Value of a nomogram based on contrast-enhanced CT radiomics score and clinical data for prediction of tumor thrombus in patients with adenocarcinoma of esophagogastric junction

  Background   The incidence of adenocarcinoma of esophagogastric junction (AEG) has been increasing. Due to its special location and unstable biological behavior, prognosis of AEG is poor. Tumor thrombus is an independent risk factor affecting the prognosis of AEG. Preoperative understanding of tumor thrombus of AEG can help clinicians scientifically and objectively formulate a more reasonable individual treatment plan.

  Objective   To evaluate the application value of CT radiomics in prediction of vascular tumor thrombus of AEG.

  Methods  CT images of 79 patients with AEG who were diagnosed and surgically treated in the Second Department of General Surgery, the First Medical Center, Chinese PLA General Hospital from January 2015 to July 2019 were selected. 3D Slicer software was used to extract the radiomics features from the images of the largest lesion layer of venous CT images, and the radiomics features were screened out by Lasso regression dimensionality reduction to construct the prediction model. Radiomic scores were calculated, then the radiomics scores and clinical data were used as parameters to construct the nomogram. The receiver operator characteristic curve (ROC) was used to evaluate the model for preoperative prediction of tumor thrombus of AEG.

  Results  Of the 79 cases, 68 cases were male and 11 cases were female, and the age ranged from 30 to 80 years, with the average of (63.8 + 9.5) years. There were 54 patients without tumor thrombus and 25 patients with tumor thrombus, and the incidence of tumor thrombus was 31.6%. Totally 873 features were extracted from each patient, while 7 features including 2D maximum diameter were finally determined by Lasso regression analysis and ten fold cross validation to construct a radiomics model, with AUC of 0.852, sensitivity of 0.880 and specificity of 0.849. The AUC, sensitivity and specificity of the nomogram were 0.885, 0.880 and 0.892, respectively. Two-thirds of the samples were randomly sampled for ROC analysis, and the results were very close to the established model.

  Conclusion   The nomogram based on contrast-enhanced CT radiomics score and clinical data has better predictive efficacy for vascular tumor thrombus of AEG.