《计算机应用》唯一官方网站 ›› 2026, Vol. 46 ›› Issue (9): 3043-3053.DOI: 10.11772/j.issn.1001-9081.2025070894
• 前沿与综合应用 • 上一篇
收稿日期:2025-08-06
修回日期:2025-09-21
接受日期:2025-09-22
发布日期:2025-11-05
出版日期:2026-09-10
通讯作者:
汪祖民
作者简介:季长清(1980—),男,辽宁丹东人,教授,博士,CCF会员,主要研究方向:智慧医疗、低空经济、数字孪生基金资助:
Changqing JI1,2, Ziying ZHANG2, Zumin WANG2(
)
Received:2025-08-06
Revised:2025-09-21
Accepted:2025-09-22
Online:2025-11-05
Published:2026-09-10
Contact:
Zumin WANG
About author:JI Changqing, born in 1980, Ph. D., professor. His research interests include smart healthcare, low-altitude economy, digital twin.Supported by:摘要:
高血压是心血管疾病(CVD)的重要诱因,持续的血压监测对CVD的预防至关重要。针对现有模型在多尺度特征建模能力、长期依赖的建模能力和任务间的信息交互等方面的问题,提出一种名为增强图神经网络的多尺度Transformer (MTEG)的数据驱动血压预测模型,通过融合多尺度卷积、改进Transformer编码器、通道注意力机制和图注意力多任务学习策略,实现光电容积脉搏波(PPG)信号的局部与全局特征联合建模。具体地,通过多尺度特征提取模块捕获PPG信号中不同时间尺度的局部特征,采用基于旋转位置编码(RoPE)和相对位置偏置的Transformer编码器建模血压变化的长期依赖关系,利用特征感知增强模块强化对血压变化相关通道的响应,并构建多任务特征聚合模块实现收缩压(SBP)、舒张压(DBP)和平均动脉压(MAP)这3个预测任务间的信息协同。实验结果表明,MTEG在SBP、DBP和MAP预测中的平均绝对误差(MAE)分别为4.92、2.68和2.59 mmHg,DBP和MAP预测符合美国医疗器械促进协会(AAMI)标准,并达到英国高血压学会(BHS)标准的A级,SBP预测达到BHS标准的B级,验证了所提模型的可靠性与临床应用潜力。
中图分类号:
季长清, 张紫滢, 汪祖民. 基于增强图神经网络的多尺度Transformer的连续无创血压预测方法[J]. 计算机应用, 2026, 46(9): 3043-3053.
Changqing JI, Ziying ZHANG, Zumin WANG. Continuous non-invasive blood pressure prediction method based on multi-scale Transformer with enhanced graph neural network[J]. Journal of Computer Applications, 2026, 46(9): 3043-3053.
阈值范围/ mmHg | SBP和DBP的预测误差在不同阈值范围内的累计比例/% | ||
|---|---|---|---|
| A级 | B级 | C级 | |
| [0,5] | 60 | 50 | 40 |
| [0,10] | 85 | 75 | 65 |
| [0,15] | 95 | 90 | 85 |
表1 BHS标准
Tab. 1 BHS standard
阈值范围/ mmHg | SBP和DBP的预测误差在不同阈值范围内的累计比例/% | ||
|---|---|---|---|
| A级 | B级 | C级 | |
| [0,5] | 60 | 50 | 40 |
| [0,10] | 85 | 75 | 65 |
| [0,15] | 95 | 90 | 85 |
| 阈值范围/mmHg | 不同血压参数在不同阈值范围内的累计误差比例/% | ||
|---|---|---|---|
| SBP | DBP | MAP | |
| 等级 | B | A | A |
| [0,5] | 70.96 | 85.67 | 85.92 |
| [0,10] | 86.94 | 95.59 | 95.01 |
| [0,15] | 92.86 | 98.38 | 98.11 |
表2 基于BHS标准的评估
Tab. 2 Evaluation based on BHS standard
| 阈值范围/mmHg | 不同血压参数在不同阈值范围内的累计误差比例/% | ||
|---|---|---|---|
| SBP | DBP | MAP | |
| 等级 | B | A | A |
| [0,5] | 70.96 | 85.67 | 85.92 |
| [0,10] | 86.94 | 95.59 | 95.01 |
| [0,15] | 92.86 | 98.38 | 98.11 |
| 血压参数 | ME/mmHg | SD/mmHg | 受试者数 |
|---|---|---|---|
| SBP | -1.19 | 8.42 | 278 |
| DBP | 0.54 | 4.53 | 278 |
| MAP | -0.23 | 4.77 | 278 |
表3 基于AAMI标准的评估
Tab. 3 Evaluation based on AAMI standard
| 血压参数 | ME/mmHg | SD/mmHg | 受试者数 |
|---|---|---|---|
| SBP | -1.19 | 8.42 | 278 |
| DBP | 0.54 | 4.53 | 278 |
| MAP | -0.23 | 4.77 | 278 |
| 模型 | MAE/mmHg | SD/mmHg | ||||
|---|---|---|---|---|---|---|
| SBP | DBP | MAP | SBP | DBP | MAP | |
| RDAE[ | 7.95 | 4.11 | 3.83 | 10.37 | 5.50 | 5.13 |
| RSABPM[ | 14.39 | 6.57 | — | 17.87 | 8.43 | — |
| GA-BP[ | 5.16 | 2.89 | 3.47 | 8.50 | 4.77 | 4.78 |
| PPG2ABP[ | 5.73 | 3.45 | 2.31 | 10.69 | 6.86 | 4.96 |
| MSA-ResNet[ | 5.98 | 3.24 | — | 9.79 | 4.94 | — |
| MTEG | 4.92 | 2.68 | 2.59 | 8.42 | 4.53 | 4.77 |
表4 MIMIC数据集上的实验结果
Tab. 4 Experimental results on MIMIC dataset
| 模型 | MAE/mmHg | SD/mmHg | ||||
|---|---|---|---|---|---|---|
| SBP | DBP | MAP | SBP | DBP | MAP | |
| RDAE[ | 7.95 | 4.11 | 3.83 | 10.37 | 5.50 | 5.13 |
| RSABPM[ | 14.39 | 6.57 | — | 17.87 | 8.43 | — |
| GA-BP[ | 5.16 | 2.89 | 3.47 | 8.50 | 4.77 | 4.78 |
| PPG2ABP[ | 5.73 | 3.45 | 2.31 | 10.69 | 6.86 | 4.96 |
| MSA-ResNet[ | 5.98 | 3.24 | — | 9.79 | 4.94 | — |
| MTEG | 4.92 | 2.68 | 2.59 | 8.42 | 4.53 | 4.77 |
| 模型 | MAE/mmHg | SD/mmHg | ||||
|---|---|---|---|---|---|---|
| SBP | DBP | MAP | SBP | DBP | MAP | |
| TEG | 5.11 | 2.85 | 3.00 | 9.15 | 4.75 | 5.16 |
| MEG | 5.00 | 3.02 | 2.91 | 8.50 | 4.63 | 4.80 |
| MTE | 4.98 | 2.77 | 2.72 | 8.46 | 4.60 | 4.52 |
| MTEG | 4.92 | 2.68 | 2.59 | 8.42 | 4.53 | 4.77 |
表5 MIMIC数据集上的消融实验结果
Tab. 5 Ablation experimental results on MIMIC dataset
| 模型 | MAE/mmHg | SD/mmHg | ||||
|---|---|---|---|---|---|---|
| SBP | DBP | MAP | SBP | DBP | MAP | |
| TEG | 5.11 | 2.85 | 3.00 | 9.15 | 4.75 | 5.16 |
| MEG | 5.00 | 3.02 | 2.91 | 8.50 | 4.63 | 4.80 |
| MTE | 4.98 | 2.77 | 2.72 | 8.46 | 4.60 | 4.52 |
| MTEG | 4.92 | 2.68 | 2.59 | 8.42 | 4.53 | 4.77 |
| [1] | 杨诗韵,牛静菘,车欣然,等.心血管疾病的液体活检研究进展[J].自然杂志, 2025, 47(3): 196-206. |
| Yang Shiyun, Niu Jingsong, Che Xinran, et al. Research advances in liquid biopsy for cardiovascular diseases [J]. Chinese Journal of Nature, 2025, 47(3): 196-206. | |
| [2] | 吴燕,吴庞,何攀,等.一款无创连续血压测量设备研制及其准确性验证[J].中国医学装备, 2024, 21(3): 1-7. |
| Wu Yan, Wu Pang, He Pan, et al. Development and accuracy verification of a continuous non-invasive blood pressure measurement device [J]. China Medical Equipment, 2024, 21(3): 1-7. | |
| [3] | El-Hajj C, Kyriacou P A. A review of machine learning techniques in photoplethysmography for the non-invasive cuff-less measurement of blood pressure [J]. Biomedical Signal Processing and Control, 2020, 58: No.101870. |
| [4] | 叶青,章祎枫,沙金亮,等.基于光电容积脉搏波的无创血压连续测量研究进展[J].科学技术与工程, 2024, 24(5): 1756-1774. |
| Ye Qing, Zhang Yifeng, Sha Jinliang, et al. Research progress of non-invasive blood pressure continuous measurement based on photo plethysmo graphy [J]. Science Technology and Engineering, 2024, 24(5): 1756-1774. | |
| [5] | 王梦婷.基于光电容积脉搏波的多参数连续无创血压算法研究[D].杭州:浙江大学, 2017: 26-28. |
| Wang Mengting. Multi-parameter non-invasive continuous blood pressure algorithm research based on photoplethysmography [D]. Hangzhou: Zhejiang University, 2017: 26-28. | |
| [6] | Mahmud S, Ibtehaz N, Khandakar A, et al. A shallow U-Net architecture for reliably predicting Blood Pressure (BP) from PhotoPlethysmoGram (PPG) and ElectroCardioGram (ECG) signals [J]. Sensors, 2022, 22(3): No.919. |
| [7] | Goldberger A L, Amaral L A N, Glass L, et al. PhysioBank, PhysioToolkit, and PhysioNet: components of a new research resource for complex physiologic signals [J]. Circulation, 2000, 101(23): e215-e220. |
| [8] | O'Brien E, Petrie J, Littler W, et al. The British Hypertension Society protocol for the evaluation of blood pressure measuring devices [J]. Journal of Hypertension, 1993, 11(S2): S43-S62. |
| [9] | Thambiraj G, Gandhi U, Mangalanathan U, et al. Investigation on the effect of Womersley number, ECG and PPG features for cuff less blood pressure estimation using machine learning [J]. Biomedical Signal Processing and Control, 2020, 60: No.101942. |
| [10] | 黄晓祥.血压与多生理参数的关联性及无创连续血压测量模型的研究[D].济南:山东大学, 2021: 37-45. |
| Huang Xiaoxiang. The research of the correlation between BP and multiple physiological parameters and non-invasive continuous BP measurement models [D]. Jinan: Shandong University, 2021: 37-45. | |
| [11] | Alpert B S. Validation of the Welch Allyn Spot Vital Signs blood pressure device according to the ANSI/AAMI SP10: 2002. accuracy and cost-efficiency successfully combined [J]. Blood Pressure Monitoring, 2007, 12(5): 345-347. |
| [12] | Zhang Y, Zhou C, Huang Z, et al. Development of a continuous blood pressure monitoring system based on pulse transit time and hemodynamic covariates [C]// BIODEVICES 2020. Setúbal: SciTePress, 2020: 33-39. |
| [13] | Miao F, Fu N, Zhang Y T, et al. A novel continuous blood pressure estimation approach based on data mining techniques [J]. IEEE Journal of Biomedical and Health Informatics, 2017, 21(6): 1730-1740. |
| [14] | 赵彦峰,于双,王慧泉,等.基于脉搏波传导时间和脉搏波特征参数的连续血压无创检测[J].生物医学工程研究, 2018, 37(1): 42-45. |
| Zhao Yanfeng, Yu Shuang, Wang Huiquan, et al. Non-invasive measurement of continuous blood pressure based on pulse transit time and pulse wave characteristic parameters [J]. Journal of Biomedical Engineering Research, 2018, 37(1): 42-45. | |
| [15] | Khalid S G, Zhang J, Chen F, et al. Blood pressure estimation using photoplethysmography only: comparison between different machine learning approaches [J]. Journal of Healthcare Engineering, 2018, 2018: No.1548647. |
| [16] | Su P, Ding X R, Zhang Y T, et al. Long-term blood pressure prediction with deep recurrent neural networks [C]// EMBS-BHI 2018. Piscataway: IEEE, 2018: 323-328. |
| [17] | Lee D, Kwon H, Son D, et al. Beat-to-beat continuous blood pressure estimation using bidirectional long short-term memory network [J]. Sensors, 2020, 21(1): No.96. |
| [18] | Tazarv A, Levorato M. A deep learning approach to predict blood pressure from PPG signals [C]// EMBC 2021. Piscataway: IEEE, 2021: 5658-5662. |
| [19] | Esmaelpoor J, Moradi M H, Kadkhodamohammadi A. A multistage deep neural network model for blood pressure estimation using photoplethysmogram signals [J]. Computers in Biology and Medicine, 2020, 120: No.103719. |
| [20] | Tanveer M S, Hasan M K. Cuffless blood pressure estimation from electrocardiogram and photoplethysmogram using waveform based ANN-LSTM network [J]. Biomedical Signal Processing and Control, 2019, 51: 382-392. |
| [21] | 熊嘉豪,姜晨希,陈永毅,等.基于TCN-SE神经网络模型的智能连续血压估计方法[J].传感技术学报, 2022, 35(11): 1499-1505. |
| Xiong Jiahao, Jiang Chenxi, Chen Yongyi, et al. Intelligent and continuous estimation of blood pressure method based on TCN-SE neural network model [J]. Chinese Journal of Sensors and Actuators, 2022, 35(11): 1499-1505. | |
| [22] | Liu Z, Zhang Y, Zhou C. BiGRU-attention for continuous blood pressure trends estimation through single channel PPG [J]. Computers in Biology and Medicine, 2024, 168: No.107795. |
| [23] | Hu J, Shen L, Sun G. Squeeze-and-excitation networks [C]// CVPR 2018. Piscataway: IEEE, 2018: 7132-7141. |
| [24] | Vardhan K R, Vedanth S, Poojah G, et al. BP-Net: efficient deep learning for continuous arterial blood pressure estimation using photoplethysmogram [C]// ICMLA 2021. Piscataway: IEEE, 2021: 1495-1500. |
| [25] | Kachuee M, Kiani M M, Mohammadzade H, et al. Cuffless blood pressure estimation algorithms for continuous health-care monitoring [J]. IEEE Transactions on Biomedical Engineering, 2017, 64(4): 859-869. |
| [26] | Altman D G, Bland J M. Measurement in medicine: the analysis of method comparison studies [J]. Journal of the Royal Statistical Society Series D: The Statistician, 1983, 32(3): 307-317. |
| [27] | Ibtehaz N, Mahmud S, Chowdhury M E H, et al. PPG2ABP: translating PhotoPlethysmoGram (PPG) signals to Arterial Blood Pressure (ABP) waveforms [J]. Bioengineering, 2022, 9(11): 692. |
| [28] | Slapničar G, Mlakar N, Luštrek M. Blood pressure estimation from photoplethysmogram using a spectro-temporal deep neural network [J]. Sensors, 2019, 19(15): No.3420. |
| [29] | Mousavi S S, Firouzmand M, Charmi M, et al. Blood pressure estimation from appropriate and inappropriate PPG signals using a whole-based method [J]. Biomedical Signal Processing and Control, 2019, 47: 196-206. |
| [30] | Qin K, Huang W, Zhang T. Deep generative model with domain adversarial training for predicting arterial blood pressure waveform from photoplethysmogram signal [J]. Biomedical Signal Processing and Control, 2021, 70: 102972. |
| [31] | Hermida R C, Ayala D E, Mojón A, et al. Cardiovascular disease risk stratification by the Framingham score is markedly improved by ambulatory compared with office blood pressure [J]. Revista Española de Cardiología (English Edition), 2021, 74(11): 953-961. |
| [32] | Xinyi Chun, Zheng Xiujuan, et al. Non-contact blood pressure estimation method based on genetic algorithm optimized bp neural network [J]. Journal of Electronic Measurement and Instrumentation, 2023, 35(7): 53-59. |
| [33] | Qin C, Li Y, Liu C, et al. Cuff-less blood pressure prediction based on photoplethysmography and modified ResNet [J]. Bioengineering, 2023, 10(4): 400. |
| [1] | 董思语, 贾鑫, 张密, 陈新文, 孙昕霙, 程方骁. 基于CNN-Transformer编码器和多维脑电特征融合的抑郁状态识别模型[J]. 《计算机应用》唯一官方网站, 2026, 46(9): 3025-3034. |
| [2] | 欧宇枫, 王嘉宏. 基于双分支特征提取的印刷电路板小目标缺陷检测方法[J]. 《计算机应用》唯一官方网站, 2026, 46(9): 2977-2986. |
| [3] | 何戡, 马洪峰, 宗学军, 史洪岩, 连莲, 宁博伟. 融合双通道语义特征的工业控制协议模糊测试方法DCSFFuzzer[J]. 《计算机应用》唯一官方网站, 2026, 46(9): 2877-2888. |
| [4] | 谢丽霞, 曹超越, 杨宏宇. 基于改进DeepLabV3+的ADS-B攻击检测方法[J]. 《计算机应用》唯一官方网站, 2026, 46(9): 2867-2876. |
| [5] | 陈江彦, 王彦丹, 刘翼虎, 马应龙. 基于协同查询优化的长尾目标检测框架[J]. 《计算机应用》唯一官方网站, 2026, 46(9): 2769-2775. |
| [6] | 吴昊, 黄渝璐, 冯暄, 张琼, 张晓博. 雷达信号特征识别的动态任务加权混合CNN-Transformer多任务学习方法[J]. 《计算机应用》唯一官方网站, 2026, 46(9): 3017-3024. |
| [7] | 王伊璠, 韩虎, 李栋, 范雅婷, 李琳. 基于异构增强与多源知识融合的方面级情感分析[J]. 《计算机应用》唯一官方网站, 2026, 46(9): 2752-2760. |
| [8] | 高艳, 姚育东. 基于时空Mamba模型的生理信号提取和心率测量[J]. 《计算机应用》唯一官方网站, 2026, 46(8): 2699-2707. |
| [9] | 刘明, 沈东奇, 孟子洋. 双分支结构下多层次特征融合的点云配准网络[J]. 《计算机应用》唯一官方网站, 2026, 46(8): 2584-2593. |
| [10] | 刘凤春, 邵馨莹, 张春英, 王立亚, 任静. 多尺度特征优化的单目深度估计框架FCMdepth[J]. 《计算机应用》唯一官方网站, 2026, 46(8): 2603-2611. |
| [11] | 刘威, 李维刚, 田志强. 面向点云分类与分割的层次化旋转不变几何结构的表征学习方法[J]. 《计算机应用》唯一官方网站, 2026, 46(8): 2594-2602. |
| [12] | 白杰龙, 方晨韵, 乔志伟. 基于多尺度注意力自适应融合的稀疏CT伪影抑制Transformer网络[J]. 《计算机应用》唯一官方网站, 2026, 46(8): 2660-2667. |
| [13] | 李光辉, 屈立成. 基于改进Lite-Mono架构的动态空洞卷积和分层注意力模型[J]. 《计算机应用》唯一官方网站, 2026, 46(8): 2567-2576. |
| [14] | 黄远航, 荣娜. 结合注意力机制与深度强化学习的无模型光伏接入容量评估方法[J]. 《计算机应用》唯一官方网站, 2026, 46(8): 2708-2715. |
| [15] | 黄麟, 李英顺, 佟维妍, 张树园, 王子涵. 基于改进实时检测Transformer的遥感图像目标检测算法[J]. 《计算机应用》唯一官方网站, 2026, 46(8): 2652-2659. |
| 阅读次数 | ||||||
|
全文 |
|
|||||
|
摘要 |
|
|||||