Multiparametric radiomics nomogram may be used for predicting the severity of esophageal varices in cirrhotic patients

Abstract

Background: To explore whether a multiparametric radiomics nomogram on computed tomography (CT) images based on radiomics and relevant parameters of esophageal varices (EV) can be used for predicting the EV severity in patients with cirrhotic livers. Methods:From January 2016 to August 2018, 136 consecutive patients with clinicopathologically confirmed liver cirrhosis were included for the development of a predictive model. The patients were then divided into two groups, including non-conspicuous EV group (mild-to-moderate EV, n=30) and conspicuous EV group (severe EV, n=106) by using the endoscopic validation as the reference standard.The radiomic scores (Rad scores) were constructed using the binary logistic regression model from the radiomics features of regions of interest (ROIs) in the left liver (LL) and right liver (RL), respectively. The multiparametric nomogram combined the best performance Rad-score and EV-relevant factors, and the calibration, discrimination, and clinical usefulness of developed nomogram were evaluated using calibration curves, decision curve analysis (DCA) and net reclassification index (NRI) analysis respectively. Results:The LL Rad-score calculated from radiomics features was selected with a relatively higher area under the curve (AUC) (AUC; 0.88, training cohort; 0.87, the validation cohort) compared with RL Radscore (AUC; 0.86, training cohort; 0.83, the validation cohort). In addition, cross-sectional surface area (CSA) was identified as the important predictor (P<0.05), the multiparametric nomogram containing LL Rad-score and CSA was shown to have a better predictive performance and good calibration in the training model (C-index, 0.953, 95% CI, 0.892 to 0.973) and the validation cohort (C-index, 0.938, 95% CI, 0.841 to 0.961), resulting in an improved NRI (categorical NRI of 25.9%, P=0.0128; continuous NRI of 120%, P<0.001) and integrated discriminatory improvement (IDI) (IDI =13.9%, P<0.001). DCA demonstrated that the multiparametric radiomics nomogram was clinically useful.Conclusions: A multiparametric radiomics nomogram, which incorporates the liver radiomics signature and EV-relevant indices, is a useful tool for noninvasively predicting EV severity and may complement the standard endoscopy for evaluating EV severity in patients with cirrhosis.

A Multi-Organ Fusion and LightGBM Based Radiomics Algorithm for High-Risk Esophageal Varices Prediction in Cirrhotic Patients

“…Esophageal varices (EV) bleeding is the principal and lifethreatening complication of portal hypertension in cirrhosis [1], [2], [3]. Esophageal varices can be categorized as low-risk and high-risk varices on the basis of the risk of variceal bleeding which is decided by the size of the varices and the presence of red color signs [4].…”

“…Radiomics is the study of a large number of quantitative features extraction from medical images which converts image data into a high-dimensional discoverable feature space [1]. Numerous studies indicate that radiomics analysis combined with machine learning have achieved remarkable progress in quantifying the state of various diseases, especially in the improvement of the diagnostic, prognostic, and predictive accuracy of cancer [15], [16], [17], [18], [19], [20].…”

“…To date, few studies applied radiomics to classify and predict the risk and severity levels of esophageal varices [27], [28]. Wan et al proposed a multiparametric radiomics nomogram on CT images based on radiomics and relevant parameters of EV which can be used for predicting the EV severity in patients with cirrhotic livers [1]. However, only the left and right liver were applied for feature selection which might capture partial information about EV and therefore influence the resulting classification performance.…”

“…Therefore, the proposed multi-organ fusion and LightGBM based method is effective for the improvement of classification performance, which indicates that high-risk EV patients are more correctly detected and the lower probability for high-risk being misdiagnosed as lowrisk, verifying its feasibility of an auxiliary tool for clinical EV diagnosis. We also implemented the EV severity prediction with the proposed methods and the existing methods based on the same dataset, including binary logistic regression model based on liver (BLR-L) [1], binary logistic regression model based on liver and spleen (BLR-LS) [21] and tree-based method based on liver (TB-L) [ Ⅹ, the results show that the EV prediction performance with the esophagus organ has been greatly improved compared with the existing methods. The proposed multi-organ feature fusion and classifier ensemble method further improves the classification accuracy.…”

A novel radiomics–platelet nomogram for the prediction of gastroesophageal varices needing treatment in cirrhotic patients

“…Radiomics is a newly emerging technology of image analysis which refers to extracting high-throughput and quantitative features from medical images, revealing the correlation between these features and the disease using data mining algorithms and statistics analysis, then builds an appropriate model with re ning features [15,16]. Previous studies suggest that the potential application of radiomics in predicting VNT [17][18][19]. However, the previous radiomics models do not contain the esophageal and gastric radiomics features which are important evidence for the radiologist to determine the existence and severity of GEV.…”