Quantitative feature analysis of continuous analytic wavelet transforms of electrocardiography and electromyography

Hybrid deep transfer learning-based early diagnosis of autism spectrum disorder using scalogram representation of electroencephalography signals

Early diagnosis of autism spectrum disorder (ASD) plays an important role in the rehabilitation of the patient. This goal necessitates higher-level pattern representation and a strong modeling approach. The proposed approach applies scalogram images of electroencephalography signals for the first purpose and a two-level deep learning architecture for better classification. Scalogram images embed both the temporal and spectral information of the signal. On the other hand, the hybrid deep learning hierarchy of convolutional neural network followed by long short-term memory models both spatial and temporal information of the scalogram image. The approach is evaluated on a dataset of 34 ASD samples and 11 normal cases in without-voice and with-voice conditions. To validate the early diagnosis hypothesis, signals from children older than 5 years are used as the training set, and signals from younger subjects are used as the validation set. The proposed method achieves excellent performance of 99.50% and 98.43% for automatically detecting ASD with and without voice, respectively. This classification performance is higher than most recent reported approaches, and the results show the effectiveness of the approach in early diagnosis of ASD and demonstrate the auditory impact on the diagnosis of autism.

Cardiac arrhythmia detection using deep learning approach and time frequency representation of ECG signals

BACKGROUND

Cardiac arrhythmia is a cardiovascular disorder characterized by disturbances in the heartbeat caused by electrical conduction anomalies in cardiac muscle. Clinically, ECG machines are utilized to diagnose and monitor cardiac arrhythmia noninvasively. Since ECG signals are dynamic in nature and depict various complex information, visual assessment and analysis are time consuming and very difficult. Therefore, an automated system that can assist physicians in the easy detection of arrhythmia is needed.

METHOD

The main objective of this study was to create an automated deep learning model capable of accurately classifying ECG signals into three categories: cardiac arrhythmia (ARR), congestive heart failure (CHF), and normal sinus rhythm (NSR). To achieve this, ECG data from the MIT-BIH and BIDMC databases available on PhysioNet were preprocessed and segmented before being utilized for deep learning model training. Pretrained models, ResNet 50 and AlexNet, were fine-tuned and configured to achieve optimal classification results. The main outcome measures for evaluating the performance of the model were F-measure, recall, precision, sensitivity, specificity, and accuracy, obtained from a multi-class confusion matrix.

RESULT

The proposed deep learning model showed overall classification accuracy of 99.2%, average sensitivity of 99.2%, average specificity of 99.6%, average recall, precision and F- measure of 99.2% of test data.

CONCLUSION

The proposed work introduced a robust approach for the classification of arrhythmias in comparison with the most recent state of the art and will reduce the diagnosis time and error that occurs in the visual investigation of ECG signals.

Characteristics of glucose change in diabetes mellitus generalized through continuous wavelet transform processing: A preliminary study

BACKGROUND

The continuous glucose monitoring (CGM) system has become a popular evaluation tool for glucose fluctuation, providing a detailed description of glucose change patterns. We hypothesized that glucose fluctuations may contain specific information on differences in glucose change between type 1 diabetes mellitus (T1DM) and type 2 diabetes mellitus (T2DM), despite similarities in change patterns, because of different etiologies. Unlike Fourier transform, continuous wavelet transform (CWT) is able to simultaneously analyze the time and fre-quency domains of oscillating data.

AIM

To investigate whether CWT can detect glucose fluctuations in T1DM.

METHODS

The 60-d and 296-d glucose fluctuation data of patients with T1DM (n = 5) and T2DM (n = 25) were evaluated respectively. Glucose data obtained every 15 min for 356 d were analyzed. Data were assessed by CWT with Morlet form (n = 7) as the mother wavelet. This methodology was employed to search for limited frequency glucose fluctuation in the daily glucose change. The frequency and enclosed area (0.02625 scalogram value) of 18 emerged signals were compared. The specificity for T1DM was evaluated through multiple regression analysis using items that demonstrated significant differences between them as explanatory variables.

RESULTS

The high frequency at midnight (median: 75 Hz, cycle time: 19 min) and middle frequency at noon (median: 45.5 Hz, cycle time: 32 min) were higher in T1DM vs T2DM (median: 73 and 44 Hz; P = 0.006 and 0.005, respectively). The area of the > 100 Hz zone at midnight to forenoon was more frequent and larger in T1DM vs T2DM. In a day, the lower frequency zone (15-35 Hz) was more frequent and the area was larger in T2DM than in T1DM. The three-dimensional scatter diagrams, which consist of the time of day, frequency, and area of each signal after CWT, revealed that high frequency signals belonging to T1DM at midnight had a loose distribution of wave cycles that were 17-24 min. Multivariate analysis revealed that the high frequency signal at midnight could characterize T1DM (odds ratio: 1.33, 95% confidence interval: 1.08-1.62; P = 0.006).

CONCLUSION

CWT might be a novel tool for differentiate glucose fluctuation of each type of diabetes mellitus using CGM data.

Deep learning classification of systemic sclerosis from multi-site photoplethysmography signals

Introduction: A pilot study assessing a novel approach to identify patients with Systemic Sclerosis (SSc) using deep learning analysis of multi-site photoplethysmography (PPG) waveforms ("DL-PPG"). Methods: PPG recordings having baseline, unilateral arm pressure cuff occlusion and reactive hyperaemia flush phases from 6 body sites were studied in 51 Controls and 20 SSc patients. RGB scalogram images were obtained from the PPG, using the continuous wavelet transform (CWT). 2 different pre-trained convolutional neural networks (CNNs, namely, GoogLeNet and EfficientNetB0) were trained to classify the SSc and Control groups, evaluating their performance using 10-fold stratified cross validation (CV). Their classification performance (i.e., accuracy, sensitivity, and specificity, with 95% confidence intervals) was also compared to traditional machine learning (ML), i.e., Linear Discriminant Analysis (LDA) and K-Nearest Neighbour (KNN). Results: On a participant basis DL-PPG accuracy, sensitivity and specificity for GoogLeNet were 83.1 (72.3-90.9), 75.0 (50.9-91.3) and 86.3 (73.7-94.3)% respectively, and for EfficientNetB0 were 87.3 (77.2-94.0), 80.0 (56.3-94.3) and 90.1 (78.6-96.7)%. The corresponding results for ML classification using LDA were 66.2 (53.9-77.0), 65.0 (40.8-84.6) and 66.7 (52.1-79.2)% respectively, and for KNN were 76.1 (64.5-85.4), 40.0 (19.1-63.9), and 90.2 (78.6-96.7)% respectively. Discussion: This study shows the potential of DL-PPG classification using CNNs to detect SSc. EfficientNetB0 gave an overall improved performance compared to GoogLeNet, with both CNNs performing better than the traditional ML methods tested. Our automatic AI approach, using transfer learning, could offer significant benefits for SSc diagnostics in a variety of clinical settings where low-cost portable and easy-to-use diagnostics can be beneficial.

Myogenic and cortical evoked potentials vary as a function of stimulus pulse geometry delivered in the subthalamic nucleus of Parkinson's disease patients

Introduction

The therapeutic efficacy of deep brain stimulation (DBS) of the subthalamic nucleus (STN) for Parkinson's disease (PD) may be limited for some patients by the presence of stimulation-related side effects. Such effects are most often attributed to electrical current spread beyond the target region. Prior computational modeling studies have suggested that changing the degree of asymmetry of the individual phases of the biphasic, stimulus pulse may allow for more selective activation of neural elements in the target region. To the extent that different neural elements contribute to the therapeutic vs. side-effect inducing effects of DBS, such improved selectivity may provide a new parameter for optimizing DBS to increase the therapeutic window.

Methods

We investigated the effect of six different pulse geometries on cortical and myogenic evoked potentials in eight patients with PD whose leads were temporarily externalized following STN DBS implant surgery. DBS-cortical evoked potentials were quantified using peak to peak measurements and wavelets and myogenic potentials were quantified using RMS.

Results

We found that the slope of the recruitment curves differed significantly as a function of pulse geometry for both the cortical- and myogenic responses. Notably, this effect was observed most frequently when stimulation was delivered using a monopolar, as opposed to a bipolar, configuration.

Discussion

Manipulating pulse geometry results in differential physiological effects at both the cortical and neuromuscular level. Exploiting these differences may help to expand DBS' therapeutic window and support the potential for incorporating pulse geometry as an additional parameter for optimizing therapeutic benefit.

Deep learning based fetal distress detection from time frequency representation of cardiotocogram signal using Morse wavelet: research study

Background

Clinically cardiotocography is a technique which is used to monitor and evaluate the level of fetal distress. Even though, CTG is the most widely used device to monitor determine the fetus health, existence of high false positive result from the visual interpretation has a significant contribution to unnecessary surgical delivery or delayed intervention.

Objective

In the current study an innovative computer aided fetal distress diagnosing model is developed by using time frequency representation of FHR signal using generalized Morse wavelet and the concept of transfer learning of pre-trained ResNet 50 deep neural network model.

Method

From the CTG data that is obtained from the only open access CTU-UHB data base only FHR signal is extracted and preprocessed to remove noises and spikes. After preprocessing the time frequency information of FHR signal is extracted by using generalized Morse wavelet and fed to a pre-trained ResNet 50 model which is fine tuned and configured according to the dataset.

Main outcome measures

Sensitivity (Se), specificity (Sp) and accuracy (Acc) of the model adopted from binary confusion matrix is used as outcome measures.

Result

After successfully training the model, a comprehensive experimentation of testing is conducted for FHR data for which a recording is made during early stage of labor and last stage of labor. Thus, a promising classification result which is accuracy of 98.7%, sensitivity of 97.0% and specificity 100% are achieved for FHR signal of 1st stage of labor. For FHR recorded in last stage of labor, accuracy of 96.1%, sensitivity of 94.1% and specificity 97.7% are achieved.

Conclusion

The developed model can be used as a decision-making aid system for obstetrician and gynecologist.

The impact of pulse timing on cortical and subthalamic nucleus deep brain stimulation evoked potentials

The impact of pulse timing is an important factor in our understanding of how to effectively modulate the basal ganglia thalamocortical (BGTC) circuit. Single pulse low-frequency DBS-evoked potentials generated through electrical stimulation of the subthalamic nucleus (STN) provide insight into circuit activation, but how the long-latency components change as a function of pulse timing is not well-understood. We investigated how timing between stimulation pulses delivered in the STN region influence the neural activity in the STN and cortex. DBS leads implanted in the STN of five patients with Parkinson's disease were temporarily externalized, allowing for the delivery of paired pulses with inter-pulse intervals (IPIs) ranging from 0.2 to 10 ms. Neural activation was measured through local field potential (LFP) recordings from the DBS lead and scalp EEG. DBS-evoked potentials were computed using contacts positioned in dorsolateral STN as determined through co-registered post-operative imaging. We quantified the degree to which distinct IPIs influenced the amplitude of evoked responses across frequencies and time using the wavelet transform and power spectral density curves. The beta frequency content of the DBS evoked responses in the STN and scalp EEG increased as a function of pulse-interval timing. Pulse intervals <1.0 ms apart were associated with minimal to no change in the evoked response. IPIs from 1.5 to 3.0 ms yielded a significant increase in the evoked response, while those >4 ms produced modest, but non-significant growth. Beta frequency activity in the scalp EEG and STN LFP response was maximal when IPIs were between 1.5 and 4.0 ms. These results demonstrate that long-latency components of DBS-evoked responses are pre-dominantly in the beta frequency range and that pulse interval timing impacts the level of BGTC circuit activation.

Wavelet-based CNN for Predicting PAP Adherence Using Overnight Polysomnography Recordings: a Pilot Study

Obstructive sleep apnea (OSA) is a common sleep disorder. Positive airway pressure (PAP) therapy is the first-line treatment, while its effectiveness is significantly limited by incomplete adherence in many patients. This work aims to find a predictive association between data from in-laboratory sleep studies during treatment (PAP titration polysomnogram, or PSG) and PAP adherence. Based on a PAP titration PSG database, we present a pipeline to develop a wavelet-based deep learning model and address two challenges. First, to tackle the problem of extremely long overnight PSG signals, it randomly draws segments and extracts features locally. The global representation for the entire signal is achieved by local feature P-norm pooling. Second, to tackle the problem of limited dataset size, the pre-trained EfficienNet-B7 is used as an unsupervised feature extractor to transfer ImageNet knowledge to PSG signals in the wavelet domain. The trained pipeline achieves 78% balanced accuracy and 83% AUC on the test set using airflow and frontal EEG signals, which, we believe, is a compelling result as a pilot study.

A Motion Artifact Correction Procedure for fNIRS Signals Based on Wavelet Transform and Infrared Thermography Video Tracking

Functional near infrared spectroscopy (fNIRS) is a neuroimaging technique that allows to monitor the functional hemoglobin oscillations related to cortical activity. One of the main issues related to fNIRS applications is the motion artefact removal, since a corrupted physiological signal is not correctly indicative of the underlying biological process. A novel procedure for motion artifact correction for fNIRS signals based on wavelet transform and video tracking developed for infrared thermography (IRT) is presented. In detail, fNIRS and IRT were concurrently recorded and the optodes' movement was estimated employing a video tracking procedure developed for IRT recordings. The wavelet transform of the fNIRS signal and of the optodes' movement, together with their wavelet coherence, were computed. Then, the inverse wavelet transform was evaluated for the fNIRS signal excluding the frequency content corresponding to the optdes' movement and to the coherence in the epochs where they were higher with respect to an established threshold. The method was tested using simulated functional hemodynamic responses added to real resting-state fNIRS recordings corrupted by movement artifacts. The results demonstrated the effectiveness of the procedure in eliminating noise, producing results with higher signal to noise ratio with respect to another validated method.

Clustering Continuous Wavelet Transform Characteristics of Heart Rate Variability through Unsupervised Learning

The analysis and interpretation of physiological signals acquired non-invasively are increasingly important in Smart Health, precision medicine, and medical research. However, this analysis is hampered due to the length, complexity, and inter-subject variation of these signals, and, consequently, dimensionality reduction and clustering offer substantial benefits. Machine learning, used widely in biomedicine, is increasingly being applied to physiological time series. Among the applications of unsupervised learning, clustering is one of the most important. In this paper, an unsupervised autoen-coder architecture, deep convolutional embedded clustering, is presented as a data-driven approach to study time-frequency characteristics of heart rate variability records. An autoen-coder network is trained on continuous wavelet transforms of heart rate variability signals calculated from publicly-available annotated ECG records with a wide variety of conditions. The latent variables learned by the clustering autoencoder are low-dimensional representations of wavelet transform characteristics that can be visualized and further analyzed. The results indicate that the learned clusters correspond to beat morphologies in the electrocardiogram in many cases, but also that the reduced dimensions of the time-frequency features can potentially provide additional insights into cardiac activity and the autonomic nervous system.

Introduction to redundancy rules: the continuous wavelet transform comes of age

Redundancy: it is a word heavy with connotations of lacking usefulness. I often hear that the rationale for not using the continuous wavelet transform (CWT)-even when it appears most appropriate for the problem at hand-is that it is 'redundant'. Sometimes the conversation ends there, as if self-explanatory. However, in the context of the CWT, 'redundant' is not a pejorative term, it simply refers to a less compact form used to represent the information within the signal. The benefit of this new form-the CWT-is that it allows for intricate structural characteristics of the signal information to be made manifest within the transform space, where it can be more amenable to study: resolution over redundancy. Once the signal information is in CWT form, a range of powerful analysis methods can then be employed for its extraction, interpretation and/or manipulation. This theme issue is intended to provide the reader with an overview of the current state of the art of CWT analysis methods from across a wide range of numerate disciplines, including fluid dynamics, structural mechanics, geophysics, medicine, astronomy and finance.This article is part of the theme issue 'Redundancy rules: the continuous wavelet transform comes of age'.