Prediction of microvascular invasion in hepatocellular carcinoma using contrast-enhanced ultrasound and deep learning.
Authors
Affiliations (15)
Affiliations (15)
- Department of Interventional Ultrasound, Senior Department of Oncology, Chinese PLA General Hospital, Beijing, China.
- CAS Key Laboratory of Molecular Imaging, Institute of Automation, Chinese Academy of Sciences, Beijing, China.
- School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing, China.
- School of Medicine, Nankai University, Tianjin, China.
- Department of Ultrasound, Fifth Medical Center of Chinese PLA General Hospital, Beijing, China.
- Department of Medical Ultrasound, Xuzhou Central Hospital, Xuzhou, China.
- Department of Ultrasound, The First Affiliated Hospital of Henan University of Science and Technology, Luoyang, China.
- Department of Pathology and Hepatology, Fifth Medical Center of Chinese PLA General Hospital, Beijing, China.
- Center of Hepato-Pancreato-Biliary Surgery, The First Affiliated Hospital of Sun Yat-sen University, Guangzhou, China.
- Institute of Diagnostic and Interventional Ultrasound, The First Affiliated Hospital, Sun Yat-sen University, Guangzhou, China.
- School of Engineering Medicine, Beihang University, Beijing, China.
- CAS Key Laboratory of Molecular Imaging, Institute of Automation, Chinese Academy of Sciences, Beijing, China. [email protected].
- School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing, China. [email protected].
- Department of Interventional Ultrasound, Senior Department of Oncology, Chinese PLA General Hospital, Beijing, China. [email protected].
- Department of Interventional Ultrasound, Senior Department of Oncology, Chinese PLA General Hospital, Beijing, China. [email protected].
Abstract
Microvascular invasion (MVI) is a key prognostic factor in hepatocellular carcinoma but is currently only detectable after surgery. Here, we develop MAPUSE, a deep learning model using contrast-enhanced ultrasound (CEUS) to predict MVI non-invasively. We train and test the model on 5148 CEUS videos from 1716 patients across multiple centers. Results show that MAPUSE achieves accurate MVI prediction (AUCs 0.835-0.978) across different tumor sizes, contrast agents, and prospective validations. Transcriptomic analysis links the model's predictions to CD8 + T cell immune infiltration, confirmed via the model's attention maps. In a clinical cohort, patients predicted as MVI-positive can benefit from post-ablation immunotherapy. MAPUSE thus enables preoperative, non-invasive MVI assessment and provides insights into the tumor immune microenvironment, offering a valuable tool for clinical decision-making.