Lightweight end-to-end model-based korotkoff sounds phase identification and blood pressure measurement
- VernacularTitle:基于轻量化端到端模型的柯氏音时相识别与血压测量
- Author:
Zhiyu JIANG
1
;
Wenyi KOU
1
;
Li LI
2
;
Qijun ZHAO
3
;
Yongjun QIAN
4
;
Fan PAN
1
Author Information
1. School of Electronic Information, Sichuan University, Chengdu, 610065, P. R. China
2. Department of Pediatrics, West China Second University Hospital, Sichuan University, Chengdu, 610041, P. R. China
3. School of Computer Science, Sichuan University, Chengdu, 610065, P. R. China
4. Department of Cardiovascular Surgery, West China Hospital, Sichuan University, Chengdu, 610041, P. R. China
- Publication Type:Journal Article
- Keywords:
Korotkoff sounds;
phase identification;
blood pressure measurement;
end-to-end model;
deep learning;
artificial intelligence;
convolutional neural network;
gated recurrent unit
- From:
Chinese Journal of Clinical Thoracic and Cardiovascular Surgery
2026;33(02):248-254
- CountryChina
- Language:Chinese
-
Abstract:
Objective To propose a lightweight end-to-end neural network model for automated Korotkoff sound phase recognition and subsequent blood pressure (BP) measurement, aiming to improve measurement accuracy and population adaptability. Methods We developed a streamlined architecture integrating depthwise separable convolution (DSConv), multi-head attention (MHA), and bidirectional gated recurrent unit (BiGRU). The model directly processes Korotkoff sound time-series signals to identify auscultatory phases. Systolic BP (SBP) and diastolic BP (DBP) were determined using phase Ⅰ and phaseⅤdetections, respectively. Given the clinical relevance of phase Ⅳ for specific populations (e.g., children and pregnant women, denoted as DBPⅣ), BP values from this phase were also recorded.Results The study enrolled 106 volunteers with 70 males and 36 females at mean age of (40.0±12.0) years. The model achieved 94.25% phase recognition accuracy. Measurement errors were (0.1±2.5) mm Hg (SBP), (0.9±3.4) mm Hg (DBPⅣ), and (0.8±2.6) mm Hg (DBP). Conclusion Our method enables precise phase recognition and BP measurement, demonstrating potential for developing population-adaptive blood pressure monitoring systems.
