Multicenter study on the prediction of microvascular invasion in hepatocellular carcinoma using multiphase ultrasound imaging radiomics models
10.3760/cma.j.cn131148-20250601-00300
- VernacularTitle:多期相超声影像组学模型预测肝细胞癌微血管侵犯的多中心研究
- Author:
Yanhong HAO
1
;
Juan CHEN
;
Qin LU
;
Ruining WANG
;
Yuan SU
;
Shanshan SHI
;
Rui SHI
;
Lingjie WANG
;
Jianhong WANG
;
Li YANG
;
Liping LIU
Author Information
1. 山西医科大学第一医院超声科，太原 030001
- Publication Type:Journal Article
- Keywords:
Carcinoma，hepatocellular;
Microvascular invasion;
Radiomics;
Contrast-enhanced ultrasound;
Combined model
- From:
Chinese Journal of Ultrasonography
2025;34(11):983-991
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To construct and evaluate the predictive performance of a multiphase ultrasound radiomics model for microvascular invasion（MVI）in hepatocellular carcinoma（HCC）.Methods:A total of 126 patients with pathologically confirmed HCC were retrospectively enrolled from 4 medical centers between May 2018 and July 2025，including the First Hospital of Shanxi Medical University，Shanxi Province Third People's Hospital，Changzhi People's Hospital，and the Organ Transplant Center of the Affiliated Hospital of Qingdao University. A total of 630 ultrasound images of the lesions in different phases were collected，from which 1 561 radiomic features were extracted. The patients from medical institutions in Shanxi Province were chosen as the training set（ n=91），and the patients from the Organ Transplant Center of the Affiliated Hospital of Qingdao University were chosen as the validation set（ n=35）. In the training set，37.4%（34/91）patients presented MVI（+），whereas in the validation set，54.3%（19/35）patients presented MVI（+）. Radiomics features were extracted from ultrasound images，and features related to the MVI（+）were selected through dimensionality reduction analysis. Five multiple machine learning algorithms were used to construct predictive models，which were then evaluated using an external validation set. The Radscore was calculated，and a nomogram was constructed combining Radscore with ultrasound and clinical characteristics to predict MVI. Results:The model combining radiomics features from the portal venous phase and the delay phase showed the best predictive performance in both the training and validation sets，with area under curve（AUC）values of 0.835 and 0.727，respectively. The prediction model developed using radiomics Radscore and clinical indicators could be represented and presented as a nomogram.Conclusions:The radiomics model based on multi-phase ultrasound offers a novel approach for non-invasive preoperative prediction of MVI in liver cancer. Furthermore，its integration with clinical features aids in optimizing clinical treatment strategies.
