CT-Based Radiomics Nomogram Improves Risk Stratification and Prediction of Early Recurrence in Hepatocellular Carcinoma After Partial Hepatectomy

Abstract

Objectives: To develop and validate an intuitive computed tomography (CT)-based radiomics nomogram for the prediction and risk stratification of early recurrence (ER) in hepatocellular carcinoma (HCC) patients after partial hepatectomy.

Methods: A total of 132 HCC patients treated with partial hepatectomy were retrospectively enrolled and assigned to training and test sets. Least absolute shrinkage and selection operator and gradient boosting decision tree were used to extract quantitative radiomics features from preoperative contrast-enhanced CT images of the HCC patients. The radiomics features with predictive value for ER were used, either alone or in combination with other predictive features, to construct predictive models. The best performing model was then selected to develop an intuitive, simple-to-use nomogram, and its performance in the prediction and risk stratification of ER was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curve, and decision curve analysis (DCA).

Results: The radiomics model based on the radiomics score (Rad-score) achieved AUCs of 0.870 and 0.890 in the training and test sets, respectively. Among the six predictive models, the combined model based on the Rad-score, Edmondson grade, and tumor size had the highest AUCs of 0.907 in the training set and 0.948 in the test set and was used to develop an intuitive nomogram. Notably, the calibration curve and DCA for the nomogram showed good calibration and clinical application. Moreover, the risk of ER was significantly different between the high- and low-risk groups stratified by the nomogram (p <0.001).

Conclusions: The CT-based radiomics nomogram developed in this study exhibits outstanding performance for ER prediction and risk stratification. As such, this intuitive nomogram holds promise as a more effective and user-friendly tool in predicting ER for HCC patients after partial hepatectomy.

Keywords: hepatocellular carcinoma; machine learning; models; nomograms; radiomics; recurrence.

Figure 1

Flowchart of patient enrollment. HCC, hepatocellular carcinoma; CT, computed tomography; ER, early recurrence; Non-ER, non-early recurrence.

Figure 2

Schematic diagram of construction and evaluation of models.

Figure 3

Box-Scatter plots showing that the Rad-score of the ER group is substantially higher than that of the non-ER group both in the training and test sets. Red represents a high Rad-score in the ER group, and blue represents a low Rad-score in the non-ER group, indicating that the higher the Rad-score is, the more likely ER is in HCC patients (ER, early recurrence; non-ER, non-early recurrence).

Figure 4

Predictive performance of different machine learning methods. (A) Four machine learning methods and (B) heat maps comparing p-values of AUCs in the training set between groups.

Figure 5

ROC curves for predicting early recurrence with different models in the training (A) and test (B) sets; the decision curve analysis (DCA) in the training (C) and test (D) sets. ROC, receiver operating characteristic; Logistic_1, clinical model; Logistic_2, radiological model; Logistic_3, radiomics model; Logistic_4, radiomics-clinical model; Logistic_5, radiomics-radiological model; Logistic_6, combined model.

Figure 6

Development and performance of the nomogram based on the combined model in predicting the risk of ER in HCC patients after partial liver resection.

Figure 7

Risk stratification according to the nomogram. The risk of early recurrence was substantially higher in the high-risk group than in the low-risk group.

Figure 8

The ROC curves of the nomogram in the training (A) and test (B) sets; the calibration curves of the nomogram for predicting early recurrence in the training (C) and test (D) sets, which demonstrated good agreement with the ideal curve. The prediction performance improved as the solid line approached the dotted line. The decision curve analysis (DCA) of the nomogram in the training (E) and test (F) sets. The net benefit of the nomogram was higher compared to the treat-all-patients scenario (gray line) and treat-no-patients scenario (horizontal dotted line).

Figure 9

Preoperative contra-enhanced CT images and pathological pictures of two cases with hepatocellular carcinoma (HCC). (A–C) A 72-year-old man with non-early recurrence of HCC, the nomogram judged as low risk. (D–F) A 54-year-old man with early recurrence of HCC, the nomogram judged as high risk.