Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
41 tokens/sec
GPT-4o
60 tokens/sec
Gemini 2.5 Pro Pro
44 tokens/sec
o3 Pro
8 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

TransfoRhythm: A Transformer Architecture Conductive to Blood Pressure Estimation via Solo PPG Signal Capturing (2404.15352v1)

Published 15 Apr 2024 in eess.SP and cs.LG
TransfoRhythm: A Transformer Architecture Conductive to Blood Pressure Estimation via Solo PPG Signal Capturing

Abstract: Recent statistics indicate that approximately 1.3 billion individuals worldwide suffer from hypertension, a leading cause of premature death globally. Blood pressure (BP) serves as a critical health indicator for accurate and timely diagnosis and/or treatment of hypertension. Driven by recent advancements in AI and Deep Neural Networks (DNNs), there has been a surge of interest in developing data-driven and cuff-less BP estimation solutions. In this context, current literature predominantly focuses on coupling Electrocardiography (ECG) and Photoplethysmography (PPG) sensors, though this approach is constrained by reliance on multiple sensor types. An alternative, utilizing standalone PPG signals, presents challenges due to the absence of auxiliary sensors (ECG), requiring the use of morphological features while addressing motion artifacts and high-frequency noise. To address these issues, the paper introduces the TransfoRhythm framework, a Transformer-based DNN architecture built upon the recently released physiological database, MIMIC-IV. Leveraging Multi-Head Attention (MHA) mechanism, TransfoRhythm identifies dependencies and similarities across data segments, forming a robust framework for cuff-less BP estimation solely using PPG signals. To our knowledge, this paper represents the first study to apply the MIMIC IV dataset for cuff-less BP estimation, and TransfoRhythm is the first MHA-based model trained via MIMIC IV for BP prediction. Performance evaluation through comprehensive experiments demonstrates TransfoRhythm's superiority over its state-of-the-art counterparts. Specifically, TransfoRhythm achieves highly accurate results with Root Mean Square Error (RMSE) of [1.84, 1.42] and Mean Absolute Error (MAE) of [1.50, 1.17] for systolic and diastolic blood pressures, respectively.

PDF HTML Abstract

Exploring Cuffless Blood Pressure Estimation Using the TransfoRhythm Framework

Overview of Research

This paper introduces a novel attention-based Deep Neural Network (DNN) architecture termed the "TransfoRhythm" to estimate blood pressure (BP) from photoplethysmography (PPG) signals. The research employs the recently released MIMIC-IV dataset, marking a significant development in the application of attention mechanisms for cuffless BP monitoring. By leveraging Multi-Head Attention (MHA), the proposed model is designed to capture the rich temporal dynamics and variations in PPG signals, thus enhancing the prediction accuracy of systolic and diastolic blood pressure (SBP and DBP).

Methodology

The TransfoRhythm framework utilizes an advanced neural architecture strategy combining MHA with position embeddings and a time compressor for effective feature translation into BP estimates. The model processes input data through several bespoke preprocessing and feature extraction stages, ensuring that inputs are robust and reflective of significant physiological signals. The architecture of the model supports parallel processing of inputs, crucial for handling the large volumes of data typical in healthcare applications.

Data Handling

The framework employs the MIMIC-IV dataset comprising 198 individual records with PPG, ECG, and invasive arterial blood pressure (ABP) signals. The preprocessing pipeline includes steps for noise reduction, signal smoothing via Butterworth and Moving Average Filters, and artifact removal, ensuring high-quality inputs for the DNN model.

Feature Engineering

Twelve features are extracted from the PPG waveforms focusing on temporal characteristics and morphological attributes which are vital for capturing the waveform dynamics associated with BP changes. These features include cycle duration, peak-to-notch durations, and integrative metrics over defined parts of the PPG waveform.

Results and Performance Evaluation

The TransfoRhythm framework demonstrated high accuracy in BP estimation with mean R-squared values above 0.991 for both SBP and DBP, and low error metrics (MAE and RMSE) across both measurement categories. Performance metrics closely adhere to both AAMI and BHS standards, underscoring the model's utility in clinical settings.

Benchmark Comparisons

Comparative analyses against other deep learning models using the same dataset highlight the superior predictive capability of TransfoRhythm, with significant improvements noted in MAE and RMSE values compared to established benchmarks like ResNet1D, U-Net, and hybrid CNN-RNN models.

Practical Implications and Theoretical Contributions

The TransfoRhythm model's strong performance suggests significant practical applications, particularly in continuous, non-invasive blood pressure monitoring, potentially enhancing patient care in settings where traditional cuff devices are impractical. Theoretically, the work extends knowledge on the efficacy of attention mechanisms in medical signal processing, particularly in modeling time-series data such as PPG signals where traditional methods might falter.

Future Directions

While the current paper prominently showcases the potential of MHA in healthcare, future research might explore hybrid models integrating RNN and CNN features, use larger datasets for training and validation, or extend applications to other physiological signal estimations. Further refining the model to address diverse patient demographics or different health conditions could enhance its robustness and applicability in global health contexts.

Conclusion

In conclusion, the TransfoRhythm framework marks a robust advancement in the use of deep learning technologies for health monitoring, specifically leveraging MHA within a blood pressure estimation context from PPG signals. Its success opens avenues for more personalized, efficient, and non-invasive health monitoring solutions, crucial for modern healthcare infrastructures.

File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown Bookmark Chat Bubble Oval Streamline Icon: https://streamlinehq.com Chat (Pro)
Definition Search Book Streamline Icon: https://streamlinehq.com
References (61)
  1. Zabihi, S. et al. Bp-net: Cuff-less and non-invasive blood pressure estimation via a generic deep convolutional architecture. \JournalTitleBiomedical Signal Processing and Control 78, 103850 (2022).
  2. Heidenreich, P. A. et al. Forecasting the future of cardiovascular disease in the united states: a policy statement from the american heart association. \JournalTitleCirculation 123, 933–944 (2011).
  3. Lewis, P. S. & British and Irish Hypertension Society’s Blood Pressure Measurement Working Party. Oscillometric measurement of blood pressure: a simplified explanation. a technical note on behalf of the british and irish hypertension society. \JournalTitleJournal of human hypertension 33, 349–351 (2019).
  4. Cuffless blood pressure measurement. \JournalTitleAnnual Review of Biomedical Engineering 24, 203–230 (2022).
  5. Kario, K. et al. Effect of esaxerenone on nocturnal blood pressure and natriuretic peptide in different dipping phenotypes. \JournalTitleHypertension Research 45, 97–105 (2022).
  6. Hu, J.-R. et al. Validating cuffless continuous blood pressure monitoring devices. \JournalTitleCardiovascular Digital Health Journal (2023).
  7. Shao, J. et al. An optimization study of estimating blood pressure models based on pulse arrival time for continuous monitoring. \JournalTitleJournal of Healthcare Engineering 2020 (2020).
  8. Yousefian, P. et al. Pulse transit time-pulse wave analysis fusion based on wearable wrist ballistocardiogram for cuff-less blood pressure trend tracking. \JournalTitleIEEE Access 8, 138077–138087 (2020).
  9. Nonlinear cuffless blood pressure estimation of healthy subjects using pulse transit time and arrival time. \JournalTitleIEEE Transactions on Instrumentation and Measurement 66, 3299–3308 (2017).
  10. Pielmus, A.-G. et al. Surrogate based continuous noninvasive blood pressure measurement. \JournalTitleBiomedical Engineering/Biomedizinische Technik 66, 231–245 (2021).
  11. Pulse transit time as an indicator of arterial blood pressure. \JournalTitlepsychophysiology 18, 71–74 (1981).
  12. Cuffless blood pressure monitoring using ptt and pwv methods. In 2017 International Conference on Recent Innovations in Signal processing and Embedded Systems (RISE), 395–401 (IEEE, 2017).
  13. A hybrid neural network for continuous and non-invasive estimation of blood pressure from raw electrocardiogram and photoplethysmogram waveforms. \JournalTitleComputer Methods and Programs in Biomedicine 207, 106191 (2021).
  14. Mlp-bp: A novel framework for cuffless blood pressure measurement with ppg and ecg signals based on mlp-mixer neural networks. \JournalTitleBiomedical Signal Processing and Control 73, 103404 (2022).
  15. Tolstikhin, I. O. et al. Mlp-mixer: An all-mlp architecture for vision. \JournalTitleAdvances in neural information processing systems 34, 24261–24272 (2021).
  16. Goldberger, A. L. et al. Physiobank, physiotoolkit, and physionet: components of a new research resource for complex physiologic signals. \JournalTitlecirculation 101, e215–e220 (2000).
  17. Le, T. et al. Continuous non-invasive blood pressure monitoring: a methodological review on measurement techniques. \JournalTitleIEEE Access 8, 212478–212498 (2020).
  18. Welykholowa, K. et al. Multimodal photoplethysmography-based approaches for improved detection of hypertension. \JournalTitleJournal of Clinical Medicine 9, 1203 (2020).
  19. Combination of ecg and ppg signals for smart healthcare systems: Techniques, applications, and challenges. In 2021 Tenth International Conference on Intelligent Computing and Information Systems (ICICIS), 448–455 (IEEE, 2021).
  20. Chao, P. C.-P. et al. The machine learnings leading the cuffless ppg blood pressure sensors into the next stage. \JournalTitleIEEE Sensors Journal 21, 12498–12510 (2021).
  21. Deep generative model with domain adversarial training for predicting arterial blood pressure waveform from photoplethysmogram signal. \JournalTitleBiomedical Signal Processing and Control 70, 102972 (2021).
  22. Mahmud, S. et al. Nabnet: a nested attention-guided biconvlstm network for a robust prediction of blood pressure components from reconstructed arterial blood pressure waveforms using ppg and ecg signals. \JournalTitleBiomedical Signal Processing and Control 79, 104247 (2023).
  23. Improved ppg-based estimation of the blood pressure using latent space features. \JournalTitleSignal, Image and Video Processing 13, 1141–1147 (2019).
  24. Cuffless and continuous blood pressure estimation from ppg signals using recurrent neural networks. In 2020 42nd annual international conference of the IEEE engineering in medicine & biology society (EMBC), 4269–4272 (IEEE, 2020).
  25. Generalized deep neural network model for cuffless blood pressure estimation with photoplethysmogram signal only. \JournalTitleSensors 20, 5668 (2020).
  26. Cuffless blood pressure estimation from ppg signals and its derivatives using deep learning models. \JournalTitleBiomedical Signal Processing and Control 70, 102984 (2021).
  27. Higher order derivative-based integrated model for cuff-less blood pressure estimation and stratification using ppg signals. \JournalTitleIEEE Sensors Journal 22, 22030–22039 (2022).
  28. Cuffless blood pressure estimation based on photoplethysmography signal and its second derivative. \JournalTitleInternational Journal of Computer Theory and Engineering 9, 202 (2017).
  29. Hybrid cnn-svr blood pressure estimation model using ecg and ppg signals. \JournalTitleSensors 23, 1259 (2023).
  30. A comparison of deep learning techniques for arterial blood pressure prediction. \JournalTitleCognitive Computation 14, 1689–1710 (2022).
  31. Deep learning models for cuffless blood pressure monitoring from ppg signals using attention mechanism. \JournalTitleBiomedical Signal Processing and Control 65, 102301 (2021).
  32. Cuff-less blood pressure estimation from photoplethysmography via visibility graph and transfer learning. \JournalTitleIEEE Journal of Biomedical and Health Informatics 26, 2075–2085 (2021).
  33. Johnson, A. et al. Mimic-iv. \JournalTitlePhysioNet. Available online at: https://physionet.org/content/mimic-iv-ed/2.0/ (accessed August, 2023) (2020).
  34. Ma, C. et al. Stp: Self-supervised transfer learning based on transformer for noninvasive blood pressure estimation using photoplethysmography. \JournalTitleExpert Systems with Applications 123809 (2024).
  35. Tahir, M. A. et al. Cuff-less arterial blood pressure waveform synthesis from single-site ppg using transformer & frequency-domain learning. \JournalTitlearXiv preprint arXiv:2401.05452 (2024).
  36. PPG-DaLiA. UCI Machine Learning Repository (2019). DOI: https://doi.org/10.24432/C53890.
  37. Arterialnet: Arterial blood pressure reconstruction. In 2023 IEEE EMBS International Conference on Biomedical and Health Informatics (BHI), 1–4 (IEEE, 2023).
  38. Moody, B. et al. Mimic-iv waveform database (version 0.1.0), DOI: 10.13026/a2mw-f949 (2022).
  39. Secnlp: A survey of embeddings in clinical natural language processing. \JournalTitleJournal of biomedical informatics 101, 103323 (2020).
  40. Vaswani, A. et al. Attention is all you need. \JournalTitleAdvances in neural information processing systems 30 (2017).
  41. Assessment of non-invasive blood pressure prediction from ppg and rppg signals using deep learning. \JournalTitleSensors 21, 6022 (2021).
  42. Ibtehaz, N. et al. Ppg2abp: Translating photoplethysmogram (ppg) signals to arterial blood pressure (abp) waveforms. \JournalTitleBioengineering 9, 692 (2022).
  43. Blood pressure estimation from photoplethysmogram using a spectro-temporal deep neural network. \JournalTitleSensors 19, 3420 (2019).
  44. Concatenated convolutional neural network model for cuffless blood pressure estimation using fuzzy recurrence properties of photoplethysmogram signals. \JournalTitleScientific Reports 12, 6633 (2022).
  45. Pay less attention with lightweight and dynamic convolutions. \JournalTitlearXiv preprint arXiv:1901.10430 (2019).
  46. Bearing fault diagnosis method based on a multi-head graph attention network. \JournalTitleMeasurement Science and Technology 33, 075012 (2022).
  47. Time series forecasting with multi-headed attention-based deep learning for residential energy consumption. \JournalTitleEnergies 13, 4722 (2020).
  48. Masked multi-head self-attention for causal speech enhancement. \JournalTitleSpeech Communication 125, 80–96 (2020).
  49. Multi-task time series forecasting with shared attention. In 2020 International Conference on Data Mining Workshops (ICDMW), 917–925 (IEEE, 2020).
  50. A multi-head attention-based transformer model for traffic flow forecasting with a comparative analysis to recurrent neural networks. \JournalTitleExpert Systems with Applications 202, 117275 (2022).
  51. Comparing attention-based convolutional and recurrent neural networks: Success and limitations in machine reading comprehension. \JournalTitlearXiv preprint arXiv:1808.08744 (2018).
  52. Attention-based sequence-to-sequence model for time series imputation. \JournalTitleEntropy 24, 1798 (2022).
  53. Attention as robust representation for time series forecasting. \JournalTitlearXiv preprint arXiv:2402.05370 (2024).
  54. Attention-based interpretable neural network for building cooling load prediction. \JournalTitleApplied Energy 299, 117238 (2021).
  55. Ahmed, S. et al. Transformers in time-series analysis: A tutorial. \JournalTitlearXiv preprint arXiv:2205.01138 (2022).
  56. Beat-to-beat continuous blood pressure estimation with optimal feature set of ppg and ecg signals using deep recurrent neural networks. \JournalTitleVessel Plus 7 (2023).
  57. A ppg-based calibration-free cuffless blood pressure estimation method using cardiovascular dynamics. \JournalTitleSensors 23, 4145 (2023).
  58. Ppg signals-based blood-pressure estimation using grid search in hyperparameter optimization of cnn–lstm. \JournalTitleDiagnostics 13, 2566 (2023).
  59. A svm method for continuous blood pressure estimation from a ppg signal. In Proceedings of the 9th international conference on machine learning and computing, 128–132 (2017).
  60. He, R. et al. Beat-to-beat ambulatory blood pressure estimation based on random forest. In 2016 IEEE 13th international conference on wearable and implantable body sensor networks (BSN), 194–198 (IEEE, 2016).
  61. Continuous blood pressure estimation from two-channel ppg parameters by xgboost. In 2019 IEEE International Conference on Robotics and Biomimetics (ROBIO), 2707–2712 (IEEE, 2019).
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (5)
  1. Amir Arjomand (1 paper)
  2. Amin Boudesh (2 papers)
  3. Farnoush Bayatmakou (4 papers)
  4. Kenneth B. Kent (5 papers)
  5. Arash Mohammadi (69 papers)