An official website of the United States government
Abstract This paper introduces a study on the classification of aortic stenosis (AS) based on cardio-mechanical signals collected using non-invasive wearable inertial sensors. Measurements were taken from 21 AS patients and 13 non-AS subjects. A feature analysis framework utilizing Elastic Net was implemented to reduce the features generated by continuous wavelet transform (CWT). Performance comparisons were conducted among several machine learning (ML) algorithms, including decision tree, random forest, multi-layer perceptron neural network, and extreme gradient boosting. In addition, a two-dimensional convolutional neural network (2D-CNN) was developed using the CWT coefficients as images. The 2D-CNN was made with a custom-built architecture and a CNN based on Mobile Net via transfer learning. After the reduction of features by 95.47%, the results obtained report 0.87 on accuracy by decision tree, 0.96 by random forest, 0.91 by simple neural network, and 0.95 by XGBoost. Via the 2D-CNN framework, the transfer learning of Mobile Net shows an accuracy of 0.91, while the custom-constructed classifier reveals an accuracy of 0.89. Our results validate the effectiveness of the feature selection and classification framework. They also show a promising potential for the implementation of deep learning tools on the classification of AS.
more »
« less
Classification of Aortic Stenosis Using Time-Frequency Features from Chest Cardio-mechanical Signals
Objectives: This paper introduces a novel method for the detection and classification of aortic stenosis (AS) using the time-frequency features of chest cardio-mechanical signals collected from wearable sensors, namely seismo-cardiogram (SCG) and gyro-cardiogram (GCG) signals. Such a method could potentially monitor high-risk patients out of the clinic. Methods: Experimental measurements were collected from twenty patients with AS and twenty healthy subjects. Firstly, a digital signal processing framework is proposed to extract time-frequency features. The features are then selected via the analysis of variance test. Different combinations of features are evaluated using the decision tree, random forest, and artificial neural network methods. Two classification tasks are conducted. The first task is a binary classification between normal subjects and AS patients. The second task is a multi-class classification of AS patients with co-existing valvular heart diseases. Results: In the binary classification task, the average accuracies achieved are 96.25% from decision tree, 97.43% from random forest, and 95.56% from neural network. The best performance is from combined SCG and GCG features with random forest classifier. In the multi-class classification, the best performance is 92.99% using the random forest classifier and SCG features. Conclusion: The results suggest that the solution could be a feasible method for classifying aortic stenosis, both in the binary and multi-class tasks. It also indicates that most of the important time-frequency features are below 11 Hz. Significance: The proposed method shows great potential to provide continuous monitoring of valvular heart diseases to prevent patients from sudden critical cardiac situations.
more »
« less
- Award ID(s):
- 1855394
- PAR ID:
- 10135605
- Date Published:
- Journal Name:
- IEEE Transactions on Biomedical Engineering
- ISSN:
- 0018-9294
- Page Range / eLocation ID:
- 1 to 12
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
This paper describes an open-access database for seismo-cardiogram (SCG) and gyro-cardiogram (GCG) signals. The archive comprises SCG and GCG recordings sourced from and processed at multiple sites worldwide, including Columbia University Medical Center and Stevens Institute of Technology in the United States, as well as Southeast University, Nanjing Medical University, and the first affiliated hospital of Nanjing Medical University in China. It includes electrocardiogram (ECG), SCG, and GCG recordings collected from 100 patients with various conditions of valvular heart diseases such as aortic and mitral stenosis. The recordings were collected from clinical environments with the same types of wearable sensor patch. Besides the raw recordings of ECG, SCG, and GCG signals, a set of hand-corrected fiducial point annotations is provided by manually checking the results of the annotated algorithm. The database also includes relevant echocardiogram parameters associated with each subject such as ejection fraction, valve area, and mean gradient pressure.more » « less
-
In this empirical study, a framework was developed for binary and multi-class classification of Twitter data. We first introduce a manually built gold standard dataset of 4000 tweets related to the environmental health hazards in Barbados for the period 2014 - 2018. Then, the binary classification was used to categorize each tweet as relevant or irrelevant. Next, the multiclass classification was then used to further classify relevant tweets into four types of community engagement: reporting information, expressing negative engagement, expressing positive engagement, and asking for information. Results indicate that (combination of TF-IDF, psychometric, linguistic, sentiment and Twitter-specific features) using a Random Forest algorithm is the best feature for detecting and predicting binary classification with (87% F1 score). For multi-class classification, TF-IDF using Decision Tree algorithm was the best with (74% F1 score).more » « less
-
Abstract Cardiovascular diseases, the leading cause of global mortality, demand refined diagnostic methods. Seismocardiography (SCG), a noninvasive method of measuring cardiovascular-induced vibrations on the chest surface, offers promise in assessing cardiac function. The cardiac wall movements are transmitted to the organs around the heart and eventually damped onto the chest surface, where they manifest as visible vibrations. These chest surface vibrations can be measured using an accelerometer via SCG. Although SCG signals are widely used in literature, further investigations are needed to understand the genesis of their patterns under different pathophysiological conditions. The goal of this study is to improve our understanding of the origin of SCG signals by simulating the transmission of cardiac motion reaching the chest surface using finite element method, and linking back the patterns of the simulated SCG signals to specific cardiac events. The computational domain, extracted from 4D computed tomography (CT) images of a healthy subject, comprised the lungs, ribcage, and chest muscles and fat. Using the Lukas-Kanade algorithm, the cardiac wall motion was extracted from the 4D CT scan images and was used as a displacement boundary condition. The elastic material properties were assigned to the lungs, muscles, fat, and rib cage. The dorsoventral SCG component from the finite element modeling was compared with two actual SCG signals obtained from the literature. The left ventricular volume was also calculated from the CT scans and was used to interpret the SCG waveforms. Important cardiac phases were labeled on the SCG signal extracted from the computationally modeled acceleration map near the xiphoid. This type of analysis can provide insights into various cardiac parameters and SCG patterns corresponding to the mitral valve closing, mitral valve opening, aortic valve opening, and aortic valve closure. These findings suggested the effectiveness of this modeling approach in understanding the underlying sources of the SCG waveforms.more » « less
-
Automated interpretation of ultrasound imaging of the heart (echocardiograms) could improve the detection and treatment of aortic stenosis (AS), a deadly heart disease. However, existing deep learning pipelines for assessing AS from echocardiograms have two key limitations. First, most methods rely on limited 2D cineloops, thereby ignoring widely available Spectral Doppler imaging that contains important complementary information about pressure gradients and blood flow abnormalities associated with AS. Second, obtaining labeled data is difficult. There are often far more unlabeled echocardiogram recordings available, but these remain underutilized by existing methods. To overcome these limitations, we introduce Semi-supervised Multimodal Multiple-Instance Learning (SMMIL), a new deep learning framework for automatic interpretation for structural heart diseases like AS. During training, SMMIL can combine a smaller labeled set and an abundant unlabeled set of both 2D and Doppler modalities to improve its classifier. When deployed, SMMIL can combine information from all available images to produce an accurate study-level diagnosis of this life-threatening condition. Experiments demonstrate that SMMIL outperforms recent alternatives, including two medical foundation models.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1109/TBME.2019.2942741
