Fang Hu1,2, Yuhan Zhang1, Man Li3, Chen Liu1, Handan Zhang1, Xiaoming Li1, Sanyuan Liu3, Xiaofei Hu1, Jian Wang1. 1. Department of Radiology, Southwest Hospital, Third Military Medical University (Army Medical University), Chongqing, China. 2. Department of Radiology, Tongliang District People's Hospital, Chongqing, China. 3. Department of Research and Development, Shanghai United Imaging Intelligence Co., Ltd., Shanghai, China.
Abstract
Objective: To predict preoperative microvascular invasion (MVI) risk grade by analyzing the radiomics signatures of tumors and peritumors on enhanced magnetic resonance imaging (MRI) images of hepatocellular carcinoma (HCC). Methods: A total of 501 HCC patients (training cohort n = 402, testing cohort n = 99) who underwent preoperative Gd-EOB-DTPA-enhanced MRI and curative liver resection within a month were studied retrospectively. Radiomics signatures were selected using the least absolute shrinkage and selection operator (Lasso) algorithm. Unimodal radiomics models based on tumors and peritumors (10mm or 20mm) were established using the Logistic algorithm, using plain T1WI, arterial phase (AP), portal venous phase (PVP), and hepatobiliary phase (HBP) images. Multimodal radiomics models based on different regions of interest (ROIs) were established using a combinatorial modeling approach. Moreover, we merged radiomics signatures and clinico-radiological features to build unimodal and multimodal clinical radiomics models. Results: In the testing cohort, the AUC of the dual-region (tumor & peritumor 20 mm)radiomics model and single-region (tumor) radiomics model were 0.741 vs 0.694, 0.733 vs 0.725, 0.667 vs 0.710, and 0.559 vs 0.677, respectively, according to AP, PVP, T1WI, and HBP images. The AUC of the final clinical radiomics model based on tumor and peritumoral 20mm incorporating radiomics features in AP&PVP&T1WI images for predicting MVI classification in the training and testing cohorts were 0.962 and 0.852, respectively. Conclusion: The radiomics signatures of the dual regions for tumor and peritumor on AP and PVP images are of significance to predict MVI.
Objective: To predict preoperative microvascular invasion (MVI) risk grade by analyzing the radiomics signatures of tumors and peritumors on enhanced magnetic resonance imaging (MRI) images of hepatocellular carcinoma (HCC). Methods: A total of 501 HCC patients (training cohort n = 402, testing cohort n = 99) who underwent preoperative Gd-EOB-DTPA-enhanced MRI and curative liver resection within a month were studied retrospectively. Radiomics signatures were selected using the least absolute shrinkage and selection operator (Lasso) algorithm. Unimodal radiomics models based on tumors and peritumors (10mm or 20mm) were established using the Logistic algorithm, using plain T1WI, arterial phase (AP), portal venous phase (PVP), and hepatobiliary phase (HBP) images. Multimodal radiomics models based on different regions of interest (ROIs) were established using a combinatorial modeling approach. Moreover, we merged radiomics signatures and clinico-radiological features to build unimodal and multimodal clinical radiomics models. Results: In the testing cohort, the AUC of the dual-region (tumor & peritumor 20 mm)radiomics model and single-region (tumor) radiomics model were 0.741 vs 0.694, 0.733 vs 0.725, 0.667 vs 0.710, and 0.559 vs 0.677, respectively, according to AP, PVP, T1WI, and HBP images. The AUC of the final clinical radiomics model based on tumor and peritumoral 20mm incorporating radiomics features in AP&PVP&T1WI images for predicting MVI classification in the training and testing cohorts were 0.962 and 0.852, respectively. Conclusion: The radiomics signatures of the dual regions for tumor and peritumor on AP and PVP images are of significance to predict MVI.
Authors: Alessandro Cucchetti; Guo-Liang Qiao; Matteo Cescon; Jun Li; Yong Xia; Giorgio Ercolani; Feng Shen; Antonio Daniele Pinna Journal: Surgery Date: 2013-10-14 Impact factor: 3.982