Multiregional radiomics score based on multisequence MRI to predict vessels that encapsulate tumor cluster and/or microvascular invasion-positive hepatocellular carcinoma
10.3760/cma.j.cn113884-20240731-00232
- VernacularTitle:基于多序列MRI的多区域影像组学评分预测血管包绕肿瘤细胞簇和/或微血管侵犯阳性肝细胞癌
- Author:
Zixin LIU
1
;
Zuyi YAN
;
Tao ZHANG
;
Xueqin ZHANG
;
Chunyan GU
;
Qi QU
;
Jifeng JIANG
Author Information
1. 南通大学附属南通第三医院（南通市第三人民医院）放射科，南通 226006
- Publication Type:Journal Article
- Keywords:
Carcinoma, hepatocellular;
Vessels that encapsulate tumor cluster;
Microvascular invasion;
Gd-EOB-DTPA;
Machine learning
- From:
Chinese Journal of Hepatobiliary Surgery
2024;30(12):886-892
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To develop a nomogram to predict vessels that encapsulate tumor cluster and/or microvascular invasion-positive hepatocellular carcinoma (VM-HCC) based on multiregional radiomics score derived from multisequence MRI.Methods:Clinical data of 209 patients with HCC undergoing radical liver resection at Affiliated Nantong Hospital 3 of Nantong University from January 2016 to December 2021 were retrospectively analyzed, including 149 males and 60 females, aged (58.5±9.2) years. Patients were divided into a training set ( n=146) and a testing set ( n=63). The patients in training set were further classified into two groups based on pathological results: the VM-HCC group ( n=76) and the non-VM-HCC group ( n=70). Radiomics features were extracted from the arterial phase and hepatobiliary phase images within the tumor, peritumor, and combined regions of interest. The arterial phase and hepatobiliary phase features from the same regions were integrated to obtain dual-sequence features. After feature selection, linear support vector machines (SVM) and linear regression machine learning classifiers were employed to construct radiomics models for different sequences and regions. The optimal radiomics model was selected based on the area under the receiver operating characteristic (ROC) curve from the testing set. Logistic regression analysis was performed to identify independent predictors of VM-HCC, and a visual nomogram was constructed using the results of the multivariate logistic regression analysis and the radiomics score of the optimal radiomics model. ROC curves were plotted, and area under curve (AUC) were calculated to evaluate the models’ discriminatory ability. Calibration curves and decision curve analysis (DCA) were utilized to assess the model’s calibration and clinical utility. Results:Among the radiomics models, the dual-sequence-combined region model based on the SVM classifier exhibited the best AUC in the testing set (AUC=0.764, 95% CI: 0.646-0.882). Multivariate logistic regression analysis indicated that HCC patients with protein induced by vitamin K absence or antagonist-II (PIVKA-II) levels >40 mAU/ml ( OR=4.266, 95% CI: 1.921-9.473, P<0.001) had a higher risk of VM-HCC. The nomogram combining PIVKA-II>40 mAU/ml and radiomics score achieved AUC of 0.806 (95% CI: 0.733-0.867) in the training set and 0.817 (95% CI: 0.699-0.903) in the testing set for predicting VM-HCC. The calibration curves of the nomogram showed good fit in both the training and testing sets. DCA indicated that the model possesses good clinical utility. Conclusion:The nomogram based on multiregional radiomics score derived from multisequence MRI demonstrates a good predictive performance for VM-HCC, which could facilitate the risk stratification of recurrence in HCC patients.
