Towards Generating Realistic Wrist Pulse Signals Using Enhanced One Dimensional Wasserstein GAN
<p>Architecture of the Vanilla GAN.</p> "> Figure 2
<p>The pulse wave acquisition system.</p> "> Figure 3
<p>Example of the original and generated signals with different GAN models. In the figure, the real pulse data is shown above, and the false pulse generated by the corresponding GAN model is shown below.</p> "> Figure 4
<p>Comparison of the distribution of frequency spectra real signals and signals generated by different GAN architectures: (<b>a</b>) GAN; (<b>b</b>) DCGAN; (<b>c</b>) WGAN; (<b>d</b>) WGAN-GP.</p> "> Figure 5
<p>The auto-correlation plots of generated wrist pulse signals with different GAN models. In the figure, the real pulse data are shown above, and the false pulse generated by the corresponding GAN model is shown below.</p> "> Figure 6
<p>Examples of time series pair generated from WGAN and WGAN-GP models for associated epochs, respectively: (<b>a</b>) 500 epochs; (<b>b</b>) 1000 epochs; (<b>c</b>) 1500 epochs; (<b>d</b>) 2000 epochs. In every subfigure, the top is the simulated pulse generated by WGAN and the bottom is the simulated pulse generated by WGAN-GP.</p> ">
Abstract
:1. Introduction
2. Methods
2.1. Vanilla GANs
2.2. Proposed Wasserstein GAN with Gradient Penalty Term
2.3. Architecture Details
- Generator: The generator component takes an N-dimensional noise vector (in our case ). Then it is feed to the first block consists of a 1D transposed convolutional layer with kernel size. It has four hidden layers; each block consists of with a 1D transposed convolutional layer with kernel size, then a batch normalization followed by ReLU activations. Note that Tanh activations are employed for the last block. Finally, the generated signal of the output serves as input to the subsequent discriminator component.
- Discriminator: The discriminator component is learned in a supervised manner to minimize the error for classifying false signals from real signal samples. Specifically, it has three hidden fully connected layers. Each block consists of a 1D convolutional layer with kernel size, then a instance normalization followed by Leaky ReLU activations. The last block consists of a 1D transposed convolutional layer with kernel size. Finally, the output layer computes the probabilities of the two classification categories: simulated or real.
3. Materials and Experiment Design
3.1. Data Collected
3.2. Preprocessing
3.3. Model Training
3.4. Performance Evaluations for the Generated Samples
3.4.1. Qualitative Visual Inspection
3.4.2. Quantitative Metric Evaluations
4. Results
4.1. Visual Inspection
4.1.1. Time Samples
4.1.2. Frequency Spectra
4.1.3. Auto-Correlation Plots
4.2. Quantitative Evaluation
4.3. Stability of Pulse Signal Generation
5. Discussion
- We adapted a novel, specifically designed Wasserstein generative adversarial networks with gradient penalty (WGAN-GP), which has been shown to be effective in improving the training stability and convergence [39] in comparison to vanilla GAN. Furthermore, most of the GAN models are currently optimized for 2D images, which is difficult for direct application in time domain signals. In the present study, one-dimensional convolutional neural network (1D-CNN) is adopted as the building block of both generator and discriminator and carefully tailored to leverage its ability to learn local and hierarchical representations from raw data, thereby allowing the adaptation of the GAN framework to pulse data.
- We adopted a set of metrics to quantitatively and visually evaluate the quality of the generated samples comprehensively. For visual comparison, the frequency and time-domain, also including auto-correlation plots, were carefully examined between real and generated pulse signals. In terms of quantitative metrics, three statistical analysis: maximum mean deviation [49], percent root mean square difference [51] and sliced Wasserstein distance [50], were employed.
- We compared the performance of the proposed method with the existing several popular variants of GAN in literature, including vanilla GAN [25], DCGAN [44] and Wasserstein GAN [40]. The various conducted experimental results using real collected wrist pulse dataset demonstrate the effectiveness and advantage of WGAN-GP, showing that the data generation ability of the proposed framework well reflects the distribution of real data.
- A major limitation of time series GANs is the restrictions placed on the length of the sequence specified that the architecture can manage. In our study, the length of pulse signal (2000) is relatively long compared to other physiological signals (for example, ECG/PPG/EEG usually several hundreds lengths in [30,31,33,34,36,38]). In practice, the longer signal usually requires a longer training time while worsening the performance of unstable network and requiring larger epochs. For example, we found that the modal collapse emerges after about 700 epochs for DCGAN. The pulse signal consists of different cycles, and the length of a cycle is roughly 150. Therefore, decomposing pulse into different cycles and training GANs to generate a single cycle signal may be a possible solution.On the other hand, to balance the training cost, the depth of the generator network we used is relatively shallow. As mentioned earlier, we started with a resolution of 100 time samples and doubled it in five steps to reach 2000 samples. Several design choices, such as convolution size, may also impact the performance. In the future, increasing the depth of the network may better capture the pulse characteristics.
- In the present study, WP-GAN was trained independently of the classification task. In practice, the possibility to generate wrist signals with associated labels, especially in the scarcity of abnormal cases, is worthy of further study. That is, whether the computational pulse diagnosis performance can be further improved with these generated pulse signals of different categories. For instance, class-conditioned GAN classifier with WGAN-GP can be employed to use category labels as the auxiliary information and therefore allow for generating labeled data [52]. This is what our following work will focus on, with subsequent collection of wrist pulse signals of different diseases.
- Better network structures: As mentioned, the used wrist pulse signals are relative long (2000 samples), and also the pulse signal is similar to quasi-periodic with each cycle highly correlated. It has been pointed out that relying solely on the convolution operator limits the ability of GAN to capture long-range dependencies across input sequences due to the local receptive field of the convolution operator [53]. To address this problem, self-attention generative adversarial networks (SAGANs) introduce a self-attention mechanism into convolutional GANs, and exhibits a better balance between the ability to model long-range dependencies [53]. In the image [53] and speech domain [54], it has been empirically proved to achieve a superior performance than the baseline. Consequently, introducing a self-attention module into the convolution-based GAN model for pulse data generation may perform better.
- Better cost function: Another important aspect that requires further investigation is to choose appropriate cost function to improve the generation performance, which is orthogonal to network structures, especially considering that evaluating the quality of the generated samples is still an open research topic [46]. Without knowing what the ultimate goal of the learning process is, the selection of optimization functions is to some extent a random process [55]. Furthermore, the loss function utilized in GANs is important for reducing model training error and speeding up network convergence [55]. For example, a reward term with encouraging intraclass/interclass diversity was tailed to achieve better performance in PPG-GAN models [37]. Using different loss functions with various regularizations may result in a better convergence and further improve data generation performance.
- Better model training: In comparison to vanilla GAN, WGAN-GP serves as an effective improvement to enhance the training stability and reduce mode collapse. Moreover, a variety of improved variants are also emerging, with the same aim. For example, different solutions such as normalization and regularization schemes [56,57] have been proposed and have demonstrated their superior performance. Consequently, the proposed WP-GAN model has space for further optimization, which could be achieved by adjusting these regularization and normalization techniques.
6. Conclusions
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Acknowledgments
Conflicts of Interest
References
- Zieman, S.J.; Melenovsky, V.; Kass, D.A. Mechanisms, pathophysiology, and therapy of arterial stiffness. Arterioscler. Thromb. Vasc. Biol. 2005, 25, 932–943. [Google Scholar] [CrossRef] [Green Version]
- Walsh, S.; King, E. Pulse Diagnosis E-Book: A Clinical Guide; Elsevier Health Sciences: Amsterdam, The Netherlands, 2007. [Google Scholar]
- Yao, Y.; Zhou, S.; Alastruey, J.; Hao, L.; Greenwald, S.E.; Zhang, Y.; Xu, L.; Xu, L.; Yao, Y. Estimation of central pulse wave velocity from radial pulse wave analysis. Comput. Methods Programs Biomed. 2022, 219, 106781. [Google Scholar] [CrossRef]
- Bilton, K.; Zaslawski, C. Reliability of manual pulse diagnosis methods in traditional East Asian medicine: A systematic narrative literature review. J. Altern. Complement. Med. 2016, 22, 599–609. [Google Scholar] [CrossRef]
- Guo, C.; Jiang, Z.; He, H.; Liao, Y.; Zhang, D. Wrist pulse signal acquisition and analysis for disease diagnosis: A review. Comput. Biol. Med. 2022, 143, 105312. [Google Scholar] [CrossRef]
- Zhao, C.; Li, G.Z.; Wang, C.; Niu, J. Advances in patient classification for traditional Chinese medicine: A machine learning perspective. Evid.-Based Complement. Altern. Med. eCAM 2015, 2015. [Google Scholar] [CrossRef] [Green Version]
- Jin, J.; Zhang, H.; Geng, X.; Ye, T. The Pulse Waveform Quantification Method Basing on Contour and Derivative. Comput. Methods Programs Biomed. 2022, 220, 106784. [Google Scholar] [CrossRef]
- Jiang, Z.; Guo, C.; Zhang, D. Pressure wrist pulse signal analysis by sparse decomposition using improved Gabor function. Comput. Methods Programs Biomed. 2022, 219, 106766. [Google Scholar] [CrossRef]
- Jiang, Z.; Guo, C.; Zang, J.; Lu, G.; Zhang, D. Features fusion of multichannel wrist pulse signal based on KL-MGDCCA and decision level combination. Biomed. Signal Process. Control. 2020, 57, 101751. [Google Scholar] [CrossRef]
- Liu, L.; Zuo, W.; Zhang, D.; Li, N.; Zhang, H. Combination of heterogeneous features for wrist pulse blood flow signal diagnosis via multiple kernel learning. IEEE Trans. Inf. Technol. Biomed. 2012, 16, 598–606. [Google Scholar] [CrossRef]
- Zhang, D.; Zuo, W.; Wang, P. Generalized feature extraction for wrist pulse analysis: From 1-D time series to 2-D matrix. In Computational Pulse Signal Analysis; Springer: Berlin/Heidelberg, Germany, 2018; pp. 169–189. [Google Scholar]
- Zhang, D.; Zuo, W.; Wang, P. Computational Pulse Signal Analysis; Springer: Berlin/Heidelberg, Germany, 2018. [Google Scholar]
- Lee, C.T.; Wei, L.Y. Spectrum analysis of human pulse. IEEE Trans. Biomed. Eng. 1983, BME-30, 348–352. [Google Scholar] [CrossRef]
- Zhang, D.Y.; Zuo, W.M.; Zhang, D.; Zhang, H.Z.; Li, N.M. Wrist blood flow signal-based computerized pulse diagnosis using spatial and spectrum features. J. Biomed. Sci. Eng. 2010, 3, 361. [Google Scholar] [CrossRef] [Green Version]
- Zhang, D.; Zhang, L.; Zhang, D.; Zheng, Y. Wavelet based analysis of doppler ultrasonic wrist-pulse signals. In Proceedings of the 2008 International Conference on BioMedical Engineering and Informatics, Hainan, China, 28–30 May 2008; Volume 2, pp. 539–543. [Google Scholar]
- Guo, Q.L.; Wang, K.Q.; Zhang, D.Y.; Li, N.M. A wavelet packet based pulse waveform analysis for cholecystitis and nephrotic syndrome diagnosis. In Proceedings of the 2008 International Conference on Wavelet Analysis and Pattern Recognition, Hong Kong, China, 30–31 August 2008; Volume 2, pp. 513–517. [Google Scholar]
- Priyadarshini, S.; Dutt, D.; Rajan, A. Processing of wrist pulse signals using linear and nonlinear techniques—A review. Internet J. Eng. Sci. Comput. 2016, 4, 7477–7482. [Google Scholar]
- Zhang, D.; Zuo, W.; Wang, P. Edit Distance for Pulse Diagnosis. In Computational Pulse Signal Analysis; Springer: Berlin/Heidelberg, Germany, 2018; pp. 217–230. [Google Scholar]
- Li, X.; Fu, Z.; Qian, P.; Wang, L.; Zhang, H.; Zhou, X.; Zhang, W.; Li, F. Computerized wrist pulse signal diagnosis using gradient boosting decision tree. In Proceedings of the 2018 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), Madrid, Spain, 3–6 December 2018; pp. 1941–1947. [Google Scholar]
- LeCun, Y.; Bengio, Y.; Hinton, G. Deep learning. Nature 2015, 521, 436–444. [Google Scholar] [CrossRef] [PubMed]
- Zhang, S.R.; Sun, Q.F. Human pulse recognition based on convolutional neural networks. In Proceedings of the 2016 International Symposium on Computer, Consumer and Control (IS3C), Xi’an, China, 4–6 July 2016; pp. 366–369. [Google Scholar]
- Yan, J.; Chen, S.; Yan, H.; Wang, Y.; Guo, R. Wrist pulse analysis and recognition based on recurrence plot and convolution neural network. Comput. Eng. Appl. 2020, 56, 170–175. [Google Scholar]
- Zhang, Q.; Zhou, J.; Zhang, B. Graph based multichannel feature fusion for wrist pulse diagnosis. IEEE J. Biomed. Health Inform. 2020, 25, 3732–3743. [Google Scholar] [CrossRef]
- Ouyang, C.; Gan, Z.; Zhen, J.; Guan, Y.; Zhu, X.; Zhou, P. Inter-Patient Classification With Encoded Peripheral Pulse Series and Multi-Task Fusion CNN: Application in Type 2 Diabetes. IEEE J. Biomed. Health Inform. 2021, 25, 3130–3140. [Google Scholar] [CrossRef]
- Goodfellow, I.; Pouget-Abadie, J.; Mirza, M.; Xu, B.; Warde-Farley, D.; Ozair, S.; Courville, A.; Bengio, Y. Generative adversarial nets. Adv. Neural Inf. Process. Syst. 2014, 27, 2672–2680. [Google Scholar]
- Wang, Z.; She, Q.; Ward, T.E. Generative adversarial networks in computer vision: A survey and taxonomy. ACM Comput. Surv. (CSUR) 2021, 54, 1–38. [Google Scholar] [CrossRef]
- Wen, Q.; Sun, L.; Yang, F.; Song, X.; Gao, J.; Wang, X.; Xu, H. Time series data augmentation for deep learning: A survey. arXiv 2020, arXiv:2002.12478. [Google Scholar]
- Brophy, E.; Wang, Z.; She, Q.; Ward, T. Generative adversarial networks in time series: A survey and taxonomy. arXiv 2021, arXiv:2107.11098. [Google Scholar]
- Chen, G.; Zhu, Y.; Hong, Z.; Yang, Z. EmotionalGAN: Generating ECG to enhance emotion state classification. In Proceedings of the 2019 International Conference on Artificial Intelligence and Computer Science, Wuhan, China, 12–13 July 2019; pp. 309–313. [Google Scholar]
- Golany, T.; Radinsky, K. Pgans: Personalized generative adversarial networks for ecg synthesis to improve patient-specific deep ecg classification. In Proceedings of the AAAI Conference on Artificial Intelligence, Honolulu, HI, USA, 27 January–1 February 2019; Volume 33, pp. 557–564. [Google Scholar]
- Wulan, N.; Wang, W.; Sun, P.; Wang, K.; Xia, Y.; Zhang, H. Generating electrocardiogram signals by deep learning. Neurocomputing 2020, 404, 122–136. [Google Scholar] [CrossRef]
- Hartmann, K.G.; Schirrmeister, R.T.; Ball, T. EEG-GAN: Generative adversarial networks for electroencephalograhic (EEG) brain signals. arXiv 2018, arXiv:1806.01875. [Google Scholar]
- Fahimi, F.; Dosen, S.; Ang, K.K.; Mrachacz-Kersting, N.; Guan, C. Generative adversarial networks-based data augmentation for brain–computer Interface. IEEE Trans. Neural Netw. Learn. Syst. 2020, 32, 4039–4051. [Google Scholar] [CrossRef]
- Luo, Y.; Lu, B.L. EEG data augmentation for emotion recognition using a conditional Wasserstein GAN. In Proceedings of the 2018 40th annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), Honolulu, HI, USA, 18–21 July 2018; pp. 2535–2538. [Google Scholar]
- Lashgari, E.; Liang, D.; Maoz, U. Data augmentation for deep-learning-based electroencephalography. J. Neurosci. Methods 2020, 346, 108885. [Google Scholar] [CrossRef]
- Delaney, A.M.; Brophy, E.; Ward, T.E. Synthesis of realistic ECG using generative adversarial networks. arXiv 2019, arXiv:1909.09150. [Google Scholar]
- Kiyasseh, D.; Tadesse, G.A.; Thwaites, L.; Zhu, T.; Clifton, D. PlethAugment: GAN-based PPG augmentation for medical diagnosis in low-resource settings. IEEE J. Biomed. Health Inform. 2020, 24, 3226–3235. [Google Scholar] [CrossRef]
- Abdelfattah, S.M.; Abdelrahman, G.M.; Wang, M. Augmenting the size of EEG datasets using generative adversarial networks. In Proceedings of the 2018 International Joint Conference on Neural Networks (IJCNN), Rio de Janeiro, Brazil, 8–13 July 2018; pp. 1–6. [Google Scholar]
- Gulrajani, I.; Ahmed, F.; Arjovsky, M.; Dumoulin, V.; Courville, A.C. Improved training of wasserstein gans. Adv. Neural Inf. Process. Syst. 2017, 30, 5769–5779. [Google Scholar]
- Arjovsky, M.; Chintala, S.; Bottou, L. Wasserstein generative adversarial networks. In Proceedings of the International Conference on Machine Learning, Sydney, Australia, 6–11 August 2017; pp. 214–223. [Google Scholar]
- Karras, T.; Aila, T.; Laine, S.; Lehtinen, J. Progressive growing of gans for improved quality, stability, and variation. arXiv 2017, arXiv:1710.10196. [Google Scholar]
- Wang, D.; Zhang, D.; Lu, G. A robust signal preprocessing framework for wrist pulse analysis. Biomed. Signal Process. Control 2016, 23, 62–75. [Google Scholar] [CrossRef]
- Xu, L.; Zhang, D.; Wang, K. Wavelet-based cascaded adaptive filter for removing baseline drift in pulse waveforms. IEEE Trans. Biomed. Eng. 2005, 52, 1973–1975. [Google Scholar] [CrossRef] [Green Version]
- Radford, A.; Metz, L.; Chintala, S. Unsupervised representation learning with deep convolutional generative adversarial networks. arXiv 2015, arXiv:1511.06434. [Google Scholar]
- Borji, A. Pros and cons of gan evaluation measures. Comput. Vis. Image Underst. 2019, 179, 41–65. [Google Scholar] [CrossRef] [Green Version]
- Borji, A. Pros and cons of GAN evaluation measures: New developments. Comput. Vis. Image Underst. 2022, 215, 103329. [Google Scholar] [CrossRef]
- Ni, H.; Szpruch, L.; Wiese, M.; Liao, S.; Xiao, B. Conditional sig-wasserstein gans for time series generation. arXiv 2020, arXiv:2006.05421. [Google Scholar] [CrossRef]
- Eckerli, F. Generative Adversarial Networks in finance: An overview. arXiv 2021, arXiv:2106.06364. [Google Scholar] [CrossRef]
- Borgwardt, K.M.; Gretton, A.; Rasch, M.J.; Kriegel, H.P.; Schölkopf, B.; Smola, A.J. Integrating structured biological data by kernel maximum mean discrepancy. Bioinformatics 2006, 22, e49–e57. [Google Scholar] [CrossRef] [Green Version]
- Rabin, J.; Peyré, G.; Delon, J.; Bernot, M. Wasserstein barycenter and its application to texture mixing. In Proceedings of the International Conference on Scale Space and Variational Methods in Computer Vision, Ein-Gedi, Israel, 29 May–2 June 2011; Springer: Berlin/Heidelberg, Germany, 2011; pp. 435–446. [Google Scholar]
- Hazra, D.; Byun, Y.C. SynSigGAN: Generative adversarial networks for synthetic biomedical signal generation. Biology 2020, 9, 441. [Google Scholar] [CrossRef]
- Zheng, M.; Li, T.; Zhu, R.; Tang, Y.; Tang, M.; Lin, L.; Ma, Z. Conditional Wasserstein generative adversarial network-gradient penalty-based approach to alleviating imbalanced data classification. Inf. Sci. 2020, 512, 1009–1023. [Google Scholar] [CrossRef]
- Zhang, H.; Goodfellow, I.; Metaxas, D.; Odena, A. Self-attention generative adversarial networks. In Proceedings of the International Conference on Machine Learning, Long Beach, CA, USA, 9–15 June 2019; pp. 7354–7363. [Google Scholar]
- Phan, H.; Le Nguyen, H.; Chén, O.Y.; Koch, P.; Duong, N.Q.; McLoughlin, I.; Mertins, A. Self-attention generative adversarial network for speech enhancement. In Proceedings of the ICASSP 2021—2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Toronto, ON, Canada, 6–12 June 2021; pp. 7103–7107. [Google Scholar]
- Pan, Z.; Yu, W.; Wang, B.; Xie, H.; Sheng, V.S.; Lei, J.; Kwong, S. Loss functions of generative adversarial networks (GANs): Opportunities and challenges. IEEE Trans. Emerg. Top. Comput. Intell. 2020, 4, 500–522. [Google Scholar] [CrossRef]
- Kurach, K.; Lučić, M.; Zhai, X.; Michalski, M.; Gelly, S. A large-scale study on regularization and normalization in GANs. In Proceedings of the International Conference on Machine Learning, Long Beach, CA, USA, 9–15 June 2019; pp. 3581–3590. [Google Scholar]
- Lee, M.; Seok, J. Regularization methods for generative adversarial networks: An overview of recent studies. arXiv 2020, arXiv:2005.09165. [Google Scholar]
Generator | |
---|---|
Layer | Output Shape |
Input layer | 100 × 1 |
Transposed Conv1d | 521 × 125 |
ReLU | 521 × 125 |
Batch Norm | 521 × 125 |
Transposed Conv1d | 256 × 250 |
ReLU | 256 × 250 |
Batch Norm | 256 × 250 |
Transposed Conv1d | 128 × 500 |
ReLU | 128 × 500 |
Batch Norm | 128 × 500 |
Transposed Conv1d | 64 × 1000 |
ReLU | 64 × 1000 |
Batch Norm | 64 × 1000 |
Transposed Conv1d | 1 × 2000 |
Tanh | 1 × 2000 |
Discriminator | |
Layer | Output Shape |
Input layer | 1 × 2000 |
Conv1d | 64 × 1000 |
LeakyReLU | 64 × 1000 |
Instance Norm | 64 × 1000 |
Conv1d | 128 × 500 |
LeakyReLU | 128 × 500 |
Instance Norm | 128 × 500 |
Conv1d | 256 × 250 |
LeakyReLU | 256 × 250 |
Instance Norm | 256 × 250 |
Conv1d | 521 × 125 |
LeakyReLU | 521 × 125 |
Instance Norm | 521 × 125 |
Conv1d | 1 × 1 |
Variable | Male (154) | Female (166) |
---|---|---|
Age (year) | 20.21 ± 0.53 | 20.35 ± 0.46 |
Height (cm) | 175.32 ± 4.25 | 161.35 ± 4.36 |
Weigth (kg) | 65.32 ± 8.35 | 55.07 ± 5.35 |
Method | MMD | SWD | PRD |
---|---|---|---|
GAN | 2.4777 ± 0.1393 | 0.0519 ± 0.0097 | 8.0296 ± 0.6404 |
DCGAN | 5.1677 ± 0.0864 | 0.2545 ± 0.0053 | 11.9029 ± 0.5347 |
WGAN | 0.2815 ± 0.0596 | 0.0145 ± 0.0043 | 6.7092 ± 0.5350 |
WGAN-GP | 0.2325 ± 0.1003 | 0.0112 ± 0.0064 | 5.8748 ± 0.5630 |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Chang, J.; Hu, F.; Xu, H.; Mao, X.; Zhao, Y.; Huang, L. Towards Generating Realistic Wrist Pulse Signals Using Enhanced One Dimensional Wasserstein GAN. Sensors 2023, 23, 1450. https://doi.org/10.3390/s23031450
Chang J, Hu F, Xu H, Mao X, Zhao Y, Huang L. Towards Generating Realistic Wrist Pulse Signals Using Enhanced One Dimensional Wasserstein GAN. Sensors. 2023; 23(3):1450. https://doi.org/10.3390/s23031450
Chicago/Turabian StyleChang, Jiaxing, Fei Hu, Huaxing Xu, Xiaobo Mao, Yuping Zhao, and Luqi Huang. 2023. "Towards Generating Realistic Wrist Pulse Signals Using Enhanced One Dimensional Wasserstein GAN" Sensors 23, no. 3: 1450. https://doi.org/10.3390/s23031450
APA StyleChang, J., Hu, F., Xu, H., Mao, X., Zhao, Y., & Huang, L. (2023). Towards Generating Realistic Wrist Pulse Signals Using Enhanced One Dimensional Wasserstein GAN. Sensors, 23(3), 1450. https://doi.org/10.3390/s23031450