit and feature extraction shows a large reduction (i.e. up to 8.7X) in the hardware costly and more accurate feature extractors, offering a substantial improvement in feature extraction design.

Accurate breast lesion segmentation in ultrasound images helps radiologists to make exact diagnoses and treatments, which is important to increase the survival rate of breast cancer patients. Recently, deep learning-based methods have demonstrated remarkable results in breast lesion segmentation. However, the blurry breast lesion boundaries and noise artifacts in ultrasound images still limit the performance of the deep learning-based methods. In this paper, we propose a novel segmentation network equipped with a focal self-attention block for improving the performance of breast lesion segmentation. The focal self-attention block can incorporate fine-grained local and coarse-grained global information. The fine-grained local information is useful to enhance features of breast lesion boundaries, while the coarse-grained global information effectively reduces noise interference. To verify the performance of our network, we implement breast lesion segmentation on our collected dataset of 9836 ultrasound images. The results demonstrate that the focal self-attention block enhances features of breast lesion boundaries and improves the accuracy of breast lesion segmentation.

Multi-class segmentation of vertebrae and inter-vertebral discs (IVDs) is crucial for the diagnosis and treatment of spinal diseases. However, it is still a challenge due to similarities between neighboring vertebrae of a subject and differences among the IVDs from different subjects. In this paper, we propose a novel spine segmentation framework to achieve automatic segmentation of vertebrae and IVDs in MR images. The core component of the new framework is a Multi-View GCN (MVGCN), which utilizes multi-view features and graph convolutional network (GCN) to reason about the relations of vertebrae and IVDs. We additionally use a boundary constraint for better segmentation of the boundary between vertebrae and IVDs. We test our method on a public spine dataset of 172 MR volumetric images for the vertebrae and IVDs segmentation. The experimental results demonstrate the efficacy of our method. Code and models of our method will be available in the future.

Automatic discrimination of clean and deformed segments of capnogram signals is an essential requisite in capnogram-based respiratory assessment. However, improving the performance of this classification task remains challenging, particularly in terms of specificity and sensitivity. The goal of this paper is to address this issue by proposing a cooperative classification approach rather than relying solely on a single classifier. The presented method's main advantage is the vote participation of four distinct classifiers that affects the reliability of the final classification decision. MATLAB simulation was run on a dataset consisting of 200 15-seconds capnogram segments, 100 of which are clean and 100 are deformed. The results revealed a trade-off between the achieved specificity and sensitivity by adjusting the strictness of voting. Being highly strict in the sense of classifying a capnogram segment as clean if and only if all voting classifiers agreed on deciding so, provided specificity and sensitivity of 94% and 81%, respectively. On the contrary, lowering the strictness of voting by considering only one positive vote is sufficient to eventually classify the query capnogram segment as non-deformed gave specificity and sensitivity of 74% and 94%, respectively.

A growing area of mental health research pertains to how an individual's degree of depression might be automatically assessed through analyzing multimodal-based objective markers. However, when combined with machine learning, this research can be challenging due to the existence of unaligned multimodal sequences and the limited amount of annotated training data. In this paper, a novel cross-modal framework for automatic depression severity assessment is proposed. The low-level descriptions (LLDs) from multiple clues (such as text, audio and video) are extracted, after which multimodal fusion via cross-modal attention mechanism is utilized to facilitate the learning of more accurate feature representations. For the features extracted from each modality, the cross-modal attention mechanism is utilized to continuously update the input sequence of the target mode, until the score of the patient's health questionnaire (PHQ-8) can finally be obtained. Moreover, Self-Attention Generative Adversarial Networks (SAGAN) is employed to increase the amount of training data available for depression severity analysis. Experimental results on the depression sub-challenge dataset of the Audio/Visual Emotion Challenge (AVEC 2017 and AVEC 2019) demonstrate the effectiveness of our proposed method.

Automatic surgical phase recognition plays a key role in surgical workflow analysis and overall optimization in clinical work. In the complicated surgical procedures, similar inter-class appearance and drastic variability in phase duration make this still a challenging task. In this paper, a spatio-temporal transformer is proposed for online surgical phase recognition with different granularity. To extract rich spatial information, a spatial transformer is used to model global spatial dependencies of each time index. To overcome the variability in phase duration, a temporal transformer captures the multi-scale temporal context of different time indexes with a dual pyramid pattern. Our method is thoroughly validated on the public Cholec80 dataset with 7 coarse-grained phases and the CATARACTS2020 dataset with 19 fine-grained phases, outperforming state-of-the-art approaches with 91.4% and 84.2% accuracy, taking only 24.5M parameters.

Multiomics data integration is key for cancer prediction as it captures different aspects of molecular mechanisms. Nevertheless, the high-dimensionality of multi-omics data with a relatively small number of patients presents a challenge for the cancer prediction tasks. While feature selection techniques have been widely used to tackle the curse of dimensionality of multi-omics data, most existing methods have been applied to each type of omics data separately. In this paper, we propose a multi-agent architecture for feature selection, called MAgentOmics, to consider all omics data together. MAgentOmics extends the ant colony optimization algorithm to multi-omics data, which iteratively builds candidate solutions and evaluates them. Moreover, a new fitness function is introduced to assess the candidate feature subsets without using prediction target such as survival time of patients. Therefore, it can be considered as an unsupervised method. We evaluate the performance of MAgentOmics on the TCGA ovarian cancer multi-omics data from 176 patients using a 5-fold cross-validation. The results demonstrate that the integration power of MAgentOmics is relatively better than the state-of-the-art supervised multi-view method. The code is publicly available at https://github.com/SinaTabakhi/MAgentOmics. Clinical relevance- Discovering knowledge in existing multi-omics datasets through better feature selection enhances the clinical understanding of cancers and speeds-up decision-making in the clinic.

Crosstalk is the result of the propagation of muscle electrical signals on surface electromyogram channels simultaneously. The objective of this paper is to study the behavior of three blind source separation (BSS) methods for crosstalk reduction during finger extensor muscle contractions: FastICA, joint diagonalization of covariance matrices and optimal filtering. These methods have been tested on artificial mixtures defined by a temporal sum of the real signals from isolated contraction of two independent biomechanical muscles for the extension of the index and little finger. Artificial mixtures display a ground truth for comparison between the methods. The separation was better using the optimal filtering compared to the other two methods. The optimal filtering have then be tested on real mixtures recorded during a simultaneous contraction of the two muscles. The results are less satisfactory but open doors to new perspectives.

This paper aims to present an approach based on Reinforcement Learning (RL) concept to detect contaminants' type and minimize their effect on surface electromyography signal (sEMG) based movement recognition. The referred method was applied in the pre-processing stage of a sEMG based motion classification system using the Ninapro database 2 artificially contaminated with electrocardiography (ECG) interference, motion artifact (MOA), powerline interference (PLI) and additive white Gaussian noise (WGN). Support Vector Machine was the method for movement classification. The results showed an improvement of 8.9%, 16.7%, 15.9%, 16.5%, and 11.9% in the movement recognition accuracy with the application of the pre-processing algorithm to restore, respectively, one, three, six, nine, and 12 contaminated channels.

Quantitative Ultrasound (QUS) provides important information about the tissue properties. QUS parametric image can be formed by dividing the envelope data into small overlapping patches and computing different speckle statistics such as parameters of the Nakagami and Homodyned K-distributions (HK-distribution). The calculated QUS parametric images can be erroneous since only a few independent samples are available inside the patches. Another challenge is that the envelope samples inside the patch are assumed to come from the same distribution, an assumption that is often violated given that the tissue is usually not homogenous. In this paper, we propose a method based on Convolutional Neural Networks (CNN) to estimate QUS parametric images without patching. We construct a large dataset sampled from the HK-distribution, having regions with random shapes and QUS parameter values. We then use a well-known network to estimate QUS parameters in a multi-task learning fashion. Our results confirm that the proposed method is able to reduce errors and improve border definition in QUS parametric images.

Ultrasound exam output of large organs like liver has traditionally been limited to still images or 2D cine loops of key structures, without 3D context. Having 3D context for follow up studies makes ultrasound scanning much easier and for interventional applications such as biopsy. 3D context will reduce wrong sample selection thereby increasing patient comfort. As of today, there is no existing solution which provides 3D anatomical context to users during scanning for large organs like liver. Even for routine measurements like liver volume, patients have to undergo CT or MR scan. In this paper, we propose a novel approach to build-patient specific 3D anatomical surface models from B-mode ultrasound images and tracking information from position sensors. The complexity of the problem stems from the fact that liver boundaries are often not very clear in ultrasound images, in addition to large variability in liver size and shape across patients. Our work uses state-of-the-art deep learning algorithms to detect surface landmarks of liver followed by registering a geometric model to surface point cloud to build patient specific 3D liver model. Further, the developed models will be used to guide users to right lesion locations during the interventional procedure. Our proposed semi -automated workflow ensures the accuracy of the developed models are within acceptable limits for the targeted problem.

Synthetic medical images have an important role to play in developing data-driven medical image processing systems. Using a relatively small amount of patient data to train generative models that can produce an abundance of additional samples could bridge the gap towards big-data in niche medical domains. These generative models are evaluated in terms of the synthetic data they generate using the Visual Turing Test (VTT), Fréchet Inception Distance (FID), and other metrics. However, these are generally interpreted at the group level, and do not measure the artificiality of individual synthetic images. The present study attempts to address the challenge of automatically identifying artificial images that are obviously-artificial-looking, which may be necessary for filtering out poorly constructed synthetic images that might otherwise deteriorate the performance of assimilating systems. Synthetic computed tomography (CT) images from a progressively-grown generative adversarial network (PGGAN) were evaluated with a VTT and their image embeddings were analyzed for correlation with artificiality. Images categorized as obviously-artificial (≥0. 7 probability of being rated as fake) were classified using a battery of algorithms. The top-performing classifier, a support vector machine, exhibited accuracy of 75.5%, sensitivity of 0.743, and specificity of 0.769. This is an encouraging result that suggests a potential approach for validating synthetic medical image datasets. Clinical Relevance - Next-generation medical AI systems for image processing will utilize synthetic images produced by generative models. This paper presents an approach towards verifying artificial image legibility for quality-control before being deployed for these purposes.

Cardiac magnetic resonance (CMR) imaging is the one of the gold standard imaging modalities for the diagnosis and characterization of cardiovascular diseases. The clinical cine protocol of the CMR typically generates high-resolution 2D images of heart tissues in a finite number of separated and independent 2D planes, which are appropriate for the 3D reconstruction of biventricular heart surfaces. However, they are usually inadequate for the whole-heart reconstruction, specifically for both atria. In this regard, the paper presents a novel approach for automated patient-specific 3D whole-heart mesh reconstruction from limited number of 2D cine CMR slices with the help of a statistical shape model (SSM). After extracting the heart contours from 2D cine slices, the SSM is first optimally fitted over the sparse heart contours in 3D space to provide the initial representation of the 3D whole-heart mesh, which is further deformed to minimize the distance from the heart contours for generating the final reconstructed mesh. The reconstruction performance of the proposed approach is evaluated on a cohort of 30 subjects randomly selected from the UK Biobank study, demonstrating the generation of high-quality 3D whole-heart meshes with average contours to surface distance less than the underlying image resolution and the clinical metrics within acceptable ranges reported in previous literature. Clinical Relevance- Automated patient-specific 3D whole-heart mesh reconstruction has numerous applications in car-diac diagnosis and multimodal visualization, including treatment planning, virtual surgery, and biomedical simulations.

Two major challenges in applying deep learning to develop a computer-aided diagnosis of fundus images are the lack of enough labeled data and legal issues with patient privacy. Various efforts are being made to increase the amount of data either by augmenting training images or by synthesizing realistic-looking fundus images. However, augmentation is limited by the amount of available data and it does not address the patient privacy concern. In this paper, we propose a Generative Adversarial Network-based (GAN-based) fundus image synthesis method (Fundus GAN) that generates synthetic training images to solve the above problems. Fundus GAN is an improved way of generating retinal images by following a two-step generation process which involves first training a segmentation network to extract the vessel tree followed by vessel tree to fundus image-to-image translation using unsupervised generative attention networks. Our results show that the proposed Fundus GAN outperforms state of the art methods in different evaluation metrics. Our results also validate that generated retinal images can be used to train retinal image classifiers for eye diseases diagnosis. Clinical Relevance- Our proposed method Fundus GAN helps in solving the shortage of patient privacy-preserving training data in developing algorithms for automating image- based eye disease diagnosis. The proposed two-step GAN- based image synthesis can be used to improve the classification accuracy of retinal image classifiers without compromising the privacy of the patient.

Magnetic resonance imaging (MRI) is a powerful imaging modality that revolutionizes medicine and biology. The imaging speed of high -dimensional MRI is often limited, which constrains its practical utility. Recently, low-rank tensor models have been exploited to enable fast MR imaging with sparse sampling. Most existing methods use some pre-defined sampling design, and active sensing has not been explored for low-rank tensor imaging. In this paper, we introduce an active low-rank tensor model for fast MR imaging. We propose an active sampling method based on a Query-by-Committee model, making use of the benefits of low-rank tensor structure. Numerical experiments on a 3-D MRI data set with Cartesian sampling designs demonstrate the effectiveness of the proposed method.

Traditional methods for the development of a neuroprosthesis to perform closed-loop stimulation can be complex and the necessary technical knowledge and experience often present a high barrier for adoption. This paper takes a novel Model-Based Design approach to simplifying such closed-loop system development, and thereby lowering the adoption barrier. This work implements a computational model of different spike detection algorithms in Simulink® and compares their performances by taking advantage of synthetic neural signals to evaluate suitability for the intended embedded implementation. Clinical Relevance--- Closed-loop systems have been demonstrated to be suitable for brain repair strategies. Coupling two different brain areas by means of a neuroprosthesis can potentially lead to restoration of communication by inducing activity-dependent plasticity.

Automatic interpretation of cluster structure in rapidly arriving data streams is essential for timely detection of interesting events. Human activities often contain bursts of repeating patterns. In this paper, we propose a new relative of the Visual Assessment of Cluster Tendency (VAT) model, to interpret cluster evolution in streaming activity data where shapes of recurring patterns are important. Existing VAT algorithms are either suitable only for small batch data and unscalable to rapidly evolving streams, or cannot capture shape patterns. Our proposed incremental algorithm processes streaming data in chunks and identifies repeating patterns or shapelets from each chunk, creating a Dictionary-of-Shapes (DoS) that is updated on the fly. Each chunk is transformed into a lower dimensional representation based on it's distance from the shapelets in the current DoS. Then a small set of transformed chunks are sampled using an intelligent Maximin Random Sampling (MMRS) scheme, to create a scalable VAT image that is incrementally updated as the data stream progresses. Experiments on two upper limb activity datasets demonstrate that the proposed method can successfully and efficiently visualize clusters in long streams of data and can also identify anomalous movements.

As advances in microscopy imaging provide an ever clearer window into the human brain, accurate reconstruction of neural connectivity can yield valuable insight into the relationship between brain structure and function. However, human manual tracing is a slow and laborious task, and requires domain expertise. Automated methods are thus needed to enable rapid and accurate analysis at scale. In this paper, we explored deep neural networks for dense axon tracing and incorporated axon topological information into the loss function with a goal to improve the performance on both voxel-based segmentation and axon centerline detection. We evaluated three approaches using a modified 3D U-Net architecture trained on a mouse brain dataset imaged with light sheet microscopy and achieved a 10% increase in axon tracing accuracy over previous methods. Furthermore, the addition of centerline awareness in the loss function outperformed the baseline approach across all metrics, including a boost in Rand Index by 8%.

In this paper, we propose and validate an automatic pipeline for subcortical surface generation by making use of the average symmetrical surface distance (ASSD) loss in active shape modeling (ASM). A group of template surfaces are first generated via large deformation diffeomorphic metric mapping based surface deformation. ASM is then employed to obtain the mean shape and shape variation parameters of the template surfaces. To obtain the optimal shape variation parameters which best fit the target structure after acting upon the mean shape, a recently proposed derivative-free optimization method (the slow-fast learning method) is adopted. The ASSD loss, in addition to the typically utilized Dice similarity coefficient loss, is employed during the learning process to help enhance the boundary accuracy. We successfully validate the importance of the ASSD loss through ablation analysis. In addition, we show the effectiveness of the slow-fast learning method by comparing it with other state-of-the-art derivative-free optimization algorithms. Our proposed pipeline is found to be capable of yielding subcortical surfaces with high accuracy, correct anatomical topology, and sufficient smoothness. Clinical Relevance- This work provides a useful tool for generating subcortical surfaces which are important biomarkers for a variety of brain disorders.

Considering user-specific settings is known to enhance Brain-Computer Interface (BCI) performances. In particular, the optimal frequency band for oscillatory activity classification is highly user-dependent and many frequency band selection methods have been developed in the past two decades. However, it is not well studied whether those conventional methods can be efficiently applied to the Riemannian BCIs, a recent family of BCI systems that utilize the non-Euclidean nature of the data unlike conventional BCI pipelines. In this paper, we proposed a novel frequency band selection method working on the Riemannian manifold. The frequency band is selected considering the class distinctiveness as quantified based on the inter-class distance and the intra-class variance on the manifold. An advantage of this method is that the frequency bandwidth can be adjusted for each individual without intensive optimization steps. In a comparative experiment using a public dataset of motor imagery-based BCI, our method showed a substantial improvement in average accuracy over both a fixed broad frequency band and a popular conventional frequency band selection method. In particular, our method substantially improved performances for subjects with initially low accuracies. This preliminary result suggests the importance of developing new user-specific setting algorithms considering the manifold properties, rather than directly applying methods developed prior to the rise of the Riemannian BCIs.

Experienced radiologists can accurately diagnose relevant diseases by observing the cardiopulmonary region in chest X-ray images. Advances in deep learning techniques enable the prediction of lesions in chest X-ray images. However, deep learning-based algorithms usually require a large amount of training data, and it lacks an effective method for data generation and augmentation. In this paper, we propose a Lung Segmentation Reconstruction (LSR) module. A healthy chest X-ray image is generated based on the abnormal image as a reference. With the generated healthy reference, we propose a novel way of data augmentation for chest X-ray images. The whole images, lung regions and abnormal regions are stacked together and fed into a classification model to make a credible diagnosis. Extensive experiments have been conducted on the public dataset CheXpert. Experimental results show that our proposed abnormality enhancement can help the baseline models achieve better performance on consolidation and pleural effusion. These results highlight the potential value of the large number of healthy chest X-ray images in the dataset and the combination of different regions of chest X-ray images for prediction.

Molecular imaging has long been recognized as an important tool for diagnosis, characterization, and monitoring of treatment responses of brain tumors. Magnetic resonance spectroscopic imaging (MRSI) is a label-free molecular imaging technique capable of mapping metabolite distributions non-invasively. Several metabolites detectable by MRSI, including Choline, Lactate and N-Acetyl Aspartate, have been proved useful biomarkers for brain tumor characterization. However, clinical application of MRSI has been limited by poor resolution, small spatial coverage, low signal-to-noise ratio and long scan time. This work presents a novel MRSI method for fast, high-resolution metabolic imaging of brain tumor. This method synergistically integrates fast acquisition sequence, sparse sampling, subspace modeling and machine learning to enable 3D mapping of brain metabolites with a spatial resolution of 2.0×3.0×3.0 mm 3 in a 7-minute scan. Experimental results obtained from patients with diagnosed brain tumor showed great promise for capturing small-size tumors and revealing intra-tumor metabolic heterogeneities.Clinical Relevance - This paper presents a novel technique for label-free molecular imaging of brain tumor. With further development, this technology may enable many potential clinical applications, from tumor detection, characterization, to assessment of treatment efficacy.

Electrocardiogram (ECG) signal is one of the most important methods for diagnosing cardiovascular diseases but is usually affected by noises. Denoising is therefore necessary before further analysis. Deep learning-related methods have been applied to image processing and other domains with great success but are rarely used for denoising ECG signals. This paper proposes an effective and simple model of encoder-decoder structure for denoising ECG signals (APR-CNN). Specifically, Adaptive Parametric ReLU (APReLU) and Dual Attention Module (DAM) are introduced in the model. Rectified Linear Unit (ReLU) is replaced with the APReLU for better negative information retainment. The DAM is an attention-based module consisting of a channel attention module and spatial attention module, through which the inter-spatial and inter-channel relationship of the input data are exploited. We tested our model on the MIT-BIH dataset, and the results show that the APR-CNN can handle ECG signals with a different signal-to-noise ratio (SNR). The comparative experiment proves our model is better than other deep learning and traditional methods.Clinical Relevance- This paper proposed a method capable of denoising ECG signals with strong noise to alleviate difficulties for further medical analysis.

Simulations of electroneurogram recording could help find the optimal set of electrodes and algorithms for selective neural recording. However, no flexible methods are established for selective neural recording as for neural stimulation. This paper proposes a method to couple a compartmental and a FEM nerve model, implemented in NEURON and COMSOL, respectively, to translate Node of Ranvier currents into extraneural electric fields. The study simulate ex-vivo experimental conditions, and the method allows flexibility in electrode geometries and nerve topologies. This model has been made available in a public repository 4 . So far, the model behavior complies with available experimental results and expectations from literature. There is good agreement in terms of signal amplitude and waveform, and computational times are acceptable, leaving room for flexible simulation studies complementary to animal tests.

Automated medical skill assessment facilitates medical education by merging varying clinical experiences across instructors for standardizing medical training. However, medical datasets for training such automated assessment rarely have satisfactory sizes due to the cost of data collection, safety concerns and privacy restrictions. Current medical training relies on evaluation rubrics that usually include multiple auxiliary labels to support the overall evaluation from varying aspects of the procedure. In this paper, we explore machine learning algorithms to design a generalizable auxiliary task-based framework for medical skill assessment to address training automated systems with limited data. Our framework exhaustively mines valid auxiliary information in the evaluation rubric to pre-train the feature extractor before training the skill assessment classifier. Notably, a new regression-based multitask weighting method is the key to pre-train a meaningful feature representation comprehensively, ensuring the evaluation rubric is well imitated in the final model. The overall evaluation task can be fine-tuned based on the pre-trained rubric-based feature representation. Our experimental results on two medical skill datasets show that our work can significantly improve performance, achieving 85.9% and 97.4% accuracy in the intubation dataset and surgical skill dataset, respectively.

Techniques for 3D endoscopic systems have been widely studied for various reasons. Among them, active stereo based systems, in which structured-light patterns are projected to surfaces and endoscopic images of the pattern are analyzed to produce 3D depth images, are promising, because of robustness and simple system configurations. For those systems, finding correspondences between a projected pattern and an original pattern is an open problem. Recently, correspondence estimation by graph neural networks (GCN) using graph-based representation of the patterns were proposed for 3D endoscopic systems. One severe problem of the approach is that the graph matching by GCN is largely affected by the stability of the graph construction process using the detected patterns of a captured image. If the detected pattern is fragmented into small pieces, graph matching may fail and 3D shapes cannot be retrieved. In this paper, we propose a solution for those problems by applying deep-layered GCN and extended graph representations of the patterns, where proximity information is added. Experiments show that the proposed method outperformed the previous method in accuracies for correspondence matching for 3D reconstruction.

Traditional biometrics such as face, iris and fingerprint have been applied widely nowadays. Nevertheless, with more and more potential problems being exposed, such as privacy leak and fabricate attack, it is urgent to find new secure biometrics to meet the needs. Identification based on brain signals is a promising option due to its unique advantages of confidentiality, anti-spoofing, continuity and cancelability. Among various types of brain signals, local field potential (LFP) has long term stability, high signal to noise ratio and high spatial resolution, which is suitable for identification. In this paper, we propose a novel biometric which is extracted from LFP signals with a deep neural network. The proposed biometric can be generated in a task-related manner thus is cancelable. Experiments with ten rats demonstrate that, the proposed biometric achieves a high identification accuracy of 94.47%, and the performance is stable over several days.

The purpose of the study described in this paper is to shed more light on disease similarities by analyzing the relationship between categorical proximity of diseases in human-curated ontologies and structural proximity of the related disease module (DM) in the interactome. We propose a methodology (and related algorithms) to automatically induce a hierarchical structure from proximity relations between DMs, and to compare this structure with a human-curated disease taxonomy.Clinical relevance- Disease ontologies are extensively used for diagnostic evaluation and clinical decision support but still reflect the clinical reductionist perspective. We demonstrate that the proposed network-based methodology allows us to analyze commonalities and differences among structural and categorical similarity of human diseases, help refine human disease classification systems, and identify promising network areas where new disease-gene interactions can be discovered.

Fluorescent Molecular Tomography (FMT) is a highly sensitive and noninvasive imaging method that provides three-dimensional distribution of biomarkers by noninvasive detection of fluorescent marker probes. However, due to the light scattering effect and ill-posedness of inverse problems, it is challenging to develop an efficient construction method that can provide the exact location and morphology of the fluorescence distribution. In this paper, we proposed L 1 -L 2 norm regularization to improve FMT reconstruction. In our research, proximal operators of non-convex L 1 -L 2 norm and forward-backward splitting method was adopted to solve the inverse problem of FMT. Simulation results on heterogeneous mouse model demonstrated that the proposed FBS method is superior to IVTCG, DCA and IRW-L 1/2 reconstruction methods in location accuracy and other aspects.

In this paper, the recently developed method of, Filtered Delay Euclidian-Weighted Multiply and Sum (F-DewMAS), is newly investigated for Conventional Focused Beamforming (CFB) technique. The performance of F-DewMAS method was compared with the established Delay and Sum (DAS) method and the popular non-linear beamforming method of F-DMAS. The different methods of F-DewMAS, F-DMAS, and DAS were compared in terms of the resulting image quality metrics, Lateral Resolution (LR), Axial Resolution (AR), Contrast Ratio (CR) and contrast-to-noise ratio (CNR), in experiments on Nylon point scatterer and CIRS Triple modality Abdominal phantoms. Experimental results show that F-DewMAS resulted in improvements of AR by 35.56% and 25.33%, LR by 42.97 % and 31.05 % and CR by 119.94% and 61.46% compared to those obtained using DAS and F-DMAS, respectively. The CNR of F-DewMAS is 46.33 % more compared to F-DMAS. Hence, it can be concluded that the image quality is improved appreciably by F-DewMAS compared to DAS and F-DMAS.Clinical Relevance-The developed method can improve the resolution and contrast of the image, which results in better visualization of finer details and thus may aid in better diagnosis.