A Heart Failure Classification Model from Radial Artery Pulse Wave Using LSTM Neural Networks

A Heart Failure Classification Model from Radial Artery Pulse Wave Using LSTM Neural Networks | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article A Heart Failure Classification Model from Radial Artery Pulse Wave Using LSTM Neural Networks Yi Lyu, Wen-Yue Huang, Hai-Mei Wu, Jing Hong, Yi-Qin Wang, Hai-Xia Yan, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5442852/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 27 Aug, 2025 Read the published version in BMC Medical Informatics and Decision Making → Version 1 posted 4 You are reading this latest preprint version Abstract Background The pressing global health issue of heart failure (HF) demands innovative approaches for early detection. Non-invasive, rapid, and cost-effective deep learning (DL)-based techniques offer a promising avenue for addressing this challenge. Methods A total of 462 participants were categorized into three groups: healthy, coronary artery disease (CAD), and heart failure (HF). Raw radial artery pulse wave data were collected from each participant, followed by preprocessing steps including denoising, normalization, and balancing. Subsequently, four deep learning (DL) algorithms were applied to the processed data: Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), Gated Recurrent Unit (GRU), and Bidirectional Long Short-Term Memory (Bi-LSTM). Results LSTM achieved the highest classification performance, with an accuracy of 0.8587, precision of 0.87448, recall of 0.82164, F1-score of 0.83773, specificity of 0.92369, and AUC of 0.93365. Given its superior performance across all metrics, LSTM emerges as the preferred DL model for this study. Conclusion By employing LSTM to analyze radial artery pulse wave, we can accurately distinguish between healthy individuals, patients with CAD, and those with HF. This simple, non-invasive, and cost-effective method presents a potential strategy for early detection of HF. Heart failure Radial artery pulse wave Long Short-Term Memory Deep learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 Introduction Coronary artery disease (CAD) is a kind of cardiovascular disease which has affected human population in both developed and developing countries[ 1 ], which is also the main cause of death around the world[ 2 ]. Heart failure (HF) is a common complication of CAD[ 3 ]. It is a clinical status consisting of cardinal symptoms due to a structural and/or functional abnormality of the heart[ 4 ]. Despite considerable improvements and significant ongoing efforts to treat and manage HF, the overall incidence is still increasing due to ageing[ 5 ]. Besides, mortality and hospitalization rates among HF patients remain poor[ 6 , 7 ]. Chronic ischemia leads to the progression of left ventricular remodeling and systolic dysfunction that provide the substrate for HF development[ 8 ].. Given the growing HF burden, HF prevention and management are a major public health concern[ 9 ]. Radial artery pulse diagnosis is an integral component of Traditional Chinese Medicine (TCM) diagnosis and one of the four diagnostic methods in TCM. It involves a practitioner using their fingers to palpate the superficial arteries at specific points on a patient’s body to assess the pulse and understand the patient’s condition and differentiate various diseases. Radial artery pulse diagnosis holds a significant position in both TCM theory and practice due to its close relationship with the heart, blood vessels, and the functions of the internal organs. The pulse pattern can convey physiological and pathological information from various parts of the body, providing crucial evidence for diagnosing diseases and differentiating their severity[ 10 – 14 ], allowing TCM to integrate more seamlessly with modern medical practices[ 15 ]. Deep Learning (DL), a novel subfield of machine learning (ML), has achieved significant breakthroughs in various applications such as speech recognition and computer vision. By mimicking the neural connections of the human brain, deep learning models can hierarchically represent data features through multiple transformation stages, enabling the interpretation of complex data like images, sounds, and text. Methods The study protocol was approved by the Institutional Review Board of Shanghai University of Traditional Chinese Medicine (Approval Number:2023-3-10-08-08) and the study complied with the Declaration of Helsinki. Clinical trial number: not applicable. Study subjects This case-control study involved three groups. The healthy group included 202 participants, the CAD group included 187 patients, and the HF group included 73 patients. All participants were recruited from the Shanghai Municipal Hospital of Traditional Chinese Medicine, Shuguang Hospital, Yueyang Hospital, and Longhua Hospital, all affiliated with the Shanghai University of Traditional Medicine, between September 2019 and December 2021. Inclusion criteria The inclusion criteria were as follows: CAD patients should fit the "Nomenclature and criteria for diagnosis of ischemic heart disease"[16]. CAD patients who fit the "Chinese Guidelines for the Diagnosis and Treatment of Heart Failure 2024"[17] and be classified as having New York Heart Association (NYHA) Classification III and IV[18] were selected in the HF group. Participants in the healthy group were required to be free from any cardiovascular disease. All participants were required to finish the collection of radial artery pulse wave and sign the informed consent form. Exclusion criteria The exclusion criteria were as follows: Participants with arrhythmia, valvular heart disease or severe myocarditis. Participants with severe endocrine, blood, metabolic system diseases, severe gastrointestinal disease or kidney diseases. Participants with malignant tumors. Participants who refused to participate. Severe incomplete clinical data. Collection of pulse wave A SmartTCM-1 pulse wave digital acquisition analyzer, a device jointly developed by Shanghai University of Traditional Chinese Medicine and Shanghai Asia & Pacific Computer Information System Co., Ltd., was used to record radial artery pulse wave signals from the participant’s left wrist. The participant was instructed to maintain a supine or seated position with their forearm extended, wrist straight, palm up, and fingers slightly curved. A soft pulse cushion was placed under the wrist joint. After a 3-minute rest period, 60 seconds of pulse wave data were collected at a sampling rate of 720 Hz. The PulseSystem software, collaboratively developed by our research group and East China University of Science and Technology (Shanghai), was employed to denoise and extract raw pulse wave data. Subsequent data processing and calibration were performed using PulseAnalyseGraphic v1.1, followed by data export for further analysis. Time-domain parameters of the radial pulse wave signal Figure 1 shows the classical time-domain parameters used in pulse wave analysis. Table 1 shows the definition and meaning of classical time-domain parameters of pulse wave. Table 1 The definition and meaning of classical time-domain characteristics of pulse wave Pulse time-domain characteristics Definition and meaning h 1 Main wave amplitude h 3 Tidal wave amplitude h 4 Dicrotic notch amplitude h 5 Dicrotic wave amplitude t 1 Time between starting point to crest of main wave t 3 Time between starting point to crest of tidal wave t 4 Time between the starting point to the dicrotic notch t 5 Time between the dicrotic notch to the ending point t Time for a complete pulse cycle w Width of the main wave at its 1/3 height As Area under the curve during systolic phase Ad Area under the curve during diastolic phase h 3 / h 1 Vascular wall compliance and peripheral resistance h 4 / h 1 Level of peripheral resistance h 5 / h 1 Aortic compliance and aortic valve function w / t Duration of elevated aortic pressure The moving average was calculated based on the moving window widths of 1/3, 1/4, 1/5, and 1/6 of the sampling frequency[19–22], with the maximum values represented by h f /3 , h f /4 , h f /5 , and h f /6 , respectively. The values at which the pulse wave pressure first reaches this maximum value after the main wave peak are represented by t f /3 , t f /4 , t f /5 , and t f /6 (Fig. 2 )[23]. The time at which the waveform reached its maximum value ( t max ) was compared to traditional t 1 and t 3 values. Additionally, the time points at which the waveform rose to 80% ( t 0.8 ) and 90% ( t 0.9 ) of its maximum amplitude were determined (Fig. 3 ). Data normalization The raw data underwent min-max normalization. This transformation preserves the relative order of data points while rescaling values to the range of [0, 1]. Let \(\:{x}_{min}\) denote the minimum value of a specific value in a feature, and \(\:{x}_{max}\) denote the maximum value of a specific value in a feature, the conversion function is as follows: $$\:\begin{array}{c}{x}^{*}=\frac{x-{x}_{min}}{{x}_{max}-{x}_{min}}\#\left(3\right)\end{array}$$ Data balance processing Synthetic Minority Over-sampling Technique (SMOTE)[24] is a widely-used technique for addressing imbalanced datasets. By generating new synthetic samples to increase the number of instances in the minority class, SMOTE helps to balance the class distribution. Unlike simple replication, SMOTE employs a linear interpolation method to create new samples, mitigating the risk of overfitting. The synthetic samples will be randomly assigned to both the training and validation sets, while the test set will remain unchanged. Deep learning classification ML covers various knowledge and technologies such as probability theory, statistics, and complex algorithms[25–27]. DL is a specific type of ML that uses artificial neural networks with multiple layers to learn complex patterns from large amounts of data. These neural networks are inspired by the structure and function of the human brain. Traditional ML algorithms may be more suitable for smaller datasets or simpler problems. However, as computing power and data availability have increased, DL has become the dominant approach in many areas of ML. This study used 4 DL algorithms such as Long Short-Term Memory (LSTM)[28], Convolutional Neural Network (CNN), Gated Recurrent Unit (GRU)[29], and Bidirectional Long Short-Term Memory (Bi-LSTM). LSTM is a type of recurrent neural network (RNN) architecture that is specifically designed to learn long-term dependencies in sequential data. Unlike traditional RNNs, LSTMs have a special cell state that can store information for long periods, making them well-suited for tasks. CNN is a type of DL neural network that is primarily used for image recognition, object detection, and other computer vision tasks. CNNs can learn features directly from data, reducing the need for manual feature engineering. GRU is also a type of RNN designed to capture long-term dependencies in sequential data. It's similar to LSTM but has a simpler architecture, making it computationally more efficient. Bi-LSTM is a type of RNN architecture that is particularly well-suited for sequence modeling tasks. Unlike traditional LSTM neural networks, which process sequences in a forward direction only, Bi-LSTM networks process sequences in both forward and backward directions. The dataset was split into train set, validation set, and test set in the ratio of 6:2:2. A validation set is a portion of a dataset used to assess model performance during training, aiding in hyperparameter tuning and overfitting prevention. The final evaluation was based on the results of the test set. Figure 4 shows the flowchart of the DL classification process. Classification metrics In this study, the main metrics used for model performance evaluation are Accuracy, Precision, Recall, F1, Specificity, and area under curve (AUC). Confusion Matrix The confusion matrix (shown in Fig. 5) is used to summarize the results of a classifier and is a standard format for accuracy evaluation. TP(True Positive): Positive sample predicted by the model as the positive category. The larger the TP value, the better the model. FN(False Negative): Positive sample predicted by the model as the negative category. The smaller the FN value, the better the model. FP(False Positive): Negative sample predicted by the model as the positive category. The smaller the FP value, the better the model. TN(True Negative): Negative sample predicted by the model as the negative category. The larger the TN value, the better the model. $$\:\begin{array}{c}Precision=\frac{TP}{TP+FP}\#\left(4\right)\end{array}$$ $$\:\begin{array}{c}Recall=\frac{TP}{TP+FN}\#\left(5\right)\end{array}$$ $$\:\begin{array}{c}Accuracy=\frac{TP+TN}{P+N}\#\left(6\right)\end{array}$$ $$\:\begin{array}{c}{F}_{1}=2\times\:\frac{P\times\:R}{P+R}\#\left(7\right)\end{array}$$ $$\:\begin{array}{c}Specificity=\frac{TN}{TN+FP}\#\left(8\right)\end{array}$$ Receiver Operating Characteristic Curve The Receiver Operating Characteristic (ROC) curve is a graphical representation used to evaluate the diagnostic performance of a binary classification model. It plots the TP rate (sensitivity) against the FP rate (1-specificity) at various threshold settings. The area under curve (AUC) quantifies the model’s ability to discriminate between positive and negative classes, with values ranging from 0 to 1. An AUC of 0.5 indicates no discrimination (equivalent to random guessing), while an AUC of 1.0 signifies perfect discrimination. The ROC curve serves as a critical tool for assessing the trade-offs between sensitivity and specificity, thereby aiding in the selection of optimal thresholds for clinical decision-making[30, 31]. Results The original data underwent SMOTE oversampling to balance the class distribution (shown in Figure 6). Figure 7 displays the distribution of each feature for both the original data and the synthetic data. Figure 8 visualizes the high-dimensional data generated by t-distributed stochastic neighbor embedding (t-SNE)[32] dimensionality reduction to 2D, allowing for a more intuitive comparison between original and synthetic data. Table 2 shows the performance of four DL models. The final evaluation aimed to check the general ability of models to predict unseen data. Table 2 The performance comparison of different DL models Model Accuracy Precision Recall F1 Specificity AUC LSTM 0.85870 0.87448 0.82164 0.83773 0.92369 0.93365 GRU 0.82609 0.81482 0.78417 0.79352 0.90839 0.93125 CNN 0.81522 0.80852 0.77703 0.78498 0.90255 0.90817 Bi-LSTM 0.79348 0.78551 0.74671 0.75735 0.89029 0.91875 Figure 8 shows the confusion matrix for test data. Figure 9 shows the predicted and true value for test data. Figure 10 presents a comparison chart of metrics for different models. Figure 11 presents the ROC and AUC of test data from LSTM model. Table 3 shows the analysis for LSTM trainNetwork usage. Table 3 The analysis for LSTM trainNetwork usage Name Type Activations Learnable Properties States Sequence input with 68 dimensions Sequence Input 68(C)×1(B)×1(T) - - LSTM with 16 hidden units LSTM 16(C)×1(B)×1(T) InputWeights 64×68 RecurrentWeights 64×16 Bias 64×1 HiddenState 16×1 CellState 16×1 20% dropout Dropout 16(C)×1(B)×1(T) - - LSTM with 16 hidden units LSTM 16(C)×1(B) InputWeights 64×16 RecurrentWeights 64×16 Bias 64×1 HiddenState 16×1 CellState 16×1 20% dropout Dropout 16(C)×1(B) - - 3 fully connected layer Fully Connected 3(C)×1(B) Weights 3×16 Bias 3×1 - Softmax Softmax 3(C)×1(B) - - Classoutput Classification Output 3(C)×1(B) - - Conclusion and Discussion In this study, we employed four DL algorithms to investigate the radial artery pulse wave characteristics of healthy, CAD, and HF participants. Table 2 demonstrates that LSTM achieved the highest classification performance, with an accuracy of 0.8587, precision of 0.87448, recall of 0.82164, F1-score of 0.83773, specificity of 0.92369, and AUC of 0.93365. Given its superior performance across all metrics, LSTM emerges as the preferred DL model for this study. An early diagnosis of HF may reduce patients’ mortality and morbidity. Consequently, wide efforts have been put to develop algorithms in diagnosing [33]. HF diagnosis may be achieved through the analysis of electrocardiography, magnetic resonance imaging, ultrasound images as well as electronic health records [34-36]. These methods, however, are limited due to the requirement for specialized equipment and skilled operators. The ability to identify HF using only radial artery pulse wave could be used for early diagnosis and would enable referral for further investigation, offering a noninvasive, cost-effective solution. Wu et al. evaluated the potential of wrist pulse signals for use in the cardiac monitoring of patients with CHD. Decision tree (DT) and random forest (RF) algorithms were employed to construct models for classifying coronary heart disease (CHD) patients with different brain natriuretic peptide. The pulse features of the three groups differed significantly, suggesting different pathological states of the cardiovascular system in patients with CHD. Moreover, the RF models outperformed the DT models in performance metrics[37]. Wu et al. proposed a ML-based strategy to identify left ventricular enlargement (LVE) in HF patients by means of pulse wave signals by means of classification and regression models. The classification model exhibited superior performance with an accuracy of 0.91 and an AUC-ROC of 0.93. The regression model achieved an accuracy of 0.88 and an AUC-ROC of 0.89, indicating that both models can quickly and accurately identify LVE in HF patients[38]. In 2023, Yan et al. utilized RF, Support Vector Machine, k-nearest neighbor, and DT algorithms to develop classification models using microcirculatory characteristic parameter set. The results showed that RF showed good classification performance, the identification accuracy of the model built on the microcirculatory characteristic parameter set and RF algorithm all reached more than 88%. The highest recognition accuracy was 95.51% for coronary heart disease samples, 92.11% for healthy samples, and 88.55% for hypertensive samples. These researches demonstrated radial artery pulse waves as a key informational conduit within the cardiovascular system and ML as a more suitable method to deal with complex wrist pulse data. While previous studies have primarily relied on traditional ML techniques such as DT and RF, this study introduces a novel approach that incorporates DL algorithms. In addition to the standard features used in previous research, this study further leverages moving average features and the time points at which the radial artery pulse wave reaches 80% and 90% of its peak amplitude. These additional features provide a more comprehensive and nuanced representation of the pulse wave morphology, particularly in cases of HF where the rapid ejection phase and fusion of the main wave and pre-beat wave can be subtle and difficult to discern. By employing DL algorithms, this study achieves superior performance in terms of accuracy, recall, and AUC, while also mitigating overfitting and enhancing the scalability of the model. At the same time, this study is also beneficial for the modern inheritance and international development of TCM. This study elucidates the correlation between pulse diagnosis and HF. Given the high prevalence and severe implications of HF, this DL-based model offers a systematic method to "learn" the association between pulse wave data, CAD, and HF. While this learning approach may not fully elucidate the underlying biological mechanisms, it has the potential to significantly aid in the early diagnosis of HF in clinical settings. This approach is advantageous due to its simplicity, non-invasive nature, and low cost. Wearable devices are portable medical or health electronic devices that incorporate sensors, wireless communication, and multimedia technologies into items worn directly on the body. Supported by software, these devices can sense, record, analyze, regulate, intervene in, and even treat diseases or maintain health conditions. Leveraging human or environmental capabilities, they enable intelligent information interaction through built-in sensors and integrated chips. Wearable devices serve as the ideal platform for the ongoing integration and innovation of Internet of Things, mobile internet, cloud storage, and big data technologies. This study may serve as a foundation for developing wearable devices designed to detect cardiovascular abnormalities. Furthermore, this approach can be extended to a wider range of diagnostic applications in the future. Several potential obstacles may hinder our progress: (1) To date, our dataset comprises approximately 1500 samples from CAD and HF patients. While this provides a valuable foundation for our study, a larger dataset is required to optimize deep learning models and achieve superior performance in terms of accuracy, recall, and AUC. Consequently, we plan to expand our dataset in future research to enhance model robustness and generalization. (2) To date, our dataset comprises approximately 1800 samples from CAD and HF patients. While this provides a valuable foundation for our study, a larger dataset is required to optimize deep learning models and achieve superior performance in terms of accuracy, recall, and AUC. Consequently, we plan to expand our dataset in future research to enhance model robustness and generalization. (3) The current sampling frequency is set at 720 Hz. To enhance the precision of our measurements, future studies will explore the feasibility of increasing the sampling rate of the device. Abbreviations CAD Coronary artery disease HF Heart failure TCM Traditional Chinese Medicine DL Deep learning ML Machine learning NYHA New York Heart Association SMOTE Synthetic Minority Over-sampling Technique LSTM Long Short-Term Memory CNN Convolutional Neural Network GRU Gated Recurrent Unit Bi-LSTM Bidirectional Long Short-Term Memory AUC Area under curve TP True Positive FN False Negative FP False Positive TN True Negative ROC Receiver Operating Characteristic t-SNE t-distributed stochastic neighbor embedding DT Decision Tree RF Random Forest CHD Coronary heart disease LVE Left ventricular enlargement Declarations Authors’ contributions Y. L and W.-Y. H provided the research conceptualization and provided the research methods, contributed equally to this work. Y. L, W.-Y. H, and J. H acquired the data. Y. L and H.-M. W analyzed the data and performed deep learning. Y.-Q. W, H.-X. Y, and J. X in charge of funding acquisition. All authors edited and revised the manuscript and approved the final version. Competing interests The authors declare no competing interests. Ethics approval and consent to participate The study protocol was approved by the Institutional Review Board of Shanghai University of Traditional Chinese Medicine (Approval Number:2023-3-10-08-08) and the study complied with the Declaration of Helsinki. All participants received informed consent form and signed. Clinical trial number: not applicable. Data Availability The datasets generated and/or analyzed during the current study are not publicly available due to ethical concern but are available from the corresponding author on reasonable request. Consent for publication Not applicable. Funding This work is supported by the National Natural Science Foundation of China (NSFC, Grant No.81673880), Shanghai Key Laboratory of Health Identification and Assessment Project, China (Grant No.21DZ2271000). Acknowledgements Not applicable. References Global, regional, and national age-sex specific all-cause and cause-specific mortality for 240 causes of death, 1990-2013: a systematic analysis for the Global Burden of Disease Study 2013 . Lancet (London, England) 2015, 385 (9963):117-171. Mastoi QU, Wah TY, Gopal Raj R, Iqbal U: Automated Diagnosis of Coronary Artery Disease: A Review and Workflow . Cardiology research and practice 2018, 2018 :2016282. Hatfield J, Woods MD, Hammonds K, Wahab L, Nguyen V, Hassan OA, Butler J, Widmer RJ: Global hypokinesis in resting transthoracic echocardiography diagnosis of heart failure and coronary artery disease . The Journal of invasive cardiology 2024. McDonagh TA, Metra M, Adamo M, Gardner RS, Baumbach A, Böhm M, Burri H, Butler J, Čelutkienė J, Chioncel O et al : 2023 Focused Update of the 2021 ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure: Developed by the task force for the diagnosis and treatment of acute and chronic heart failure of the European Society of Cardiology (ESC) With the special contribution of the Heart Failure Association (HFA) of the ESC . European journal of heart failure 2024, 26 (1):5-17. Martin SS, Aday AW, Almarzooq ZI, Anderson CAM, Arora P, Avery CL, Baker-Smith CM, Barone Gibbs B, Beaton AZ, Boehme AK et al : 2024 Heart Disease and Stroke Statistics: A Report of US and Global Data From the American Heart Association . Circulation 2024, 149 (8):e347-e913. Roger VL: Epidemiology of Heart Failure: A Contemporary Perspective . Circulation research 2021, 128 (10):1421-1434. Conrad N, Judge A, Canoy D, Tran J, Pinho-Gomes AC, Millett ERC, Salimi-Khorshidi G, Cleland JG, McMurray JJV, Rahimi K: Temporal Trends and Patterns in Mortality After Incident Heart Failure: A Longitudinal Analysis of 86 000 Individuals . JAMA cardiology 2019, 4 (11):1102-1111. Triposkiadis F, Xanthopoulos A, Parissis J, Butler J, Farmakis D: Pathogenesis of chronic heart failure: cardiovascular aging, risk factors, comorbidities, and disease modifiers . Heart failure reviews 2022, 27 (1):337-344. Tomasoni D, Adamo M, Lombardi CM, Metra M: Highlights in heart failure . ESC heart failure 2019, 6 (6):1105-1127. Zhang J, Liao J, Wang T, Yuan R, Zhao Y, Han Z, Tang L, Zhao L: Effects of joy and sorrow on pulse-graph parameters in healthy female college students based on emotion-evoked experiments . EXPLORE 2020, 17 (4):303-311. Wang W, Zeng W, Chen X, Tu L, Xu J, Yin X: Parameter study on characteristic pulse diagram of polycystic ovary syndrome based on logistic regression analysis . Journal of Obstetrics and Gynaecology 2022, 42 (8):3712-3719. Lim J, Li J, Feng X, Feng L, Xia Y, Xiao X, Wang Y, Xu Z: Machine learning classification of polycystic ovary syndrome based on radial pulse wave analysis . BMC Complement Med Ther 2023, 23 (1). Wan WK, Hsu TL, Chang HC, Wan YY: Effect of acupuncture at Hsien-Ku (St-43) on the pulse spectrum and a discussion of the evidence for the frequency structure of Chinese medicine . The American journal of Chinese medicine 2000, 28 (1):41-55. Lim J, Li J, Feng X, Feng L, Xiao X, Xia Y, Wang Y, Qian L, Yang H, Xu Z: Machine learning-based evaluation of application value of traditional Chinese medicine clinical index and pulse wave parameters in the diagnosis of polycystic ovary syndrome . European Journal of Integrative Medicine 2023, 64 :102311. Baik Y: A study on The Characteristic of Traditional Pediatric Pulse Diagnosis . Journal of Korean Medical classics 2014, 27 (1):111-122. Nomenclature and criteria for diagnosis of ischemic heart disease. Report of the Joint International Society and Federation of Cardiology/World Health Organization task force on standardization of clinical nomenclature . Circulation 1979, 59 (3):607-609. Chinese Society of Cardiology CMA, Chinese College of Cardiovascular Physician, Chinese Heart Failure Association of Chinese Medical Doctor Association, Editorial Board of Chinese Journal of Cardiology: Chinese guidelines for the diagnosis and treatment of heart failure 2024 . Chinese Journal of Cardiology 2024, 52 (3):235-275. White PD, Myers MJJ: The classification of cardiac diagnosis . 1921, 77 (18):1414-1415. Ding X, Wang Y, Hao Y, Lv Y, Chen R, Yan H: A New Measure of Pulse Rate Variability and Detection of Atrial Fibrillation Based on Improved Time Synchronous Averaging . Computational and mathematical methods in medicine 2021, 2021 :5597559. Nichols WW ORM, Vlachopoulos C, et al.: McDonald's Blood Flow in Arteries: Theoretical, Experimental and Clinical Principles . London: Hodder Arnold; 2011. Nichols WW: Clinical measurement of arterial stiffness obtained from noninvasive pressure waveforms . American journal of hypertension 2005, 18 (1 Pt 2):3s-10s. Xiao H, Butlin M, Qasem A, Tan I, Li D, Avolio AP: N-Point Moving Average: A Special Generalized Transfer Function Method for Estimation of Central Aortic Blood Pressure . IEEE transactions on bio-medical engineering 2018, 65 (6):1226-1234. Quail MA, Steeden JA, Knight D, Segers P, Taylor AM, Muthurangu V: Development and validation of a novel method to derive central aortic systolic pressure from the MR aortic distension curve . Journal of magnetic resonance imaging : JMRI 2014, 40 (5):1064-1070. Chawla N, Bowyer K, Hall L, Kegelmeyer W: SMOTE: Synthetic Minority Over-sampling Technique . jair 2002, abs/1106.1813 :321-357. Choi RY, Coyner AS, Kalpathy-Cramer J, Chiang MF, Campbell JP: Introduction to Machine Learning, Neural Networks, and Deep Learning . Translational vision science & technology 2020, 9 (2):14. Deo RC: Machine Learning in Medicine . Circulation 2015, 132 (20):1920-1930. Sun X, Yin Y, Yang Q, Huo T: Artificial intelligence in cardiovascular diseases: diagnostic and therapeutic perspectives . European journal of medical research 2023, 28 (1):242. Hochreiter SJNCM-P: Long Short-term Memory . 1997. Cho KJapa: Learning phrase representations using RNN encoder-decoder for statistical machine translation . 2014. James G WD, Hastie T, Tibshirani R: An Introduction to Statistical Learning: With Applications in R . New York: Springer; 2013. Hastie T TR, Friedman JH: The Elements of Statistical Learning: Data Mining, Inference, and Prediction. 2nd ed . New York: Springer; 2009. Maaten Lvd, Hinton GEJJoMLR: Visualizing Data using t-SNE . 2008, 9 :2579-2605. Penso M, Solbiati S, Moccia S, Caiani EG: Decision Support Systems in HF based on Deep Learning Technologies . Current heart failure reports 2022, 19 (2):38-51. Kwon JM, Kim KH, Jeon KH, Kim HM, Kim MJ, Lim SM, Song PS, Park J, Choi RK, Oh BH: Development and Validation of Deep-Learning Algorithm for Electrocardiography-Based Heart Failure Identification . Korean circulation journal 2019, 49 (7):629-639. Lau ES, Di Achille P, Kopparapu K, Andrews CT, Singh P, Reeder C, Al-Alusi M, Khurshid S, Haimovich JS, Ellinor PT et al : Deep Learning-Enabled Assessment of Left Heart Structure and Function Predicts Cardiovascular Outcomes . Journal of the American College of Cardiology 2023, 82 (20):1936-1948. McGilvray MMO, Heaton J, Guo A, Masood MF, Cupps BP, Damiano M, Pasque MK, Foraker R: Electronic Health Record-Based Deep Learning Prediction of Death or Severe Decompensation in Heart Failure Patients . JACC Heart failure 2022, 10 (9):637-647. Wu WJ, Chen R, Guo R, Yan JJ, Zhang CK, Wang YQ, Yan HX, Zhang YQ: A novel method for assessing cardiac function in patients with coronary heart disease based on wrist pulse analysis . Irish journal of medical science 2023, 192 (6):2697-2706. Wu D, Ono R, Wang S, Kobayashi Y, Sughimoto K, Liu H: Pulse wave signal-driven machine learning for identifying left ventricular enlargement in heart failure patients . Biomedical engineering online 2024, 23 (1):60. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 27 Aug, 2025 Read the published version in BMC Medical Informatics and Decision Making → Version 1 posted Editorial decision: Revision requested 24 Nov, 2024 Editor assigned by journal 24 Nov, 2024 Submission checks completed at journal 16 Nov, 2024 First submitted to journal 12 Nov, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5442852","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":382069948,"identity":"1d1465a0-c73a-443e-a2f6-879f0097f5a2","order_by":0,"name":"Yi Lyu","email":"","orcid":"","institution":"Shanghai University of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Yi","middleName":"","lastName":"Lyu","suffix":""},{"id":382069949,"identity":"b33ff3aa-1cbb-4a07-8ae4-f7f00e2fc9ca","order_by":1,"name":"Wen-Yue Huang","email":"","orcid":"","institution":"Shuguang Hospital","correspondingAuthor":false,"prefix":"","firstName":"Wen-Yue","middleName":"","lastName":"Huang","suffix":""},{"id":382069950,"identity":"9a1fb7ae-fd4e-4f4a-8314-40783fe3b8f7","order_by":2,"name":"Hai-Mei Wu","email":"","orcid":"","institution":"Shanghai Lingyun Sub-district Health Service Center","correspondingAuthor":false,"prefix":"","firstName":"Hai-Mei","middleName":"","lastName":"Wu","suffix":""},{"id":382069951,"identity":"171db05b-fc49-4f5e-8098-3977ebb62763","order_by":3,"name":"Jing Hong","email":"","orcid":"","institution":"Shanghai University of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Jing","middleName":"","lastName":"Hong","suffix":""},{"id":382069952,"identity":"b1e66fc2-f175-438b-ad63-1abe4d537498","order_by":4,"name":"Yi-Qin Wang","email":"","orcid":"","institution":"Shanghai University of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Yi-Qin","middleName":"","lastName":"Wang","suffix":""},{"id":382069953,"identity":"8bd4c791-308f-49de-b059-62411ec5171c","order_by":5,"name":"Hai-Xia Yan","email":"","orcid":"","institution":"Shanghai University of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Hai-Xia","middleName":"","lastName":"Yan","suffix":""},{"id":382069954,"identity":"ae4726f8-2d48-430f-8db2-d9f6e7957c28","order_by":6,"name":"Jin Xu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA7ElEQVRIiWNgGAWjYHCChAMMP2zkGNt7wDwePqK0MPakGTP3nGFgOADUwkacRWyHE9ln5IC1MBDUYnAj4eHhAh7mBN6Zbw8+/phjJ8PGwPzw0Q08WiRnJCQcnmHBlic5Oy/Z4OC2ZKDD2IyNc/Bo4ZcAauHh4Sk2nJ1jJnFwGzNQCw+bND4tbGAtbBKJ+2+eAWmpJ6wFYgubQWLjDB6QlsOEtUj2PEg4zNuTYMzYk2NscHbbcR42ZgJ+MTiek/yZ58d/YFSeMXxQua3anp+9+eFjfFqAcZeAJsCMVzkIsB8gqGQUjIJRMApGOAAAXwxIG4LNC0YAAAAASUVORK5CYII=","orcid":"","institution":"Shanghai University of Traditional Chinese Medicine","correspondingAuthor":true,"prefix":"","firstName":"Jin","middleName":"","lastName":"Xu","suffix":""}],"badges":[],"createdAt":"2024-11-13 01:53:18","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5442852/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5442852/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12911-025-03167-5","type":"published","date":"2025-08-27T15:56:52+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":71799242,"identity":"9a8a9634-42e7-4c23-bf3f-074392e2a6d7","added_by":"auto","created_at":"2024-12-18 16:32:22","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":62928,"visible":true,"origin":"","legend":"\u003cp\u003eThe classical time-domain parameters\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-5442852/v1/2e1d70180db4600103dad7e2.png"},{"id":71799239,"identity":"6f19c40a-ed50-4456-8013-c763d4490457","added_by":"auto","created_at":"2024-12-18 16:32:22","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":73445,"visible":true,"origin":"","legend":"\u003cp\u003eMoving average of the radial artery pulse wave\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-5442852/v1/7095a4697f96446a3f6837ee.png"},{"id":71798801,"identity":"fbfc91d4-a3c0-45d6-bcdb-b2c52aec85f7","added_by":"auto","created_at":"2024-12-18 16:24:22","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":43838,"visible":true,"origin":"","legend":"\u003cp\u003eSchematic diagram of the main wave correlation time value indicator\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-5442852/v1/b190649ad63ebb86e26fe1ee.png"},{"id":71798800,"identity":"d336142e-0177-4459-82fe-60ce5fe72311","added_by":"auto","created_at":"2024-12-18 16:24:21","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":70990,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of the collection and analysis of pulse wave and the process of ML\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-5442852/v1/c9ecc9b64a2e76a203fb3363.png"},{"id":71798803,"identity":"03ddb3c8-5e62-4d7f-9f6c-88340a72797f","added_by":"auto","created_at":"2024-12-18 16:24:22","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":14524,"visible":true,"origin":"","legend":"\u003cp\u003eConfusion matrix\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-5442852/v1/a507a8509285c3dbc64768b6.png"},{"id":71800390,"identity":"002bdf59-3d3b-4db1-b15a-8c6aae855e9b","added_by":"auto","created_at":"2024-12-18 16:40:22","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":63913,"visible":true,"origin":"","legend":"\u003cp\u003eThe difference between original data and synthetic data\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-5442852/v1/2bf18a43e648d917fc8b0550.png"},{"id":71798804,"identity":"4c655a4c-d1ce-4ff9-8342-3721ac9e3748","added_by":"auto","created_at":"2024-12-18 16:24:22","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":242022,"visible":true,"origin":"","legend":"\u003cp\u003eDistribution of original data and synthetic data\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-5442852/v1/dde15d5757250413b4d6220e.png"},{"id":71798807,"identity":"d1622da7-bfac-4b32-8b0d-3558036db90a","added_by":"auto","created_at":"2024-12-18 16:24:22","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":238283,"visible":true,"origin":"","legend":"\u003cp\u003eThe t-SNE dimension of original data and synthetic data\u003c/p\u003e","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-5442852/v1/90883d75fae6316ca89211c9.png"},{"id":71799240,"identity":"755efa61-031d-4556-ae3d-37928cd1fe6b","added_by":"auto","created_at":"2024-12-18 16:32:22","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":49980,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFig.8\u003c/strong\u003e Confusion matrix for test data\u003c/p\u003e","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-5442852/v1/62950f0797085c6e7635539c.png"},{"id":71798809,"identity":"c5b198c4-27bb-468d-b899-184d9a5d6222","added_by":"auto","created_at":"2024-12-18 16:24:22","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":52377,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFig.9 \u003c/strong\u003eResult for test data\u003c/p\u003e","description":"","filename":"floatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-5442852/v1/9dd4c6a421ab0d47a1b47fed.png"},{"id":71799241,"identity":"42a3ae1d-59de-476b-954a-3586e1922f05","added_by":"auto","created_at":"2024-12-18 16:32:22","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":74301,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFig.10\u003c/strong\u003e A comparison chart of metrics for different models\u003c/p\u003e","description":"","filename":"floatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-5442852/v1/ec4b4188a2ae9fa10bccef7f.png"},{"id":71798806,"identity":"24ac9727-616b-430f-bdac-a73b48b056f9","added_by":"auto","created_at":"2024-12-18 16:24:22","extension":"png","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":32628,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFig.11 \u003c/strong\u003eThe ROC and AUC of test data from LSTM model\u003c/p\u003e","description":"","filename":"floatimage11.png","url":"https://assets-eu.researchsquare.com/files/rs-5442852/v1/c347e0623c128c3700b9700c.png"},{"id":90345091,"identity":"cc5de992-b84a-4190-b63d-e0ffc4efb8f3","added_by":"auto","created_at":"2025-09-01 16:09:54","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3470814,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5442852/v1/4149e8a5-389c-4196-84bd-7085f8e69ae0.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A Heart Failure Classification Model from Radial Artery Pulse Wave Using LSTM Neural Networks","fulltext":[{"header":"Introduction","content":"\u003cp\u003eCoronary artery disease (CAD) is a kind of cardiovascular disease which has affected human population in both developed and developing countries[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e], which is also the main cause of death around the world[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Heart failure (HF) is a common complication of CAD[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. It is a clinical status consisting of cardinal symptoms due to a structural and/or functional abnormality of the heart[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Despite considerable improvements and significant ongoing efforts to treat and manage HF, the overall incidence is still increasing due to ageing[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Besides, mortality and hospitalization rates among HF patients remain poor[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Chronic ischemia leads to the progression of left ventricular remodeling and systolic dysfunction that provide the substrate for HF development[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e].. Given the growing HF burden, HF prevention and management are a major public health concern[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eRadial artery pulse diagnosis is an integral component of Traditional Chinese Medicine (TCM) diagnosis and one of the four diagnostic methods in TCM. It involves a practitioner using their fingers to palpate the superficial arteries at specific points on a patient\u0026rsquo;s body to assess the pulse and understand the patient\u0026rsquo;s condition and differentiate various diseases. Radial artery pulse diagnosis holds a significant position in both TCM theory and practice due to its close relationship with the heart, blood vessels, and the functions of the internal organs. The pulse pattern can convey physiological and pathological information from various parts of the body, providing crucial evidence for diagnosing diseases and differentiating their severity[\u003cspan additionalcitationids=\"CR11 CR12 CR13\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e], allowing TCM to integrate more seamlessly with modern medical practices[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eDeep Learning (DL), a novel subfield of machine learning (ML), has achieved significant breakthroughs in various applications such as speech recognition and computer vision. By mimicking the neural connections of the human brain, deep learning models can hierarchically represent data features through multiple transformation stages, enabling the interpretation of complex data like images, sounds, and text.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eThe study protocol was approved by the Institutional Review Board of Shanghai University of Traditional Chinese Medicine (Approval Number:2023-3-10-08-08) and the study complied with the Declaration of Helsinki. Clinical trial number: not applicable.\u003c/p\u003e\n\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\n \u003ch2\u003eStudy subjects\u003c/h2\u003e\n \u003cp\u003eThis case-control study involved three groups. The healthy group included 202 participants, the CAD group included 187 patients, and the HF group included 73 patients. All participants were recruited from the Shanghai Municipal Hospital of Traditional Chinese Medicine, Shuguang Hospital, Yueyang Hospital, and Longhua Hospital, all affiliated with the Shanghai University of Traditional Medicine, between September 2019 and December 2021.\u003c/p\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eInclusion criteria\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe inclusion criteria were as follows:\u003c/p\u003e\n\u003col\u003e\n \u003cli\u003eCAD patients should fit the \u0026quot;Nomenclature and criteria for diagnosis of ischemic heart disease\u0026quot;[16].\u003c/li\u003e\n \u003cli\u003eCAD patients who fit the \u0026quot;Chinese Guidelines for the Diagnosis and Treatment of Heart Failure 2024\u0026quot;[17] and be classified as having New York Heart Association (NYHA) Classification III and IV[18] were selected in the HF group.\u003c/li\u003e\n \u003cli\u003eParticipants in the healthy group were required to be free from any cardiovascular disease.\u003c/li\u003e\n \u003cli\u003eAll participants were required to finish the collection of radial artery pulse wave and sign the informed consent form.\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eExclusion criteria\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe exclusion criteria were as follows:\u003c/p\u003e\n\u003col\u003e\n \u003cli\u003eParticipants with arrhythmia, valvular heart disease or severe myocarditis.\u003c/li\u003e\n \u003cli\u003eParticipants with severe endocrine, blood, metabolic system diseases, severe gastrointestinal disease or kidney diseases.\u003c/li\u003e\n \u003cli\u003eParticipants with malignant tumors.\u003c/li\u003e\n \u003cli\u003eParticipants who refused to participate.\u003c/li\u003e\n \u003cli\u003eSevere incomplete clinical data.\u003c/li\u003e\n\u003c/ol\u003e\n\u003ch3\u003eCollection of pulse wave\u003c/h3\u003e\n\u003cp\u003eA SmartTCM-1 pulse wave digital acquisition analyzer, a device jointly developed by Shanghai University of Traditional Chinese Medicine and Shanghai Asia \u0026amp; Pacific Computer Information System Co., Ltd., was used to record radial artery pulse wave signals from the participant\u0026rsquo;s left wrist. The participant was instructed to maintain a supine or seated position with their forearm extended, wrist straight, palm up, and fingers slightly curved. A soft pulse cushion was placed under the wrist joint. After a 3-minute rest period, 60 seconds of pulse wave data were collected at a sampling rate of 720 Hz.\u003c/p\u003e\n\u003cp\u003eThe PulseSystem software, collaboratively developed by our research group and East China University of Science and Technology (Shanghai), was employed to denoise and extract raw pulse wave data. Subsequent data processing and calibration were performed using PulseAnalyseGraphic v1.1, followed by data export for further analysis.\u003c/p\u003e\n\u003ch3\u003eTime-domain parameters of the radial pulse wave signal\u003c/h3\u003e\n\u003cp\u003eFigure \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e shows the classical time-domain parameters used in pulse wave analysis. Table \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e shows the definition and meaning of classical time-domain parameters of pulse wave.\u003c/p\u003e\n\u003cp\u003e\u003c/p\u003e\u0026nbsp;\u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eThe definition and meaning of classical time-domain characteristics of pulse wave\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ePulse time-domain characteristics\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eDefinition and meaning\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eh\u003c/em\u003e\u003csub\u003e1\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMain wave amplitude\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eh\u003c/em\u003e\u003csub\u003e3\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTidal wave amplitude\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eh\u003c/em\u003e\u003csub\u003e4\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDicrotic notch amplitude\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eh\u003c/em\u003e\u003csub\u003e5\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDicrotic wave amplitude\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003et\u003c/em\u003e\u003csub\u003e1\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTime between starting point to crest of main wave\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003et\u003c/em\u003e\u003csub\u003e3\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTime between starting point to crest of tidal wave\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003et\u003c/em\u003e\u003csub\u003e4\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTime between the starting point to the dicrotic notch\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003et\u003c/em\u003e\u003csub\u003e5\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTime between the dicrotic notch to the ending point\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003et\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTime for a complete pulse cycle\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003ew\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWidth of the main wave at its 1/3 height\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eAs\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eArea under the curve during systolic phase\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eAd\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eArea under the curve during diastolic phase\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eh\u003c/em\u003e\u003csub\u003e3\u003c/sub\u003e/\u003cem\u003eh\u003c/em\u003e\u003csub\u003e1\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVascular wall compliance and peripheral resistance\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eh\u003c/em\u003e\u003csub\u003e4\u003c/sub\u003e/\u003cem\u003eh\u003c/em\u003e\u003csub\u003e1\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLevel of peripheral resistance\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eh\u003c/em\u003e\u003csub\u003e5\u003c/sub\u003e/\u003cem\u003eh\u003c/em\u003e\u003csub\u003e1\u003c/sub\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAortic compliance and aortic valve function\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003ew\u003c/em\u003e/\u003cem\u003et\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDuration of elevated aortic pressure\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003eThe moving average was calculated based on the moving window widths of 1/3, 1/4, 1/5, and 1/6 of the sampling frequency[19\u0026ndash;22], with the maximum values represented by \u003cem\u003eh\u003c/em\u003e\u003csub\u003e\u003cem\u003ef\u003c/em\u003e/3\u003c/sub\u003e, \u003cem\u003eh\u003c/em\u003e\u003csub\u003e\u003cem\u003ef\u003c/em\u003e/4\u003c/sub\u003e, \u003cem\u003eh\u003c/em\u003e\u003csub\u003e\u003cem\u003ef\u003c/em\u003e/5\u003c/sub\u003e, and \u003cem\u003eh\u003c/em\u003e\u003csub\u003e\u003cem\u003ef\u003c/em\u003e/6\u003c/sub\u003e, respectively. The values at which the pulse wave pressure first reaches this maximum value after the main wave peak are represented by \u003cem\u003et\u003c/em\u003e\u003csub\u003e\u003cem\u003ef\u003c/em\u003e/3\u003c/sub\u003e, \u003cem\u003et\u003c/em\u003e\u003csub\u003e\u003cem\u003ef\u003c/em\u003e/4\u003c/sub\u003e, \u003cem\u003et\u003c/em\u003e\u003csub\u003e\u003cem\u003ef\u003c/em\u003e/5\u003c/sub\u003e, and \u003cem\u003et\u003c/em\u003e\u003csub\u003e\u003cem\u003ef\u003c/em\u003e/6\u003c/sub\u003e (Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e)[23].\u003c/p\u003e\n\u003cp\u003eThe time at which the waveform reached its maximum value (\u003cem\u003et\u003c/em\u003e\u003csub\u003e\u003cem\u003emax\u003c/em\u003e\u003c/sub\u003e) was compared to traditional \u003cem\u003et\u003c/em\u003e\u003csub\u003e1\u003c/sub\u003e and \u003cem\u003et\u003c/em\u003e\u003csub\u003e3\u003c/sub\u003e values. Additionally, the time points at which the waveform rose to 80% (\u003cem\u003et\u003c/em\u003e\u003csub\u003e0.8\u003c/sub\u003e) and 90% (\u003cem\u003et\u003c/em\u003e\u003csub\u003e0.9\u003c/sub\u003e) of its maximum amplitude were determined (Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e\n\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\n \u003ch2\u003eData normalization\u003c/h2\u003e\n \u003cp\u003eThe raw data underwent min-max normalization. This transformation preserves the relative order of data points while rescaling values to the range of [0, 1].\u003c/p\u003e\n \u003cp\u003eLet \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{x}_{min}\\)\u003c/span\u003e\u003c/span\u003e denote the minimum value of a specific value in a feature, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{x}_{max}\\)\u003c/span\u003e\u003c/span\u003e denote the maximum value of a specific value in a feature, the conversion function is as follows:\u003c/p\u003e\n \u003cdiv id=\"Equa\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e$$\\:\\begin{array}{c}{x}^{*}=\\frac{x-{x}_{min}}{{x}_{max}-{x}_{min}}\\#\\left(3\\right)\\end{array}$$\u003c/div\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003ch3\u003eData balance processing\u003c/h3\u003e\n\u003cp\u003eSynthetic Minority Over-sampling Technique (SMOTE)[24] is a widely-used technique for addressing imbalanced datasets. By generating new synthetic samples to increase the number of instances in the minority class, SMOTE helps to balance the class distribution. Unlike simple replication, SMOTE employs a linear interpolation method to create new samples, mitigating the risk of overfitting. The synthetic samples will be randomly assigned to both the training and validation sets, while the test set will remain unchanged.\u003c/p\u003e\n\u003ch3\u003eDeep learning classification\u003c/h3\u003e\n\u003cp\u003eML covers various knowledge and technologies such as probability theory, statistics, and complex algorithms[25\u0026ndash;27]. DL is a specific type of ML that uses artificial neural networks with multiple layers to learn complex patterns from large amounts of data. These neural networks are inspired by the structure and function of the human brain. Traditional ML algorithms may be more suitable for smaller datasets or simpler problems. However, as computing power and data availability have increased, DL has become the dominant approach in many areas of ML.\u003c/p\u003e\n\u003cp\u003eThis study used 4 DL algorithms such as Long Short-Term Memory (LSTM)[28], Convolutional Neural Network (CNN), Gated Recurrent Unit (GRU)[29], and Bidirectional Long Short-Term Memory (Bi-LSTM).\u003c/p\u003e\n\u003cp\u003eLSTM is a type of recurrent neural network (RNN) architecture that is specifically designed to learn long-term dependencies in sequential data. Unlike traditional RNNs, LSTMs have a special cell state that can store information for long periods, making them well-suited for tasks.\u003c/p\u003e\n\u003cp\u003eCNN is a type of DL neural network that is primarily used for image recognition, object detection, and other computer vision tasks. CNNs can learn features directly from data, reducing the need for manual feature engineering.\u003c/p\u003e\n\u003cp\u003eGRU is also a type of RNN designed to capture long-term dependencies in sequential data. It\u0026apos;s similar to LSTM but has a simpler architecture, making it computationally more efficient.\u003c/p\u003e\n\u003cp\u003eBi-LSTM is a type of RNN architecture that is particularly well-suited for sequence modeling tasks. Unlike traditional LSTM neural networks, which process sequences in a forward direction only, Bi-LSTM networks process sequences in both forward and backward directions.\u003c/p\u003e\n\u003cp\u003eThe dataset was split into train set, validation set, and test set in the ratio of 6:2:2. A validation set is a portion of a dataset used to assess model performance during training, aiding in hyperparameter tuning and overfitting prevention. The final evaluation was based on the results of the test set.\u003c/p\u003e\n\u003cp\u003eFigure 4 shows the flowchart of the DL classification process.\u0026nbsp;\u003c/p\u003e\n\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\n \u003ch2\u003eClassification metrics\u003c/h2\u003e\n \u003cp\u003eIn this study, the main metrics used for model performance evaluation are Accuracy, Precision, Recall, F1, Specificity, and area under curve (AUC).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\n \u003ch2\u003eConfusion Matrix\u003c/h2\u003e\n \u003cp\u003eThe confusion matrix (shown in Fig.\u0026nbsp;5) is used to summarize the results of a classifier and is a standard format for accuracy evaluation.\u003c/p\u003e\n \u003cp\u003eTP(True Positive): Positive sample predicted by the model as the positive category. The larger the TP value, the better the model.\u003c/p\u003e\n \u003cp\u003eFN(False Negative): Positive sample predicted by the model as the negative category. The smaller the FN value, the better the model.\u003c/p\u003e\n \u003cp\u003eFP(False Positive): Negative sample predicted by the model as the positive category. The smaller the FP value, the better the model.\u003c/p\u003e\n \u003cp\u003eTN(True Negative): Negative sample predicted by the model as the negative category. The larger the TN value, the better the model.\u003c/p\u003e\n \u003cdiv id=\"Equb\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e$$\\:\\begin{array}{c}Precision=\\frac{TP}{TP+FP}\\#\\left(4\\right)\\end{array}$$\u003c/div\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Equc\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equc\" name=\"EquationSource\"\u003e$$\\:\\begin{array}{c}Recall=\\frac{TP}{TP+FN}\\#\\left(5\\right)\\end{array}$$\u003c/div\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Equd\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equd\" name=\"EquationSource\"\u003e$$\\:\\begin{array}{c}Accuracy=\\frac{TP+TN}{P+N}\\#\\left(6\\right)\\end{array}$$\u003c/div\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Eque\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Eque\" name=\"EquationSource\"\u003e$$\\:\\begin{array}{c}{F}_{1}=2\\times\\:\\frac{P\\times\\:R}{P+R}\\#\\left(7\\right)\\end{array}$$\u003c/div\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Equf\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equf\" name=\"EquationSource\"\u003e$$\\:\\begin{array}{c}Specificity=\\frac{TN}{TN+FP}\\#\\left(8\\right)\\end{array}$$\u003c/div\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\n \u003ch2\u003eReceiver Operating Characteristic Curve\u003c/h2\u003e\n \u003cp\u003eThe Receiver Operating Characteristic (ROC) curve is a graphical representation used to evaluate the diagnostic performance of a binary classification model. It plots the TP rate (sensitivity) against the FP rate (1-specificity) at various threshold settings. The area under curve (AUC) quantifies the model\u0026rsquo;s ability to discriminate between positive and negative classes, with values ranging from 0 to 1. An AUC of 0.5 indicates no discrimination (equivalent to random guessing), while an AUC of 1.0 signifies perfect discrimination. The ROC curve serves as a critical tool for assessing the trade-offs between sensitivity and specificity, thereby aiding in the selection of optimal thresholds for clinical decision-making[30, 31].\u003c/p\u003e\n\u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003eThe original data underwent SMOTE oversampling to balance the class distribution (shown in Figure 6).\u003c/p\u003e\n\u003cp\u003eFigure 7 displays the distribution of each feature for both the original data and the synthetic data.\u003c/p\u003e\n\u003cp\u003eFigure 8 visualizes the high-dimensional data generated by t-distributed stochastic neighbor embedding (t-SNE)[32] dimensionality reduction to 2D, allowing for a more intuitive comparison between original and synthetic data.\u003c/p\u003e\n\u003cp\u003eTable 2 shows the performance of four DL models. The final evaluation aimed to check the general ability of models to predict unseen data.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2\u003c/strong\u003e The performance comparison of different DL models\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 16.6667%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.6667%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.6667%;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePrecision\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.5%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRecall\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.375%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eF1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.7083%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSpecificity\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.4167%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAUC\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 16.6667%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLSTM\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.6667%;\"\u003e\n \u003cp\u003e0.85870\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.6667%;\"\u003e\n \u003cp\u003e0.87448\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.5%;\"\u003e\n \u003cp\u003e0.82164\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.375%;\"\u003e\n \u003cp\u003e0.83773\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.7083%;\"\u003e\n \u003cp\u003e0.92369\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.4167%;\"\u003e\n \u003cp\u003e0.93365\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 16.6667%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eGRU\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.6667%;\"\u003e\n \u003cp\u003e0.82609\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.6667%;\"\u003e\n \u003cp\u003e0.81482\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.5%;\"\u003e\n \u003cp\u003e0.78417\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.375%;\"\u003e\n \u003cp\u003e0.79352\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.7083%;\"\u003e\n \u003cp\u003e0.90839\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.4167%;\"\u003e\n \u003cp\u003e0.93125\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 16.6667%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCNN\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.6667%;\"\u003e\n \u003cp\u003e0.81522\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.6667%;\"\u003e\n \u003cp\u003e0.80852\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.5%;\"\u003e\n \u003cp\u003e0.77703\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.375%;\"\u003e\n \u003cp\u003e0.78498\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.7083%;\"\u003e\n \u003cp\u003e0.90255\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.4167%;\"\u003e\n \u003cp\u003e0.90817\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 16.6667%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBi-LSTM\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.6667%;\"\u003e\n \u003cp\u003e0.79348\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.6667%;\"\u003e\n \u003cp\u003e0.78551\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 12.5%;\"\u003e\n \u003cp\u003e0.74671\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 9.375%;\"\u003e\n \u003cp\u003e0.75735\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 17.7083%;\"\u003e\n \u003cp\u003e0.89029\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 10.4167%;\"\u003e\n \u003cp\u003e0.91875\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eFigure 8 shows the confusion matrix for test data.\u003c/p\u003e\n\u003cp\u003eFigure 9 shows the predicted and true value for test data.\u003c/p\u003e\n\u003cp\u003eFigure 10 presents a comparison chart of metrics for different models.\u003c/p\u003e\n\u003cp\u003eFigure 11 presents the ROC and AUC of test data from LSTM model.\u003c/p\u003e\n\u003cp\u003eTable 3 shows the analysis for LSTM trainNetwork usage.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3\u0026nbsp;\u003c/strong\u003eThe analysis for LSTM trainNetwork usage\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.9242%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eName\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.426%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eType\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19.6751%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eActivations\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.1733%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLearnable Properties\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.8014%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eStates\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.9242%;\"\u003e\n \u003cp\u003eSequence input with 68 dimensions\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.426%;\"\u003e\n \u003cp\u003eSequence Input\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19.6751%;\"\u003e\n \u003cp\u003e68(C)\u0026times;1(B)\u0026times;1(T)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.1733%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.8014%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.9242%;\"\u003e\n \u003cp\u003eLSTM with 16 hidden units\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.426%;\"\u003e\n \u003cp\u003eLSTM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19.6751%;\"\u003e\n \u003cp\u003e16(C)\u0026times;1(B)\u0026times;1(T)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.1733%;\"\u003e\n \u003cp\u003eInputWeights 64\u0026times;68\u003c/p\u003e\n \u003cp\u003eRecurrentWeights 64\u0026times;16\u003c/p\u003e\n \u003cp\u003eBias 64\u0026times;1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.8014%;\"\u003e\n \u003cp\u003eHiddenState 16\u0026times;1\u003c/p\u003e\n \u003cp\u003eCellState 16\u0026times;1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.9242%;\"\u003e\n \u003cp\u003e20% dropout\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.426%;\"\u003e\n \u003cp\u003eDropout\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19.6751%;\"\u003e\n \u003cp\u003e16(C)\u0026times;1(B)\u0026times;1(T)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.1733%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.8014%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.9242%;\"\u003e\n \u003cp\u003eLSTM with 16 hidden units\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.426%;\"\u003e\n \u003cp\u003eLSTM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19.6751%;\"\u003e\n \u003cp\u003e16(C)\u0026times;1(B)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.1733%;\"\u003e\n \u003cp\u003eInputWeights 64\u0026times;16\u003c/p\u003e\n \u003cp\u003eRecurrentWeights 64\u0026times;16\u003c/p\u003e\n \u003cp\u003eBias 64\u0026times;1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.8014%;\"\u003e\n \u003cp\u003eHiddenState 16\u0026times;1\u003c/p\u003e\n \u003cp\u003eCellState 16\u0026times;1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.9242%;\"\u003e\n \u003cp\u003e20% dropout\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.426%;\"\u003e\n \u003cp\u003eDropout\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19.6751%;\"\u003e\n \u003cp\u003e16(C)\u0026times;1(B)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.1733%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.8014%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.9242%;\"\u003e\n \u003cp\u003e3 fully connected layer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.426%;\"\u003e\n \u003cp\u003eFully Connected\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19.6751%;\"\u003e\n \u003cp\u003e3(C)\u0026times;1(B)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.1733%;\"\u003e\n \u003cp\u003eWeights 3\u0026times;16\u003c/p\u003e\n \u003cp\u003eBias 3\u0026times;1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.8014%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.9242%;\"\u003e\n \u003cp\u003eSoftmax\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.426%;\"\u003e\n \u003cp\u003eSoftmax\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19.6751%;\"\u003e\n \u003cp\u003e3(C)\u0026times;1(B)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.1733%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.8014%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 22.9242%;\"\u003e\n \u003cp\u003eClassoutput\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 16.426%;\"\u003e\n \u003cp\u003eClassification Output\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 19.6751%;\"\u003e\n \u003cp\u003e3(C)\u0026times;1(B)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 26.1733%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 14.8014%;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e"},{"header":"Conclusion and Discussion","content":"\u003cp\u003eIn this study, we employed four DL algorithms to investigate the radial artery pulse wave characteristics of healthy, CAD, and HF participants. Table 2 demonstrates that LSTM achieved the highest classification performance, with an accuracy of 0.8587, precision of 0.87448, recall of 0.82164, F1-score of 0.83773, specificity of 0.92369, and AUC of 0.93365. Given its superior performance across all metrics, LSTM emerges as the preferred DL model for this study.\u003c/p\u003e\n\u003cp\u003eAn early diagnosis of HF may reduce patients\u0026rsquo; mortality and morbidity. Consequently, wide efforts have been put to develop algorithms in diagnosing [33].\u003csup\u003e\u0026nbsp;\u003c/sup\u003eHF diagnosis may be achieved through the analysis of electrocardiography, magnetic resonance imaging, ultrasound images as well as electronic health records [34-36].\u0026nbsp;These methods, however, are limited due to the requirement for specialized equipment and skilled operators. The ability to identify HF using only radial artery pulse wave could be used for early diagnosis and would enable referral for further investigation, offering a noninvasive, cost-effective solution. Wu et al. evaluated the potential of wrist pulse signals for use in the cardiac monitoring of patients with CHD. Decision tree (DT) and random forest (RF) algorithms were employed to construct models for classifying coronary heart disease (CHD) patients with different brain natriuretic peptide. The pulse features of the three groups differed significantly, suggesting different pathological states of the cardiovascular system in patients with CHD. Moreover, the RF models outperformed the DT models in performance metrics[37]. Wu et al.\u0026nbsp;proposed a ML-based strategy to identify left ventricular enlargement (LVE) in HF patients by means of pulse wave signals by means of classification and regression models.\u0026nbsp;The classification model exhibited superior performance with an accuracy of 0.91 and an AUC-ROC of 0.93. The regression model achieved an accuracy of 0.88 and an AUC-ROC of 0.89, indicating that both models can quickly and accurately identify LVE in HF patients[38]. In 2023, Yan et al. utilized RF, Support Vector Machine, k-nearest neighbor, and DT algorithms to develop classification models using microcirculatory characteristic parameter set. The results showed that RF showed good classification performance, the identification accuracy of the model built on the microcirculatory characteristic parameter set and RF algorithm all reached more than 88%. The highest recognition accuracy was 95.51% for coronary heart disease samples, 92.11% for healthy samples, and 88.55% for hypertensive samples. These researches demonstrated radial artery pulse waves as a key informational conduit within the cardiovascular system and ML as a more suitable method to deal with complex wrist pulse data.\u003c/p\u003e\n\u003cp\u003eWhile previous studies have primarily relied on traditional ML techniques such as DT and RF, this study introduces a novel approach that incorporates DL algorithms. In addition to the standard features used in previous research, this study further leverages moving average features and the time points at which the radial artery pulse wave reaches 80% and 90% of its peak amplitude. These additional features provide a more comprehensive and nuanced representation of the pulse wave morphology, particularly in cases of HF where the rapid ejection phase and fusion of the main wave and pre-beat wave can be subtle and difficult to discern. By employing DL algorithms, this study achieves superior performance in terms of accuracy, recall, and AUC, while also mitigating overfitting and enhancing the scalability of the model. At the same time, this study is also beneficial for the modern inheritance and international development of TCM.\u003c/p\u003e\n\u003cp\u003eThis study elucidates the correlation between pulse diagnosis and HF. Given the high prevalence and severe implications of HF, this DL-based model offers a systematic method to \u0026quot;learn\u0026quot; the association between pulse wave data, CAD, and HF. While this learning approach may not fully elucidate the underlying biological mechanisms, it has the potential to significantly aid in the early diagnosis of HF in clinical settings. This approach is advantageous due to its simplicity, non-invasive nature, and low cost.\u003c/p\u003e\n\u003cp\u003eWearable devices are portable medical or health electronic devices that incorporate sensors, wireless communication, and multimedia technologies into items worn directly on the body. Supported by software, these devices can sense, record, analyze, regulate, intervene in, and even treat diseases or maintain health conditions. Leveraging human or environmental capabilities, they enable intelligent information interaction through built-in sensors and integrated chips. Wearable devices serve as the ideal platform for the ongoing integration and innovation of Internet of Things, mobile internet, cloud storage, and big data technologies. This study may serve as a foundation for developing wearable devices designed to detect cardiovascular abnormalities. Furthermore, this approach can be extended to a wider range of diagnostic applications in the future.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSeveral potential obstacles may hinder our progress: (1) To date, our dataset comprises approximately 1500 samples from CAD and HF patients. While this provides a valuable foundation for our study, a larger dataset is required to optimize deep learning models and achieve superior performance in terms of accuracy, recall, and AUC. Consequently, we plan to expand our dataset in future research to enhance model robustness and generalization. (2) To date, our dataset comprises approximately 1800 samples from CAD and HF patients. While this provides a valuable foundation for our study, a larger dataset is required to optimize deep learning models and achieve superior performance in terms of accuracy, recall, and AUC. Consequently, we plan to expand our dataset in future research to enhance model robustness and generalization. (3) The current sampling frequency is set at 720 Hz. To enhance the precision of our measurements, future studies will explore the feasibility of increasing the sampling rate of the device.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eCAD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eCoronary artery disease\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eHF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eHeart failure\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eTCM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eTraditional Chinese Medicine\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eDL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eDeep learning\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eML\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eMachine learning\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eNYHA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eNew York Heart Association\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eSMOTE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eSynthetic Minority Over-sampling Technique\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eLSTM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eLong Short-Term Memory\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eCNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eConvolutional Neural Network\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eGRU\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eGated Recurrent Unit\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eBi-LSTM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eBidirectional Long Short-Term Memory\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eAUC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eArea under curve\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eTP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eTrue Positive\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eFN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eFalse Negative\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eFP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eFalse Positive\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eTN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eTrue Negative\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eROC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eReceiver Operating Characteristic\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003et-SNE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003et-distributed stochastic neighbor embedding\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eDT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eDecision Tree\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eRandom Forest\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eCHD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eCoronary heart disease\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 23.8698%;\"\u003e\n \u003cp\u003eLVE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 76.1302%;\"\u003e\n \u003cp\u003eLeft ventricular enlargement\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAuthors\u0026rsquo; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eY. L and W.-Y. H provided the research conceptualization and provided the research methods, contributed equally to this work. Y. L, W.-Y. H, and J. H acquired the data. Y. L and H.-M. W analyzed the data and performed deep learning. Y.-Q. W, H.-X. Y, and J. X in charge of funding acquisition. All authors edited and revised the manuscript and approved the final version.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study protocol was approved by the Institutional Review Board of Shanghai University of Traditional Chinese Medicine (Approval Number:2023-3-10-08-08) and the study complied with the Declaration of Helsinki. All participants received informed consent form and signed. Clinical trial number: not applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets generated and/or analyzed during the current study are not publicly available due to ethical concern but are available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work is supported by the National Natural Science Foundation of China (NSFC, Grant No.81673880), Shanghai Key Laboratory of Health Identification and Assessment Project, China (Grant No.21DZ2271000).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003e\u003cstrong\u003eGlobal, regional, and national age-sex specific all-cause and cause-specific mortality for 240 causes of death, 1990-2013: a systematic analysis for the Global Burden of Disease Study 2013\u003c/strong\u003e. \u003cem\u003eLancet (London, England)\u0026nbsp;\u003c/em\u003e2015, \u003cstrong\u003e385\u003c/strong\u003e(9963):117-171.\u003c/li\u003e\n \u003cli\u003eMastoi QU, Wah TY, Gopal Raj R, Iqbal U: \u003cstrong\u003eAutomated Diagnosis of Coronary Artery Disease: A Review and Workflow\u003c/strong\u003e. \u003cem\u003eCardiology research and practice\u0026nbsp;\u003c/em\u003e2018, \u003cstrong\u003e2018\u003c/strong\u003e:2016282.\u003c/li\u003e\n \u003cli\u003eHatfield J, Woods MD, Hammonds K, Wahab L, Nguyen V, Hassan OA, Butler J, Widmer RJ: \u003cstrong\u003eGlobal hypokinesis in resting transthoracic echocardiography diagnosis of heart failure and coronary artery disease\u003c/strong\u003e. \u003cem\u003eThe Journal of invasive cardiology\u0026nbsp;\u003c/em\u003e2024.\u003c/li\u003e\n \u003cli\u003eMcDonagh TA, Metra M, Adamo M, Gardner RS, Baumbach A, B\u0026ouml;hm M, Burri H, Butler J, Čelutkienė J, Chioncel O\u003cem\u003e\u0026nbsp;et al\u003c/em\u003e: \u003cstrong\u003e2023 Focused Update of the 2021 ESC Guidelines for the diagnosis and treatment of acute and chronic heart failure: Developed by the task force for the diagnosis and treatment of acute and chronic heart failure of the European Society of Cardiology (ESC) With the special contribution of the Heart Failure Association (HFA) of the ESC\u003c/strong\u003e. \u003cem\u003eEuropean journal of heart failure\u0026nbsp;\u003c/em\u003e2024, \u003cstrong\u003e26\u003c/strong\u003e(1):5-17.\u003c/li\u003e\n \u003cli\u003eMartin SS, Aday AW, Almarzooq ZI, Anderson CAM, Arora P, Avery CL, Baker-Smith CM, Barone Gibbs B, Beaton AZ, Boehme AK\u003cem\u003e\u0026nbsp;et al\u003c/em\u003e: \u003cstrong\u003e2024 Heart Disease and Stroke Statistics: A Report of US and Global Data From the American Heart Association\u003c/strong\u003e. \u003cem\u003eCirculation\u0026nbsp;\u003c/em\u003e2024, \u003cstrong\u003e149\u003c/strong\u003e(8):e347-e913.\u003c/li\u003e\n \u003cli\u003eRoger VL: \u003cstrong\u003eEpidemiology of Heart Failure: A Contemporary Perspective\u003c/strong\u003e. \u003cem\u003eCirculation research\u0026nbsp;\u003c/em\u003e2021, \u003cstrong\u003e128\u003c/strong\u003e(10):1421-1434.\u003c/li\u003e\n \u003cli\u003eConrad N, Judge A, Canoy D, Tran J, Pinho-Gomes AC, Millett ERC, Salimi-Khorshidi G, Cleland JG, McMurray JJV, Rahimi K: \u003cstrong\u003eTemporal Trends and Patterns in Mortality After Incident Heart Failure: A Longitudinal Analysis of 86 000 Individuals\u003c/strong\u003e. \u003cem\u003eJAMA cardiology\u0026nbsp;\u003c/em\u003e2019, \u003cstrong\u003e4\u003c/strong\u003e(11):1102-1111.\u003c/li\u003e\n \u003cli\u003eTriposkiadis F, Xanthopoulos A, Parissis J, Butler J, Farmakis D: \u003cstrong\u003ePathogenesis of chronic heart failure: cardiovascular aging, risk factors, comorbidities, and disease modifiers\u003c/strong\u003e. \u003cem\u003eHeart failure reviews\u0026nbsp;\u003c/em\u003e2022, \u003cstrong\u003e27\u003c/strong\u003e(1):337-344.\u003c/li\u003e\n \u003cli\u003eTomasoni D, Adamo M, Lombardi CM, Metra M: \u003cstrong\u003eHighlights in heart failure\u003c/strong\u003e. \u003cem\u003eESC heart failure\u0026nbsp;\u003c/em\u003e2019, \u003cstrong\u003e6\u003c/strong\u003e(6):1105-1127.\u003c/li\u003e\n \u003cli\u003eZhang J, Liao J, Wang T, Yuan R, Zhao Y, Han Z, Tang L, Zhao L: \u003cstrong\u003eEffects of joy and sorrow on pulse-graph parameters in healthy female college students based on emotion-evoked experiments\u003c/strong\u003e. \u003cem\u003eEXPLORE\u0026nbsp;\u003c/em\u003e2020, \u003cstrong\u003e17\u003c/strong\u003e(4):303-311.\u003c/li\u003e\n \u003cli\u003eWang W, Zeng W, Chen X, Tu L, Xu J, Yin X: \u003cstrong\u003eParameter study on characteristic pulse diagram of polycystic ovary syndrome based on logistic regression analysis\u003c/strong\u003e. \u003cem\u003eJournal of Obstetrics and Gynaecology\u0026nbsp;\u003c/em\u003e2022, \u003cstrong\u003e42\u003c/strong\u003e(8):3712-3719.\u003c/li\u003e\n \u003cli\u003eLim J, Li J, Feng X, Feng L, Xia Y, Xiao X, Wang Y, Xu Z: \u003cstrong\u003eMachine learning classification of polycystic ovary syndrome based on radial pulse wave analysis\u003c/strong\u003e. \u003cem\u003eBMC Complement Med Ther\u0026nbsp;\u003c/em\u003e2023, \u003cstrong\u003e23\u003c/strong\u003e(1).\u003c/li\u003e\n \u003cli\u003eWan WK, Hsu TL, Chang HC, Wan YY: \u003cstrong\u003eEffect of acupuncture at Hsien-Ku (St-43) on the pulse spectrum and a discussion of the evidence for the frequency structure of Chinese medicine\u003c/strong\u003e. \u003cem\u003eThe American journal of Chinese medicine\u0026nbsp;\u003c/em\u003e2000, \u003cstrong\u003e28\u003c/strong\u003e(1):41-55.\u003c/li\u003e\n \u003cli\u003eLim J, Li J, Feng X, Feng L, Xiao X, Xia Y, Wang Y, Qian L, Yang H, Xu Z: \u003cstrong\u003eMachine learning-based evaluation of application value of traditional Chinese medicine clinical index and pulse wave parameters in the diagnosis of polycystic ovary syndrome\u003c/strong\u003e. \u003cem\u003eEuropean Journal of Integrative Medicine\u0026nbsp;\u003c/em\u003e2023, \u003cstrong\u003e64\u003c/strong\u003e:102311.\u003c/li\u003e\n \u003cli\u003eBaik Y: \u003cstrong\u003eA study on The Characteristic of Traditional Pediatric Pulse Diagnosis\u003c/strong\u003e. \u003cem\u003eJournal of Korean Medical classics\u0026nbsp;\u003c/em\u003e2014, \u003cstrong\u003e27\u003c/strong\u003e(1):111-122.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eNomenclature and criteria for diagnosis of ischemic heart disease. Report of the Joint International Society and Federation of Cardiology/World Health Organization task force on standardization of clinical nomenclature\u003c/strong\u003e. \u003cem\u003eCirculation\u0026nbsp;\u003c/em\u003e1979, \u003cstrong\u003e59\u003c/strong\u003e(3):607-609.\u003c/li\u003e\n \u003cli\u003eChinese Society of Cardiology CMA, Chinese College of Cardiovascular Physician, Chinese Heart Failure Association of Chinese Medical Doctor Association, Editorial Board of Chinese Journal of Cardiology: \u003cstrong\u003eChinese guidelines for the diagnosis and treatment of heart failure 2024\u003c/strong\u003e. \u003cem\u003eChinese Journal of Cardiology\u0026nbsp;\u003c/em\u003e2024, \u003cstrong\u003e52\u003c/strong\u003e(3):235-275.\u003c/li\u003e\n \u003cli\u003eWhite PD, Myers MJJ: \u003cstrong\u003eThe classification of cardiac diagnosis\u003c/strong\u003e. 1921, \u003cstrong\u003e77\u003c/strong\u003e(18):1414-1415.\u003c/li\u003e\n \u003cli\u003eDing X, Wang Y, Hao Y, Lv Y, Chen R, Yan H: \u003cstrong\u003eA New Measure of Pulse Rate Variability and Detection of Atrial Fibrillation Based on Improved Time Synchronous Averaging\u003c/strong\u003e. \u003cem\u003eComputational and mathematical methods in medicine\u0026nbsp;\u003c/em\u003e2021, \u003cstrong\u003e2021\u003c/strong\u003e:5597559.\u003c/li\u003e\n \u003cli\u003eNichols WW ORM, Vlachopoulos C, et al.: \u003cstrong\u003eMcDonald\u0026apos;s Blood Flow in Arteries: Theoretical, Experimental and Clinical Principles\u003c/strong\u003e. London: Hodder Arnold; 2011.\u003c/li\u003e\n \u003cli\u003eNichols WW: \u003cstrong\u003eClinical measurement of arterial stiffness obtained from noninvasive pressure waveforms\u003c/strong\u003e. \u003cem\u003eAmerican journal of hypertension\u0026nbsp;\u003c/em\u003e2005, \u003cstrong\u003e18\u003c/strong\u003e(1 Pt 2):3s-10s.\u003c/li\u003e\n \u003cli\u003eXiao H, Butlin M, Qasem A, Tan I, Li D, Avolio AP: \u003cstrong\u003eN-Point Moving Average: A Special Generalized Transfer Function Method for Estimation of Central Aortic Blood Pressure\u003c/strong\u003e. \u003cem\u003eIEEE transactions on bio-medical engineering\u0026nbsp;\u003c/em\u003e2018, \u003cstrong\u003e65\u003c/strong\u003e(6):1226-1234.\u003c/li\u003e\n \u003cli\u003eQuail MA, Steeden JA, Knight D, Segers P, Taylor AM, Muthurangu V: \u003cstrong\u003eDevelopment and validation of a novel method to derive central aortic systolic pressure from the MR aortic distension curve\u003c/strong\u003e. \u003cem\u003eJournal of magnetic resonance imaging : JMRI\u0026nbsp;\u003c/em\u003e2014, \u003cstrong\u003e40\u003c/strong\u003e(5):1064-1070.\u003c/li\u003e\n \u003cli\u003eChawla N, Bowyer K, Hall L, Kegelmeyer W: \u003cstrong\u003eSMOTE: Synthetic Minority Over-sampling Technique\u003c/strong\u003e. \u003cem\u003ejair\u0026nbsp;\u003c/em\u003e2002, \u003cstrong\u003eabs/1106.1813\u003c/strong\u003e:321-357.\u003c/li\u003e\n \u003cli\u003eChoi RY, Coyner AS, Kalpathy-Cramer J, Chiang MF, Campbell JP: \u003cstrong\u003eIntroduction to Machine Learning, Neural Networks, and Deep Learning\u003c/strong\u003e. \u003cem\u003eTranslational vision science \u0026amp; technology\u0026nbsp;\u003c/em\u003e2020, \u003cstrong\u003e9\u003c/strong\u003e(2):14.\u003c/li\u003e\n \u003cli\u003eDeo RC: \u003cstrong\u003eMachine Learning in Medicine\u003c/strong\u003e. \u003cem\u003eCirculation\u0026nbsp;\u003c/em\u003e2015, \u003cstrong\u003e132\u003c/strong\u003e(20):1920-1930.\u003c/li\u003e\n \u003cli\u003eSun X, Yin Y, Yang Q, Huo T: \u003cstrong\u003eArtificial intelligence in cardiovascular diseases: diagnostic and therapeutic perspectives\u003c/strong\u003e. \u003cem\u003eEuropean journal of medical research\u0026nbsp;\u003c/em\u003e2023, \u003cstrong\u003e28\u003c/strong\u003e(1):242.\u003c/li\u003e\n \u003cli\u003eHochreiter SJNCM-P: \u003cstrong\u003eLong Short-term Memory\u003c/strong\u003e. 1997.\u003c/li\u003e\n \u003cli\u003eCho KJapa: \u003cstrong\u003eLearning phrase representations using RNN encoder-decoder for statistical machine translation\u003c/strong\u003e. 2014.\u003c/li\u003e\n \u003cli\u003eJames G WD, Hastie T, Tibshirani R: \u003cstrong\u003eAn Introduction to Statistical Learning: With Applications in R\u003c/strong\u003e. New York: Springer; 2013.\u003c/li\u003e\n \u003cli\u003eHastie T TR, Friedman JH: \u003cstrong\u003eThe Elements of Statistical Learning: Data Mining, Inference, and Prediction. 2nd ed\u003c/strong\u003e. New York: Springer; 2009.\u003c/li\u003e\n \u003cli\u003eMaaten Lvd, Hinton GEJJoMLR: \u003cstrong\u003eVisualizing Data using t-SNE\u003c/strong\u003e. 2008, \u003cstrong\u003e9\u003c/strong\u003e:2579-2605.\u003c/li\u003e\n \u003cli\u003ePenso M, Solbiati S, Moccia S, Caiani EG: \u003cstrong\u003eDecision Support Systems in HF based on Deep Learning Technologies\u003c/strong\u003e. \u003cem\u003eCurrent heart failure reports\u0026nbsp;\u003c/em\u003e2022, \u003cstrong\u003e19\u003c/strong\u003e(2):38-51.\u003c/li\u003e\n \u003cli\u003eKwon JM, Kim KH, Jeon KH, Kim HM, Kim MJ, Lim SM, Song PS, Park J, Choi RK, Oh BH: \u003cstrong\u003eDevelopment and Validation of Deep-Learning Algorithm for Electrocardiography-Based Heart Failure Identification\u003c/strong\u003e. \u003cem\u003eKorean circulation journal\u0026nbsp;\u003c/em\u003e2019, \u003cstrong\u003e49\u003c/strong\u003e(7):629-639.\u003c/li\u003e\n \u003cli\u003eLau ES, Di Achille P, Kopparapu K, Andrews CT, Singh P, Reeder C, Al-Alusi M, Khurshid S, Haimovich JS, Ellinor PT\u003cem\u003e\u0026nbsp;et al\u003c/em\u003e: \u003cstrong\u003eDeep Learning-Enabled Assessment of Left Heart Structure and Function Predicts Cardiovascular Outcomes\u003c/strong\u003e. \u003cem\u003eJournal of the American College of Cardiology\u0026nbsp;\u003c/em\u003e2023, \u003cstrong\u003e82\u003c/strong\u003e(20):1936-1948.\u003c/li\u003e\n \u003cli\u003eMcGilvray MMO, Heaton J, Guo A, Masood MF, Cupps BP, Damiano M, Pasque MK, Foraker R: \u003cstrong\u003eElectronic Health Record-Based Deep Learning Prediction of Death or Severe Decompensation in Heart Failure Patients\u003c/strong\u003e. \u003cem\u003eJACC Heart failure\u0026nbsp;\u003c/em\u003e2022, \u003cstrong\u003e10\u003c/strong\u003e(9):637-647.\u003c/li\u003e\n \u003cli\u003eWu WJ, Chen R, Guo R, Yan JJ, Zhang CK, Wang YQ, Yan HX, Zhang YQ: \u003cstrong\u003eA novel method for assessing cardiac function in patients with coronary heart disease based on wrist pulse analysis\u003c/strong\u003e. \u003cem\u003eIrish journal of medical science\u0026nbsp;\u003c/em\u003e2023, \u003cstrong\u003e192\u003c/strong\u003e(6):2697-2706.\u003c/li\u003e\n \u003cli\u003eWu D, Ono R, Wang S, Kobayashi Y, Sughimoto K, Liu H: \u003cstrong\u003ePulse wave signal-driven machine learning for identifying left ventricular enlargement in heart failure patients\u003c/strong\u003e. \u003cem\u003eBiomedical engineering online\u0026nbsp;\u003c/em\u003e2024, \u003cstrong\u003e23\u003c/strong\u003e(1):60.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-informatics-and-decision-making","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"midm","sideBox":"Learn more about [BMC Medical Informatics and Decision Making](http://bmcmedinformdecismak.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/midm/default.aspx","title":"BMC Medical Informatics and Decision Making","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Heart failure, Radial artery pulse wave, Long Short-Term Memory, Deep learning","lastPublishedDoi":"10.21203/rs.3.rs-5442852/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5442852/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground\u003c/strong\u003e The pressing global health issue of heart failure (HF) demands innovative approaches for early detection. Non-invasive, rapid, and cost-effective deep learning (DL)-based techniques offer a promising avenue for addressing this challenge.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods\u003c/strong\u003e A total of 462 participants were categorized into three groups: healthy, coronary artery disease (CAD), and heart failure (HF). Raw radial artery pulse wave data were collected from each participant, followed by preprocessing steps including denoising, normalization, and balancing. Subsequently, four deep learning (DL) algorithms were applied to the processed data: Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), Gated Recurrent Unit (GRU), and Bidirectional Long Short-Term Memory (Bi-LSTM).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults\u003c/strong\u003e LSTM achieved the highest classification performance, with an accuracy of 0.8587, precision of 0.87448, recall of 0.82164, F1-score of 0.83773, specificity of 0.92369, and AUC of 0.93365. Given its superior performance across all metrics, LSTM emerges as the preferred DL model for this study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion\u003c/strong\u003e By employing LSTM to analyze radial artery pulse wave, we can accurately distinguish between healthy individuals, patients with CAD, and those with HF. This simple, non-invasive, and cost-effective method presents a potential strategy for early detection of HF.\u003c/p\u003e","manuscriptTitle":"A Heart Failure Classification Model from Radial Artery Pulse Wave Using LSTM Neural Networks","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-12-18 16:24:17","doi":"10.21203/rs.3.rs-5442852/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-11-25T04:05:36+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-11-25T03:54:57+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-11-16T08:10:14+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Medical Informatics and Decision Making","date":"2024-11-13T01:48:09+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-medical-informatics-and-decision-making","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"midm","sideBox":"Learn more about [BMC Medical Informatics and Decision Making](http://bmcmedinformdecismak.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/midm/default.aspx","title":"BMC Medical Informatics and Decision Making","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"92a5f2c3-1c9e-4f7d-8674-8a7c48d81c65","owner":[],"postedDate":"December 18th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-09-01T16:06:59+00:00","versionOfRecord":{"articleIdentity":"rs-5442852","link":"https://doi.org/10.1186/s12911-025-03167-5","journal":{"identity":"bmc-medical-informatics-and-decision-making","isVorOnly":false,"title":"BMC Medical Informatics and Decision Making"},"publishedOn":"2025-08-27 15:56:52","publishedOnDateReadable":"August 27th, 2025"},"versionCreatedAt":"2024-12-18 16:24:17","video":"","vorDoi":"10.1186/s12911-025-03167-5","vorDoiUrl":"https://doi.org/10.1186/s12911-025-03167-5","workflowStages":[]},"version":"v1","identity":"rs-5442852","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5442852","identity":"rs-5442852","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}