Automated microvascular invasion prediction of hepatocellular carcinoma via deep relation reasoning from dynamic contrast-enhanced ultrasound
Computerized Medical Imaging and Graphics · 2025-09 · 124 · 102606
DOI: 10.1016/j.compmedimag.2025.102606 ↗Kurz gefasst
Fachartikel in der Zeitschrift Computerized Medical Imaging and Graphics, 2025-09. Verfasst von Yaoqin Wang und 6 weiteren Autor:innen. Beteiligte Einrichtung: Department of Ultrasound, Clinical Oncology School of Fujian Medical University, Fujian Cancer Hospital, Fuzhou 350014, Fujian Province, China. Hauptthemen laut MEDLINE: Liver Neoplasms, Carcinoma, Hepatocellular, Contrast Media, Microvessels. Die Arbeit wurde bisher 8-mal zitiert (Crossref) und nennt 71 Literaturangaben. Gefördert von: Fujian Provincial Natural Science Foundation.
Abstract (Originalsprache)
Hepatocellular carcinoma (HCC) is a major global health concern, with microvascular invasion (MVI) being a critical prognostic factor linked to early recurrence and poor survival. Preoperative MVI prediction remains challenging, but recent advancements in dynamic contrast-enhanced ultrasound (CEUS) imaging combined with artificial intelligence show promise in improving prediction accuracy. CEUS offers real-time visualization of tumor vascularity, providing unique insights into MVI characteristics. This study proposes a novel deep relation reasoning approach to address the challenges of modeling intricate temporal relationships and extracting complex spatial features from CEUS video frames. Our method integrates CEUS video sequences and introduces a visual graph reasoning framework that correlates intratumoral and peritumoral features across various imaging phases. The system employs dual-path feature extraction, MVI pattern topology construction, Graph Convolutional Network learning, and an MVI pattern discovery module to capture complex features while providing interpretable results. Experimental findings demonstrate that our approach surpasses existing state-of-the-art models in accuracy, sensitivity, and specificity for MVI prediction. The system achieved superiors accuracy, sensitivity, specificity and AUC. These advancements promise to enhance HCC diagnosis and management, potentially revolutionizing patient care. The method's robust performance, even with limited data, underscores its potential for practical clinical application in improving the efficacy and efficiency of HCC patient diagnosis and treatment planning.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Substanzen
Autor:innen und Institutionen
- Department of Ultrasound, Clinical Oncology School of Fujian Medical University, Fujian Cancer Hospital, Fuzhou 350014, Fujian Province, China.
- Department of Electrical Engineering, The Chinese University of Hong Kong, Shatin, Hong Kong Special Administrative Region.
- Department of Ultrasound, Clinical Oncology School of Fujian Medical University, Fujian Cancer Hospital, Fuzhou 350014, Fujian Province, China. Electronic address: tanglina@fjzlhospital.com.
Förderung
- Fujian Provincial Natural Science Foundation
Zitieren
APA
Wang, Y., Xie, W., Li, C., Xu, Q., Du, Z., Zhong, Z., & Tang, L. (2025). Automated microvascular invasion prediction of hepatocellular carcinoma via deep relation reasoning from dynamic contrast-enhanced ultrasound. Computerized Medical Imaging and Graphics, 124, 102606. https://doi.org/10.1016/j.compmedimag.2025.102606
Vancouver
Wang Y, Xie W, Li C, Xu Q, Du Z, Zhong Z, et al. Automated microvascular invasion prediction of hepatocellular carcinoma via deep relation reasoning from dynamic contrast-enhanced ultrasound. Computerized Medical Imaging and Graphics. 2025;124:102606. doi:10.1016/j.compmedimag.2025.102606
BibTeX
@article{Wang2025,
author = {Wang, Yaoqin and Xie, Wenting and Li, Chenxin and Xu, Qing and Du, Zhongshi and Zhong, Zhaoming and Tang, Lina},
title = {Automated microvascular invasion prediction of hepatocellular carcinoma via deep relation reasoning from dynamic contrast-enhanced ultrasound},
journal = {Computerized Medical Imaging and Graphics},
year = {2025},
volume = {124},
pages = {102606},
doi = {10.1016/j.compmedimag.2025.102606}
}
EUVIMED