Novel fusion architecture of multi-location blood flow sounds for arteriovenous fistula stenosis diagnosis
Computer Methods and Programs in Biomedicine · 2025-11 · 271 · 109022
DOI: 10.1016/j.cmpb.2025.109022 ↗Kurz gefasst
Fachartikel in der Zeitschrift Computer Methods and Programs in Biomedicine, 2025-11. Verfasst von Haowei Liu und 11 weiteren Autor:innen. Beteiligte Einrichtung: Key Laboratory of Speech Acoustics and Content Understanding Institute of Acoustics, Beijing, China; University of Chinese Academy of Sciences, Beiji…. Hauptthemen laut MEDLINE: Arteriovenous Fistula, Constriction, Pathologic, Models, Cardiovascular, Sound. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 36 Literaturangaben.
Abstract (Originalsprache)
BACKGROUND AND OBJECTIVE: Arteriovenous fistula (AVF) is a surgical intervention that establishes an artificial conduit between an artery and a vein, which is essential for hemodialysis treatment. The blood flow sounds has the potential to diagnose AVF stenosis. The research primarily addresses two critical questions: whether there is stenosis within the vascular pathway, and the precise location of the stenosis within the vascular pathway. However, most existing studies focus only on the anastomosis site using single-location acoustic data. This approach fails to fully leverage the acoustic information from other segments of the vascular pathway, limiting the overall diagnostic accuracy. METHODS: Given that stenosis at one site affects acoustic properties elsewhere, we propose a multi-position fusion architecture with position metadata (MPFA). Two fusion strategies are introduced: channel fusion and temporal fusion. In the proposed method, acoustic features extracted from five consecutive locations are fused either across channels or over time dimension. Each location undergoes learnable encoding, which is then integrated into the acoustic features. Following this, a deep feature extractor is used to extract deep feature, which are then fed into a classifier for classification. This architecture is integrated into both the universal model and the individual location-based model to address the two aforementioned questions, respectively. This approach enhances diagnosis through leveraging multi-location information, not model complexity, and can be embedded into other network architectures. RESULTS: Extensive experimental results have demonstrated that multi-location fusion method has significant advantages over single-location. Furthermore, the integration of position metadata serves as a beneficial augmentation, overall improving diagnostic accuracy in the temporal fusion architecture. For the diagnosis of stenosis at the anastomosis site, the accuracy can reach 93.94%. For the antecubital vein segment, the accuracy reach 87.58%. In the universal model, the overall accuracy can reach 79.45%. CONCLUSION: The findings of our experiments demonstrated the effectiveness of a multi-position fusion architecture with position metadata in the AVF stenosis diagnosis. The MPFA does not rely on increasing model parameters and its simplicity and flexibility allow easy integration into other deep learning models. This approach provides researchers with a new perspective for diagnosing AVF stenosis using blood flow sounds.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Key Laboratory of Speech Acoustics and Content Understanding Institute of Acoustics, Beijing, China; University of Chinese Academy of Sciences, Beijing, China. Electronic address: lhwjsj@163.com.
- The First Hospital of Hebei Medical University, Shijiazhuang City, China. Electronic address: wfl977172@qq.com.
- Key Laboratory of Speech Acoustics and Content Understanding Institute of Acoustics, Beijing, China; University of Chinese Academy of Sciences, Beijing, China. Electronic address: 1102597677@qq.com.
- Zhejiang Hospital, Hangzhou City, China. Electronic address: xyk19870309@126.com.
- Key Laboratory of Speech Acoustics and Content Understanding Institute of Acoustics, Beijing, China; University of Chinese Academy of Sciences, Beijing, China. Electronic address: lushouda@mail.ioa.ac.cn.
- Key Laboratory of Speech Acoustics and Content Understanding Institute of Acoustics, Beijing, China; University of Chinese Academy of Sciences, Beijing, China. Electronic address: zhangpengyuan@mail.ioa.ac.cn.
Zitieren
APA
Liu, H., Wang, F., Zhang, X., Xu, Y., Lu, S., Zhang, P., Ruan, L., Yang, Y., Li, W., Liu, Y., Yan, Y., & Zhang, L. (2025). Novel fusion architecture of multi-location blood flow sounds for arteriovenous fistula stenosis diagnosis. Computer Methods and Programs in Biomedicine, 271, 109022. https://doi.org/10.1016/j.cmpb.2025.109022
Vancouver
Liu H, Wang F, Zhang X, Xu Y, Lu S, Zhang P, et al. Novel fusion architecture of multi-location blood flow sounds for arteriovenous fistula stenosis diagnosis. Computer Methods and Programs in Biomedicine. 2025;271:109022. doi:10.1016/j.cmpb.2025.109022
BibTeX
@article{Liu2025,
author = {Liu, Haowei and Wang, Fanli and Zhang, Xueshuai and Xu, Yuankai and Lu, Shouda and Zhang, Pengyuan and Ruan, Lin and Yang, Yanli and Li, Wen and Liu, Yongliang and Yan, Yonghong and Zhang, Lihong},
title = {Novel fusion architecture of multi-location blood flow sounds for arteriovenous fistula stenosis diagnosis},
journal = {Computer Methods and Programs in Biomedicine},
year = {2025},
volume = {271},
pages = {109022},
doi = {10.1016/j.cmpb.2025.109022}
}
EUVIMED