Signal Quality Classification Research Articles

Signal Quality Classification of Impedance Plethysmogram and Ballistocardiogram for Pulse Transit Time Measurement

Signal Quality Classification of Impedance Plethysmogram and Ballistocardiogram for Pulse Transit Time Measurement

Unsupervised Learning-Based Non-Invasive Fetal ECG Muti-Level Signal Quality Assessment.

To monitor fetal health and growth, fetal heart rate is a critical indicator. The non-invasive fetal electrocardiogram is a widely employed measurement for fetal heart rate estimation, which is extracted from the electrodes placed on the surface of the maternal abdomen. The qualities of the fetal ECG recordings, however, are frequently affected by the noises from various interference sources. In general, the fetal heart rate estimates are unreliable when low-quality fetal ECG signals are used for fetal heart rate estimation, which makes accurate fetal heart rate estimation a challenging task. So, the signal quality assessment for the fetal ECG records is an essential step before fetal heart rate estimation. In other words, some low-quality fetal ECG signal segments are supposed to be detected and removed by utilizing signal quality assessment, so as to improve the accuracy of fetal heart rate estimation. A few supervised learning-based fetal ECG signal quality assessment approaches have been introduced and shown to accurately classify high- and low-quality fetal ECG signal segments, but large fetal ECG datasets with quality annotation are required in these methods. Yet, the labeled fetal ECG datasets are limited. Proposed methods: An unsupervised learning-based multi-level fetal ECG signal quality assessment approach is proposed in this paper for identifying three levels of fetal ECG signal quality. We extracted some features associated with signal quality, including entropy-based features, statistical features, and ECG signal quality indices. Additionally, an autoencoder-based feature is calculated, which is related to the reconstruction error of the spectrograms generated from fetal ECG signal segments. The high-, medium-, and low-quality fetal ECG signal segments are classified by inputting these features into a self-organizing map. The experimental results showed that our proposal achieved a weighted average F1-score of 90% in three-level fetal ECG signal quality classification. Moreover, with the acceptable removal of detected low-quality signal segments, the errors of fetal heart rate estimation were reduced to a certain extent.

Signal Quality Index for the fetal heart rates: Development and improvements for fetal monitoring

Fetal heart rate (FHR) monitoring is the most widely used tool in clinics to assess fetal health. However, FHR with low-quality signals may somehow exaggerate the risk of the fetus suffering from acidemia, thus contributing to an increase in cesarean section rates. If there is an algorithm capable to identify and reject low-quality signals and then calculate heart rate parameters only on high-quality signals, inappropriate obstetric interventions may be greatly reduced. Currently, only the Signal Quality Index (SQI) is available for adults. When we applied SQI to fetal data, its performance drops dramatically. To fill the gap in the fetal signal quality index, we developed an easy-to-use Fetal heart rate Signal Quality Index (FSQI) in this paper. Firstly, we used a human-in-the-loop strategy to reduce the amount of data annotation effort. Secondly, we developed the FSQI algorithm oriented towards the fetus to identify low-quality signal segments. Further, we proposed a post-processing technique based on fine-grained recognition to identify low-quality signal segments more accurately. Finally, we applied the FSQI algorithm in combination with the existing FHR deceleration event detection algorithm to test the effectiveness of the algorithm. The results showed that the FSQI algorithm we developed achieved an overall accuracy of 99.8842% in signal quality classification, while also eliminating 94.92% of the incorrectly detected deceleration events of current state-of-the-art methods.

Identification of Characteristic Points in Multivariate Physiological Signals by Sensor Fusion and Multi-Task Deep Networks

Identification of characteristic points in physiological signals, such as the peak of the R wave in the electrocardiogram and the peak of the systolic wave of the photopletismogram, is a fundamental step for the quantification of clinical parameters, such as the pulse transit time. In this work, we presented a novel neural architecture, called eMTUnet, to automate point identification in multivariate signals acquired with a chest-worn device. The eMTUnet consists of a single deep network capable of performing three tasks simultaneously: (i) localization in time of characteristic points (labeling task), (ii) evaluation of the quality of signals (classification task); (iii) estimation of the reliability of classification (reliability task). Preliminary results in overnight monitoring showcased the ability to detect characteristic points in the four signals with a recall index of about 1.00, 0.90, 0.90, and 0.80, respectively. The accuracy of the signal quality classification was about 0.90, on average over four different classes. The average confidence of the correctly classified signals, against the misclassifications, was 0.93 vs. 0.52, proving the worthiness of the confidence index, which may better qualify the point identification. From the achieved outcomes, we point out that high-quality segmentation and classification are both ensured, which brings the use of a multi-modal framework, composed of wearable sensors and artificial intelligence, incrementally closer to clinical translation.

A medical radar system for non-contact vital sign monitoring and clinical performance evaluation in hospitalized older patients

Medical radar plays an important role in non-contact vital sign monitoring. However, radar is sensitive and susceptible to noise that deteriorates signal quality, making its application unstable in medical practice settings. Previous studies have focused on detecting vital signs from radar output, but mainly used high-quality signals in laboratory settings. Therefore, signal quality evaluation to maintain good-quality signals for processing is crucial for radar applications. This paper presents a vital sign monitoring system using medical radar that automatically evaluates signal quality to eliminate unsatisfactory signals using the support-vector machine and to extract heartbeat and respiration signals using the singular value decomposition method. To evaluate the system performance, we tested it on 10 healthy subjects and compared the obtained vital signs with those measured using a contact-type ECG device and a respiration belt. Two evaluation steps were conducted. First, we measured 10 subjects for evaluating the signal quality in a laboratory setting. The results showed that the accuracy for each subject archive was 93.2%–100%. Second, we extracted vital signs including respiration rate (RR) and the heart rate (HR) following signal quality classification. The results showed RR and HR deviation from 0 to 1 bpm, and 0 to 4 bpm, respectively. Moreover, the system was applied in a real medical practice setting for continuously monitoring the bedside of hospitalized older patients at a hospital.

Explainability Metrics of Deep Convolutional Networks for Photoplethysmography Quality Assessment.

Photoplethysmography (PPG) is a noninvasive way to monitor various aspects of the circulatory system, and is becoming more and more widespread in biomedical processing. Recently, deep learning methods for analyzing PPG have also become prevalent, achieving state of the art results on heart rate estimation, atrial fibrillation detection, and motion artifact identification. Consequently, a need for interpretable deep learning has arisen within the field of biomedical signal processing. In this paper, we pioneer novel explanatory metrics which leverage domain-expert knowledge to validate a deep learning model. We visualize model attention over a whole testset using saliency methods and compare it to human expert annotations. Congruence, our first metric, measures the proportion of model attention within expert-annotated regions. Our second metric, Annotation Classification, measures how much of the expert annotations our deep learning model pays attention to. Finally, we apply our metrics to compare between a signal based model and an image based model for PPG signal quality classification. Both models are deep convolutional networks based on the ResNet architectures. We show that our signal-based one dimensional model acts in a more explainable manner than our image based model; on average 50.78% of the one dimensional model’s attention are within expert annotations, whereas 36.03% of the two dimensional model’s attention are within expert annotations. Similarly, when thresholding the one dimensional model attention, one can more accurately predict if each pixel of the PPG is annotated as artifactual by an expert. Through this testcase, we demonstrate how our metrics can provide a quantitative and dataset-wide analysis of how explainable the model is.

Heart Sound Signal Quality Assessment Based on Multi-Domain Features

Heart sound is one of the most important physiological signals of our body, including a large number of physiological and pathological information that can reflect the cardiovascular status. This study aims to develop a heart sound signal quality assessment method. In view of the 3 common noises (deep breath, speaking and cough) in clinical data collection, a total of 72 features were extracted from 6 domains, i.e., time, frequency, entropy, energy, high-order statistics and cyclostationarity. Then information gain, which was used as feature selection method, as well as statistical analysis were employed for dimension reduction. A SVM with radial basis kernel function was trained for final signal quality classification. The best effect was obtained on distinguishing resting from cough and the result showed that the classification performance was significantly improved after feature selection. In contrast, statistical analysis had little effect on the improvement of classification results. The best accuracy in distinguishing between resting and deep breath, resting and speaking, resting and cough is 87.73%, 95.00%, 98.64%, respectively. These results indicate that the proposed method is effective for identifying different noise states, namely cough, speaking and deep breath.

Heart Signal analysis on Multi-Domain Features Extraction by SVM Classifier in Smart Monitoring System

According to world health organization (WHO) the heart strokes and cardiovascular diseases death rate is increases every year. Heart signal is one of the most predominant physiological signals of our body, including a large number of physiological and pathological information that can reflect the cardiovascular status. This work aims to develop a heart signal quality assessment method by three common case studies of deep breath, speaking and climbing up &down. In data collection, a total features were extracted from domain statistics. Here statistical analysis is employed for reducing dimension of a particular features. For classification of electrocardiogram (ECG) signals cardiac arrhythmias using deep learning model is used by Cubic Wavelet Transform. These parameters are used as input to these classifier with types of ECG signals. A SVM with radial basis kernel function was trained for final signal quality classification. The best effect was obtained on distinguishing resting from climbing up& down and the result showed that the classification performance was significantly improved after feature selection. These results indicate that the proposed method is effective for identifying different cases. The hardware testing was implemented and tested for the same case studies.

Nontraditional Electrocardiogram and Algorithms for Inconspicuous In-Home Monitoring: Comparative Study.

BackgroundWearable and connected in-home medical devices are typically utilized in uncontrolled environments and often measure physiologic signals at suboptimal locations. Motion artifacts and reduced signal-to-noise ratio, compared with clinical grade equipment, results in a highly variable signal quality that can change significantly from moment to moment. The use of signal quality classification algorithms and robust feature delineation algorithms designed to achieve high accuracy on poor quality physiologic signals can prove beneficial in addressing concerns associated with measurement accuracy, confidence, and clinical validity.ObjectiveThe objective of this study was to demonstrate the successful extraction of clinical grade measures using a custom signal quality classification algorithm for the rejection of poor-quality regions and a robust QRS delineation algorithm from a nonstandard electrocardiogram (ECG) integrated into a toilet seat; a device plagued by many of the same challenges as wearable technologies and other Internet of Things–based medical devices.MethodsThe present algorithms were validated using a study of 25 normative subjects and 29 heart failure (HF) subjects. Measurements captured from a toilet seat-based buttocks electrocardiogram were compared with a simultaneously captured 12-lead clinical grade ECG. The ECG lead with the highest morphological correlation to buttocks electrocardiogram was used to determine the accuracy of the heart rate (HR), heart rate variability (HRV), which used the standard deviation of the normal-to-normal (SDNN) intervals between sinus beats, QRS duration, and the corrected QT interval (QTc). These algorithms were benchmarked using the MIT-BIH Arrhythmia Database (MITDB) and European ST-T Database (EDB), which are standardized databases commonly used to test QRS detection algorithms.ResultsClinical grade accuracy was achieved for all buttocks electrocardiogram measures compared with standard Lead II. For the normative cohort, the mean was −0.0 (SD 0.3) bpm (N=141 recordings) for HR accuracy and −1.0 (SD 3.4) ms for HRV (N=135). The QRS duration and the QTc interval had an accuracy of −0.5 (SD 6.6) ms (N=85) and 14.5 (SD 11.1) ms (N=85), respectively. In the HF cohort, the accuracy for HR, HRV, QRS duration, and QTc interval was 0.0 (SD 0.3) bpm (N=109), −6.6 (SD 13.2) ms (N=99), 2.9 (SD 11.5) ms (N=59), and 11.2 (SD 19.1) ms (N=58), respectively. When tested on MITDB and EDB, the algorithms presented herein had an overall sensitivity and positive predictive value of over 99.82% (N=900,059 total beats), which is comparable to best in-class algorithms tuned specifically for use with these databases.ConclusionsThe present algorithmic approach to data analysis of noisy physiologic data was successfully demonstrated using a toilet seat-based ECG remote monitoring system. This approach to the analysis of physiologic data captured from wearable and connected devices has future potential to enable new types of monitoring devices, providing new insights through daily, inconspicuous in-home monitoring.

A machine learning approach to multi-level ECG signal quality classification

Current electrocardiogram (ECG) signal quality assessment studies have aimed to provide a two-level classification: clean or noisy. However, clinical usage demands more specific noise level classification for varying applications. This work outlines a five-level ECG signal quality classification algorithm. A total of 13 signal quality metrics were derived from segments of ECG waveforms, which were labeled by experts. A support vector machine (SVM) was trained to perform the classification and tested on a simulated dataset and was validated using data from the MIT-BIH arrhythmia database (MITDB). The simulated training and test datasets were created by selecting clean segments of the ECG in the 2011 PhysioNet/Computing in Cardiology Challenge database, and adding three types of real ECG noise at different signal-to-noise ratio (SNR) levels from the MIT-BIH Noise Stress Test Database (NSTDB). The MITDB was re-annotated for five levels of signal quality. Different combinations of the 13 metrics were trained and tested on the simulated datasets and the best combination that produced the highest classification accuracy was selected and validated on the MITDB. Performance was assessed using classification accuracy (Ac), and a single class overlap accuracy (OAc), which assumes that an individual type classified into an adjacent class is acceptable. An Ac of 80.26% and an OAc of 98.60% on the test set were obtained by selecting 10 metrics while 57.26% (Ac) and 94.23% (OAc) were the numbers for the unseen MITDB validation data without retraining. By performing the fivefold cross validation, an Ac of 88.07±0.32% and OAc of 99.34±0.07% were gained on the validation fold of MITDB.