Automatic labeling of Parkinson's Disease gait videos with weak supervision.

Motor dysfunction in Parkinson's Disease (PD) patients is typically assessed by clinicians employing the Movement Disorder Society's Unified Parkinson's Disease Rating Scale (MDS-UPDRS). Such comprehensive clinical assessments are time-consuming, expensive, semi-subjective, and may potentially result in conflicting labels across different raters. To address this problem, we propose an automatic, objective, and weakly-supervised method for labeling PD patients' gait videos. The proposed method accepts videos of patients and classifies their gait scores as normal (Gait score in MDS-UPDRS = 0) or PD (MDS-UPDRS ≥ 1). Unlike previous work, the proposed method does not require a priori MDS-UPDRS ratings for training, utilizing only domain-specific knowledge obtained from neurologists. We propose several labeling functions that classify patients' gait and use a generative model to learn the accuracy of each labeling function in a self-supervised manner. Since results depended upon the estimated values of the patients' 3D poses, and existing pre-trained 3D pose estimators did not yield accurate results, we propose a weakly-supervised 3D human pose estimation method for fine-tuning pre-trained models in a clinical setting. Using leave-one-out evaluations, the proposed method obtains an accuracy of 89% on a dataset of 29 PD subjects - a significant improvement compared to previous work by 7%-10% depending upon the dataset. The method obtained state-of-the-art results on the Human3.6M dataset. Our results suggest that the use of labeling functions may provide a robust means to interpret and classify patient-oriented videos involving motor tasks.

Latest Developments in Adapting Deep Learning for Assessing TAVR Procedures and Outcomes.

Aortic valve defects are among the most prevalent clinical conditions. A severely damaged or non-functioning aortic valve is commonly replaced with a bioprosthetic heart valve (BHV) via the transcatheter aortic valve replacement (TAVR) procedure. Accurate pre-operative planning is crucial for a successful TAVR outcome. Assessment of computational fluid dynamics (CFD), finite element analysis (FEA), and fluid-solid interaction (FSI) analysis offer a solution that has been increasingly utilized to evaluate BHV mechanics and dynamics. However, the high computational costs and the complex operation of computational modeling hinder its application. Recent advancements in the deep learning (DL) domain can offer a real-time surrogate that can render hemodynamic parameters in a few seconds, thus guiding clinicians to select the optimal treatment option. Herein, we provide a comprehensive review of classical computational modeling approaches, medical imaging, and DL approaches for planning and outcome assessment of TAVR. Particularly, we focus on DL approaches in previous studies, highlighting the utilized datasets, deployed DL models, and achieved results. We emphasize the critical challenges and recommend several future directions for innovative researchers to tackle. Finally, an end-to-end smart DL framework is outlined for real-time assessment and recommendation of the best BHV design for TAVR. Ultimately, deploying such a framework in future studies will support clinicians in minimizing risks during TAVR therapy planning and will help in improving patient care.

PPG2ECGps: An End-to-End Subject-Specific Deep Neural Network Model for Electrocardiogram Reconstruction from Photoplethysmography Signals without Pulse Arrival Time Adjustments.

Electrocardiograms (ECGs) provide crucial information for evaluating a patient's cardiovascular health; however, they are not always easily accessible. Photoplethysmography (PPG), a technology commonly used in wearable devices such as smartwatches, has shown promise for constructing ECGs. Several methods have been proposed for ECG reconstruction using PPG signals, but some require signal alignment during the training phase, which is not feasible in real-life settings where ECG signals are not collected at the same time as PPG signals. To address this challenge, we introduce PPG2ECGps, an end-to-end, patient-specific deep-learning neural network utilizing the W-Net architecture. This novel approach enables direct ECG signal reconstruction from PPG signals, eliminating the need for signal alignment. Our experiments show that the proposed model achieves mean values of 0.977 mV for Pearson's correlation coefficient, 0.037 mV for the root mean square error, and 0.010 mV for the normalized dynamic time-warped distance when comparing reconstructed ECGs to reference ECGs from a dataset of 500 records. As PPG signals are more accessible than ECG signals, our proposed model has significant potential to improve patient monitoring and diagnosis in healthcare settings via wearable devices.

Subject-Based Model for Reconstructing Arterial Blood Pressure from Photoplethysmogram.

The continuous prediction of arterial blood pressure (ABP) waveforms via non-invasive methods is of great significance for the prevention and treatment of cardiovascular disease. Photoplethysmography (PPG) can be used to reconstruct ABP signals due to having the same excitation source and high signal similarity. The existing methods of reconstructing ABP signals from PPG only focus on the similarities between systolic, diastolic, and mean arterial pressures without evaluating their global similarity. This paper proposes a deep learning model with a W-Net architecture to reconstruct ABP signals from PPG. The W-Net consists of two concatenated U-Net architectures, the first acting as an encoder and the second as a decoder to reconstruct ABP from PPG. Five hundred records of different lengths were used for training and testing. The experimental results yielded high values for the similarity measures between the reconstructed ABP signals and their reference ABP signals: the Pearson correlation, root mean square error, and normalized dynamic time warping distance were 0.995, 2.236 mmHg, and 0.612 mmHg on average, respectively. The mean absolute errors of the SBP and DBP were 2.602 mmHg and 1.450 mmHg on average, respectively. Therefore, the model can reconstruct ABP signals that are highly similar to the reference ABP signals.

Multi-Channel Vision Transformer for Epileptic Seizure Prediction.

Epilepsy is a neurological disorder that causes recurrent seizures and sometimes loss of awareness. Around 30% of epileptic patients continue to have seizures despite taking anti-seizure medication. The ability to predict the future occurrence of seizures would enable the patients to take precautions against probable injuries and administer timely treatment to abort or control impending seizures. In this study, we introduce a Transformer-based approach called Multi-channel Vision Transformer (MViT) for automated and simultaneous learning of the spatio-temporal-spectral features in multi-channel EEG data. Continuous wavelet transform, a simple yet efficient pre-processing approach, is first used for turning the time-series EEG signals into image-like time-frequency representations named Scalograms. Each scalogram is split into a sequence of fixed-size non-overlapping patches, which are then fed as inputs to the MViT for EEG classification. Extensive experiments on three benchmark EEG datasets demonstrate the superiority of the proposed MViT algorithm over the state-of-the-art seizure prediction methods, achieving an average prediction sensitivity of 99.80% for surface EEG and 90.28-91.15% for invasive EEG data.

Ethnic disparities in publicly-available pulse oximetry databases.

Inaccuracies have been reported in pulse oximetry measurements taken from people who identified as Black. Here, we identify substantial ethnic disparities in the population numbers within 12 pulse oximetry databases, which may affect the testing of new oximetry devices and impact patient outcomes.

Robust Reconstruction of Electrocardiogram Using Photoplethysmography: A Subject-Based Model.

Electrocardiography and photoplethysmography are non-invasive techniques that measure signals from the cardiovascular system. While the cycles of the two measurements are highly correlated, the correlation between the waveforms has rarely been studied. Measuring the photoplethysmogram (PPG) is much easier and more convenient than the electrocardiogram (ECG). Recent research has shown that PPG can be used to reconstruct the ECG, indicating that practitioners can gain a deep understanding of the patients' cardiovascular health using two physiological signals (PPG and ECG) while measuring only PPG. This study proposes a subject-based deep learning model that reconstructs an ECG using a PPG and is based on the bidirectional long short-term memory model. Because the ECG waveform may vary from subject to subject, this model is subject-specific. The model was tested using 100 records from the MIMIC III database. Of these records, 50 had a circulatory disease. The results show that a long ECG signal could be effectively reconstructed from PPG, which is, to our knowledge, the first attempt in this field. A length of 228 s of ECG was constructed by the model, which was trained and validated using 60 s of PPG and ECG signals. To segment the data, a different approach that segments the data into short time segments of equal length (and that do not rely on beats and beat detection) was investigated. Segmenting the PPG and ECG time series data into equal segments of 1-min width gave the optimal results. This resulted in a high Pearson's correlation coefficient between the reconstructed 228 s of ECG and referenced ECG of 0.818, while the root mean square error was only 0.083 mV, and the dynamic time warping distance was 2.12 mV per second on average.

Semisupervised Seizure Prediction in Scalp EEG Using Consistency Regularization.

Early prediction of epilepsy seizures can warn the patients to take precautions and improve their lives significantly. In recent years, deep learning has become increasingly predominant in seizure prediction. However, existing deep learning-based approaches in this field require a great deal of labeled data to guarantee performance. At the same time, labeling EEG signals does require the expertise of an experienced pathologist and is incredibly time-consuming. To address this issue, we propose a novel Consistency-based Semisupervised Seizure Prediction Model (CSSPM), where only a fraction of training data is labeled. Our method is based on the principle of consistency regularization, which underlines that a robust model should maintain consistent results for the same input under extra perturbations. Specifically, by using stochastic augmentation and dropout, we consider the entire neural network as a stochastic model and apply a consistency constraint to penalize the difference between the current prediction and previous predictions. In this way, unlabeled data could be fully utilized to improve the decision boundary and enhance prediction performance. Compared with existing studies requiring all training data to be labeled, the proposed method only needs a small portion of data to be labeled while still achieving satisfactory results. Our method provides a promising solution to alleviate the labeling cost for real-world applications.

Interpreting Bottom-Up Decision-Making of CNNs via Hierarchical Inference.

With the great success of convolutional neural networks (CNNs), interpretation of their internal network mechanism has been increasingly critical, while the network decision-making logic is still an open issue. In the bottom-up hierarchical logic of neuroscience, the decision-making process can be deduced from a series of sub-decision-making processes from low to high levels. Inspired by this, we propose the Concept-harmonized HierArchical INference (CHAIN) interpretation scheme. In CHAIN, a network decision-making process from shallow to deep layers is interpreted by the hierarchical backward inference based on visual concepts from high to low semantic levels. Firstly, we learned a general hierarchical visual-concept representation in CNN layered feature space by concept harmonizing model on a large concept dataset. Secondly, for interpreting a specific network decision-making process, we conduct the concept-harmonized hierarchical inference backward from the highest to the lowest semantic level. Specifically, the network learning for a target concept at a deeper layer is disassembled into that for concepts at shallower layers. Finally, a specific network decision-making process is explained as a form of concept-harmonized hierarchical inference, which is intuitively comparable to the bottom-up hierarchical visual recognition way. Quantitative and qualitative experiments demonstrate the effectiveness of the proposed CHAIN at both instance and class levels.

A convolutional-recurrent neural network approach to resting-state EEG classification in Parkinson's disease.

BACKGROUND: Parkinson's disease (PD) is expected to become more common, particularly with an aging population. Diagnosis and monitoring of the disease typically rely on the laborious examination of physical symptoms by medical experts, which is necessarily limited and may not detect the prodromal stages of the disease. NEW METHOD: We propose a lightweight (~20 K parameters) deep learning model to classify resting-state EEG recorded from people with PD and healthy controls (HC). The proposed CRNN model consists of convolutional neural networks (CNN) and a recurrent neural network (RNN) with gated recurrent units (GRUs). The 1D CNN layers are designed to extract spatiotemporal features across EEG channels, which are subsequently supplied to the GRUs to discover temporal features pertinent to the classification. RESULTS: The CRNN model achieved 99.2% accuracy, 98.9% precision, and 99.4% recall in classifying PD from HC. Interrogating the model, we further demonstrate that the model is sensitive to dopaminergic medication effects and predominantly uses phase information in the EEG signals. COMPARISON WITH EXISTING METHODS: The CRNN model achieves superior performance compared to baseline machine learning methods and other recently proposed deep learning model. CONCLUSION: The approach proposed in this study adequately extracts spatial and temporal features in multi-channel EEG signals that enable accurate differentiation between PD and HC. The CRNN model has excellent potential for use as an oscillatory biomarker for assisting in the diagnosis and monitoring of people with PD. Future studies to further improve and validate the model's performance in clinical practice are warranted.

PPGTempStitch: A MATLAB Toolbox for Augmenting Annotated Photoplethsmogram Signals.

An annotated photoplethysmogram (PPG) is required when evaluating PPG algorithms that have been developed to detect the onset and systolic peaks of PPG waveforms. However, few publicly accessible PPG datasets exist in which the onset and systolic peaks of the waveforms are annotated. Therefore, this study developed a MATLAB toolbox that stitches predetermined annotated PPGs in a random manner to generate a long, annotated PPG signal. With this toolbox, any combination of four annotated PPG templates that represent regular, irregular, fast rhythm, and noisy PPG waveforms can be stitched together to generate a long, annotated PPG. Furthermore, this toolbox can simulate real-life PPG signals by introducing different noise levels and PPG waveforms. The toolbox can implement two stitching methods: one based on the systolic peak and the other on the onset. Additionally, cubic spline interpolation is used to smooth the waveform around the stitching point, and a skewness index is used as a signal quality index to select the final signal output based on the stitching method used. The developed toolbox is free and open-source software, and a graphical user interface is provided. The method of synthesizing by stitching introduced in this paper is a data augmentation strategy that can help researchers significantly increase the size and diversity of annotated PPG signals available for training and testing different feature extraction algorithms.

Semi-dilated convolutional neural networks for epileptic seizure prediction.

Epilepsy is a neurological brain disorder that affects ∼75 million people worldwide. Predicting epileptic seizures holds great potential for improving the quality of life of people with epilepsy, but seizure prediction solely from the Electroencephalogram (EEG) is challenging. Classical machine learning algorithms and a variety of feature engineering methods have become a mainstay in seizure prediction, yet performance has been variable. In this work, we first propose an efficient data pre-processing method that maps the time-series EEG signals into an image-like format (a "scalogram") using continuous wavelet transform. We then develop a novel convolution module named "semi-dilated convolution" that better exploits the geometry of wavelet scalograms and nonsquare-shape images. Finally, we propose a neural network architecture named "semi-dilated convolutional network (SDCN)" that uses semi-dilated convolutions to solely expand the receptive field along the long dimension (image width) while maintaining high resolution along the short dimension (image height). Results demonstrate that the proposed SDCN architecture outperforms previous seizure prediction methods, achieving an average seizure prediction sensitivity of 98.90% for scalp EEG and 88.45-89.52% for invasive EEG.

The Effectiveness of Image Augmentation in Deep Learning Networks for Detecting COVID-19: A Geometric Transformation Perspective.

Chest X-ray imaging technology used for the early detection and screening of COVID-19 pneumonia is both accessible worldwide and affordable compared to other non-invasive technologies. Additionally, deep learning methods have recently shown remarkable results in detecting COVID-19 on chest X-rays, making it a promising screening technology for COVID-19. Deep learning relies on a large amount of data to avoid overfitting. While overfitting can result in perfect modeling on the original training dataset, on a new testing dataset it can fail to achieve high accuracy. In the image processing field, an image augmentation step (i.e., adding more training data) is often used to reduce overfitting on the training dataset, and improve prediction accuracy on the testing dataset. In this paper, we examined the impact of geometric augmentations as implemented in several recent publications for detecting COVID-19. We compared the performance of 17 deep learning algorithms with and without different geometric augmentations. We empirically examined the influence of augmentation with respect to detection accuracy, dataset diversity, augmentation methodology, and network size. Contrary to expectation, our results show that the removal of recently used geometrical augmentation steps actually improved the Matthews correlation coefficient (MCC) of 17 models. The MCC without augmentation (MCC = 0.51) outperformed four recent geometrical augmentations (MCC = 0.47 for Data Augmentation 1, MCC = 0.44 for Data Augmentation 2, MCC = 0.48 for Data Augmentation 3, and MCC = 0.49 for Data Augmentation 4). When we retrained a recently published deep learning without augmentation on the same dataset, the detection accuracy significantly increased, with a χ McNema r ' s statistic 2 = 163 . 2 and a p-value of 2.23 × 10-37. This is an interesting finding that may improve current deep learning algorithms using geometrical augmentations for detecting COVID-19. We also provide clinical perspectives on geometric augmentation to consider regarding the development of a robust COVID-19 X-ray-based detector.

Assessment of Hypertension Using Clinical Electrocardiogram Features: A First-Ever Review.

Hypertension affects an estimated 1.4 billion people and is a major cause of morbidity and mortality worldwide. Early diagnosis and intervention can potentially decrease cardiovascular events later in life. However, blood pressure (BP) measurements take time and require training for health care professionals. The measurements are also inconvenient for patients to access, numerous daily variables affect BP values, and only a few BP readings can be collected per session. This leads to an unmet need for an accurate, 24-h continuous, and portable BP measurement system. Electrocardiograms (ECGs) have been considered as an alternative way to measure BP and may meet this need. This review summarizes the literature published from January 1, 2010, to January 1, 2020, on the use of only ECG wave morphology to monitor BP or identify hypertension. From 35 articles analyzed (9 of those with no listed comorbidities and confounders), the P wave, QTc intervals and TpTe intervals may be promising for this purpose. Unfortunately, with the limited number of articles and the variety of participant populations, we are unable to make conclusions about the effectiveness of ECG-only BP monitoring. We provide 13 recommendations for future ECG-only BP monitoring studies and highlight the limited findings in pregnant and pediatric populations. With the advent of convenient and portable ECG signal recording in smart devices and wearables such as watches, understanding how to apply ECG-only findings to identify hypertension early is crucial to improving health outcomes worldwide.

PPGSynth: An Innovative Toolbox for Synthesizing Regular and Irregular Photoplethysmography Waveforms.

Photoplethysmography (PPG) is increasingly used in digital health, exceptionally in smartwatches. The PPG signal contains valuable information about heart activity, and there is lots of research interest in its means and analysis for cardiovascular diseases. Unfortunately, to our knowledge, there is no arrhythmic PPG dataset publicly available-this paper attempt to provide a toolbox that can generate synthesized arrhythmic PPG signals. The model of a single PPG pulse in this toolbox utilizes two combined Gaussian functions. This toolbox supports synthesizing PPG waveform with regular heartbeats and three irregular heartbeats: compensation, interpolation, and reset. The user can generate a large amount of PPG data with a certain irregularity, with different sampling frequency, time length, and a range of noise types (Gaussian noise and multi-frequency noise) can be added to the synthesized PPG which can all be modified from the interface, and different types of arrhythmic PPGs (as calculated by the model) generated. The generation for large PPG datasets that simulate PPG collected from real humans could be used for testing the robustness of developed algorithms that are targeting arrhythmic PPG signals. Our PPG synthesis tool is publicly available.

The Performance of Deep Neural Networks in Differentiating Chest X-Rays of COVID-19 Patients From Other Bacterial and Viral Pneumonias.

Chest radiography is a critical tool in the early detection, management planning, and follow-up evaluation of COVID-19 pneumonia; however, in smaller clinics around the world, there is a shortage of radiologists to analyze large number of examinations especially performed during a pandemic. Limited availability of high-resolution computed tomography and real-time polymerase chain reaction in developing countries and regions of high patient turnover also emphasizes the importance of chest radiography as both a screening and diagnostic tool. In this paper, we compare the performance of 17 available deep learning algorithms to help identify imaging features of COVID19 pneumonia. We utilize an existing diagnostic technology (chest radiography) and preexisting neural networks (DarkNet-19) to detect imaging features of COVID-19 pneumonia. Our approach eliminates the extra time and resources needed to develop new technology and associated algorithms, thus aiding the front-line healthcare workers in the race against the COVID-19 pandemic. Our results show that DarkNet-19 is the optimal pre-trained neural network for the detection of radiographic features of COVID-19 pneumonia, scoring an overall accuracy of 94.28% over 5,854 X-ray images. We also present a custom visualization of the results that can be used to highlight important visual biomarkers of the disease and disease progression.

Synthetic photoplethysmogram generation using two Gaussian functions.

Evaluating the performance of photoplethysmogram (PPG) event detection algorithms requires a large number of PPG signals with different noise levels and sampling frequencies. As publicly available PPG databases provide few options, artificially constructed PPG signals can also be used to facilitate this evaluation. Here, we propose a dynamic model to synthesize PPG over specified time durations and sampling frequencies. In this model, a single pulse was simulated by two Gaussian functions. Additionally, the beat-to-beat intervals were simulated using a normal distribution with a specific mean value and a specific standard deviation value. To add periodicity and to generate a complete signal, the circular motion principle was used. We synthesized three classes of pulses by emulating three different templates: excellent (systolic and diastolic waves are salient), acceptable (systolic and diastolic waves are not salient), and unfit (systolic and diastolic waves are noisy). The optimized model fitting of the Gaussian functions to the templates yielded 0.99, 0.98, and 0.85 correlations between the template and synthetic pulses for the excellent, acceptable, and unfit classes, respectively, with mean square errors of 0.001, 0.003, and 0.017, respectively. By comparing the heart rate variability of real PPG and randomly synthesized PPG for 5 min in 116 records from the MIMIC III database, strong correlations were found in SDNN, RMSSD, LF, HF, SD1, and SD2 (0.99, 0.89, 0.84, 0.89, 0.90 and 0.95, respectively).

Multimodal Photoplethysmography-Based Approaches for Improved Detection of Hypertension.

Elevated blood pressure (BP) is a major cause of death, yet hypertension commonly goes undetected. Owing to its nature, it is typically asymptomatic until later in its progression when the vessel or organ structure has already been compromised. Therefore, noninvasive and continuous BP measurement methods are needed to ensure appropriate diagnosis and early management before hypertension leads to irreversible complications. Photoplethysmography (PPG) is a noninvasive technology with waveform morphologies similar to that of arterial BP waveforms, therefore attracting interest regarding its usability in BP estimation. In recent years, wearable devices incorporating PPG sensors have been proposed to improve the early diagnosis and management of hypertension. Additionally, the need for improved accuracy and convenience has led to the development of devices that incorporate multiple different biosignals with PPG. Through the addition of modalities such as an electrocardiogram, a final measure of the pulse wave velocity is derived, which has been proved to be inversely correlated to BP and to yield accurate estimations. This paper reviews and summarizes recent studies within the period 2010-2019 that combined PPG with other biosignals and offers perspectives on the strengths and weaknesses of current developments to guide future advancements in BP measurement. Our literature review reveals promising measurement accuracies and we comment on the effective combinations of modalities and success of this technology.

Cuffless Single-Site Photoplethysmography for Blood Pressure Monitoring.

One in three adults worldwide has hypertension, which is associated with significant morbidity and mortality. Consequently, there is a global demand for continuous and non-invasive blood pressure (BP) measurements that are convenient, easy to use, and more accurate than the currently available methods for detecting hypertension. This could easily be achieved through the integration of single-site photoplethysmography (PPG) readings into wearable devices, although improved reliability and an understanding of BP estimation accuracy are essential. This review paper focuses on understanding the features of PPG associated with BP and examines the development of this technology over the 2010-2019 period in terms of validation, sample size, diversity of subjects, and datasets used. Challenges and opportunities to move single-site PPG forward are also discussed.