Radiomics Analysis of Contrast-Enhanced CT for the Preoperative Prediction of Microvascular Invasion in Mass-Forming Intrahepatic Cholangiocarcinoma

Abstract

BackgroundMicrovascular invasion (MVI) has been shown to be closely associated with postoperative recurrence and metastasis in patients with intrahepatic cholangiocarcinoma (ICC). We aimed to develop a radiomics prediction model based on contrast-enhanced CT (CECT) to distinguish MVI in patients with mass-forming ICC.Methods157 patients were included and randomly divided into training (n=110) and test (n=47) datasets. Radiomic signatures were built based on the recursive feature elimination support vector machine (Rfe-SVM) algorithm. Significant clinical-radiologic factors were screened, and a clinical model was built by multivariate logistic regression. A nomogram was developed by integrating radiomics signature and the significant clinical risk factors.ResultsThe portal phase image radiomics signature with 6 features was constructed and provided an area under the receiver operating characteristic curve (AUC) of 0.804 in the training and 0.769 in the test datasets. Three significant predictors, including satellite nodules (odds ratio [OR]=13.73), arterial hypo-enhancement (OR=4.31), and tumor contour (OR=4.99), were identified by multivariate analysis. The clinical model using these predictors exhibited an AUC of 0.822 in the training and 0.756 in the test datasets. The nomogram combining significant clinical factors and radiomics signature achieved satisfactory prediction efficacy, showing an AUC of 0.886 in the training and 0.80 in the test datasets.ConclusionsBoth CECT radiomics analysis and radiologic factors have the potential for MVI prediction in mass-forming ICC patients. The nomogram can further improve the prediction efficacy.

Conventional, functional and radiomics assessment for intrahepatic cholangiocarcinoma

“…To date, in several studies have been assessed the role of radiomics or radiogenomics in ICC [ 72 , 79 , 180 – 204 ]. The area of main interest has been the evaluation of recurrence after surgical resection [ 72 , 79 , 180 – 187 ].…”

“…To date, in several studies have been assessed the role of radiomics or radiogenomics in ICC [ 72 , 79 , 180 – 204 ]. The area of main interest has been the evaluation of recurrence after surgical resection [ 72 , 79 , 180 – 187 ]. Chu et al [ 79 ] evaluated 203 ICCs, that were allocated with a ratio of 7:3 into the training cohort and the validation cohort.…”

“…Among those models, Max-Relevance Min-Redundancy (MRMR) combined with Gradient Boosting Machine (GBM) yielded the highest AUCs of 0.802 and 0.781 in the training and test cohorts, respectively [ 181 ]. Xiang et al [ 72 ] developed a radiomics prediction model based on contrast-enhanced CT to distinguish Microvascular invasion (MVI) in ICCs. Also, Zhou et al [ 127 ] develop a radiomics signature for preoperative prediction of MVI in ICC patients.…”

A nomogram based on radiomics to predict microvascular invasion and overall survival in intrahepatic cholangiocarcinoma

“…Secondly, despite univariate logistic regression analysis showed that tumor morphology was a signi cant clinical feature, it was not an independent feature in MVI status strati cation and overall survival prediction preoperatively. However, one con icting study reported that irregular morphology may be associated with MVI status (25), this may owe to tumor morphology is a subjective variable, and more importantly, we considered lobulated and irregular as two morphological categories, while considered as one category in his/her study. Thirdly, as in previous study (17), tumor size and intrahepatic duct dilatation were independent predictors in MVI strati cation.…”

Editorial for “A Multi‐Modality Fusion Deep Learning Model Based on DCE‐MRI for Preoperative Prediction of Microvascular Invasion in Intrahepatic Cholangiocarcinoma”

“…Currently, the presence of MVI is confirmed on histology, which is invasive. A few studies based on radiomic analysis to predict the MVI in mass‐forming ICC were reported (see, eg Reference 8). However, this approach requires accurate tumor segmentation that may not always be possible.…”