Feasibility of a CT-based lymph node radiomics nomogram in detecting lymph node metastasis in PDAC patients

Abstract

Objectives: To investigate the potential value of a contrast enhanced computed tomography (CECT)-based radiological-radiomics nomogram combining a lymph node (LN) radiomics signature and LNs' radiological features for preoperative detection of LN metastasis in patients with pancreatic ductal adenocarcinoma (PDAC).

Materials and methods: In this retrospective study, 196 LNs in 61 PDAC patients were enrolled and divided into the training (137 LNs) and validation (59 LNs) cohorts. Radiomic features were extracted from portal venous phase images of LNs. The least absolute shrinkage and selection operator (LASSO) regression algorithm with 10-fold cross-validation was used to select optimal features to determine the radiomics score (Rad-score). The radiological-radiomics nomogram was developed by using significant predictors of LN metastasis by multivariate logistic regression (LR) analysis in the training cohort and validated in the validation cohort independently. Its diagnostic performance was assessed by receiver operating characteristic curve (ROC), decision curve (DCA) and calibration curve analyses.

Results: The radiological model, including LN size, and margin and enhancement pattern (three significant predictors), exhibited areas under the curves (AUCs) of 0.831 and 0.756 in the training and validation cohorts, respectively. Nine radiomic features were used to construct a radiomics model, which showed AUCs of 0.879 and 0.804 in the training and validation cohorts, respectively. The radiological-radiomics nomogram, which incorporated the LN Rad-score and the three LNs' radiological features, performed better than the Rad-score and radiological models individually, with AUCs of 0.937 and 0.851 in the training and validation cohorts, respectively. Calibration curve analysis and DCA revealed that the radiological-radiomics nomogram showed satisfactory consistency and the highest net benefit for preoperative diagnosis of LN metastasis.

Conclusions: The CT-based LN radiological-radiomics nomogram may serve as a valid and convenient computer-aided tool for personalized risk assessment of LN metastasis and help clinicians make appropriate clinical decisions for PADC patients.

Keywords: computed tomography; lymph node metastasis; nomogram; pancreatic ductal adenocarcinoma; radiomics.

Figure 1

Flowchart of patients’ selection.

Figure 2

The workflow of the radiomics analysis of LNs in this study. Step 1: LNs were semi-automatically segmented slice by slice in portal venous phase images. Step 2: Radiomics features were extracted from the identified VOIs. Step 3: The LASSO logistic regression with penalty parameter tuning conducted by 10-fold cross-validation was used to select the optimal radiomics features. Step 4: The LN Rad-score and the nomogram incorporating radiological features with Rad-score were established.

Figure 3

The framework for radiomics features selection. (A) The LASSO logistic regression was used to select LN radiomics. A tuning parameter was selected via 10-fold cross-validation and nine with nonzero coefficients were selected finally. (B) Histogram shows the role of nine selected radiomics features used to calculate the Rad-score. The y-axis represents individual radiomics features, with their coefficients in the LASSO regression analysis plotted on the x-axis.

Figure 4

ROC curves of the radiological model, Rad-score model and radiological-radiomics nomogram for diagnosing metastatic LNs in the training cohort and validation cohort.

Figure 5

A radiological-radiomics nomogram was plotted combining independent radiological features with Rad-score in the training cohort (A). Calibration curves for the radiological-radiomics nomogram in the training cohort and in the validation cohort (B). The 45° straight line indicates the ideal performance of the radiological-radiomics nomogram. A closer distance between two curves indicates higher accuracy.

Figure 6

DCA for the Rad-score model and radiological-radiomics nomogram in the training cohort (A) and validation cohort (B). The y-axis measures the net benefit and the x-axis represents the threshold probability. The grey line that all patients had LN metastasis and the black line indicate no patients had LN metastasis. The red line and the green line indicate the net benefit of the Rad-score model and the radiological-radiomics nomogram at different threshold probabilities, respectively. The radiomics nomogram had a higher overall net benefit in differentiating LN metastasis than Rad-score model.