Prediction models for clear cell renal cell carcinoma ISUP/WHO grade: comparison between CT radiomics and conventional contrast-enhanced CT

Abstract

Objective: Comparing the prediction models for the ISUP/WHO grade of clear cell renal cell carcinoma (ccRCC) based on CT radiomics and conventional contrast-enhanced CT (CECT).

Methods: The corticomedullary phase images of 119 cases of low-grade (I and II) and high-grade (III and IV) ccRCC based on 2016 ISUP/WHO pathological grading criteria were analyzed retrospectively. The patients were randomly divided into training and validation set by stratified sampling according to 7:3 ratio. Prediction models of ccRCC differentiation were constructed using CT radiomics and conventional CECT findings in the training setandwere validated using validation set. The discrimination, calibration, net reclassification index (NRI) and integrated discrimination improvement index (IDI) of the two prediction models were further compared. The decision curve was used to analyze the net benefit of patients under different probability thresholds of the two models.

Results: In the training set, the C-statistics of radiomics prediction model was statistically higher than that of CECT (p < 0.05), with NRI of 9.52% and IDI of 21.6%, both with statistical significance (p < 0.01).In the validation set, the C-statistics of radiomics prediction model was also higher but did not show statistical significance (p = 0.07). The NRI and IDI was 14.29 and 33.7%, respectively, both statistically significant (p < 0.01). Validation set decision curve analysis showed the net benefit improvement of CT radiomics prediction model in the range of 3-81% over CECT.

Conclusion: The prediction model using CT radiomics in corticomedullary phase is more effective for ccRCC ISUP/WHO grade than conventional CECT.

Advances in knowledge: As a non-invasive analysis method, radiomics can predict the ISUP/WHO grade of ccRCC more effectively than traditional enhanced CT.

Figure 1.

Flowchart for ccRCC patient enrollment. ccRCC, clear cell renal cell carcinoma

Figure 2.

Heatmap of radiomics for visualizing the distribution difference of the low- and high-grade ccRCC in training set and validation set. ccRCC, clear cell renal cell carcinoma.

Figure 3.

LASSO regression cross-validation diagram and regression coefficient diagram, the upper horizontal axis is the number of radiomics features corresponding to the models. The two vertical dashed lines in Figure 3A show the two log (λ) values for minimum mean-squared error minimum and the increase of 1 SD (one standard deviation) mean-squared error minimum determined by cross-validation. Figure 3B shows that with the increase of log (λ), the radiomics features coefficients were gradually compressed to 0, and the number of features was reduced to 17 by the log (λ) with minimum mean-squared error minimum. LASSO, least absolute shrinkage and selection operator.

Figure 4.

(A) ROC of the two prediction models in training set, (B) ROC of the two prediction models in validation set. CECT, contrast-enhanced CT; ROC, receiver operating characteristic.

Figure 5.

Decision curve analysis using the validation set. CECT, contrast-enhanced CT.