MRI radiomics-based nomogram for individualised prediction of synchronous distant metastasis in patients with clear cell renal cell carcinoma

Objective To evaluate the performance of a multiparametric MRI radiomics-based nomogram for the individualised prediction of synchronous distant metastasis (SDM) in patients with clear cell renal cell carcinoma (ccRCC). Methods Two-hundred and one patients (training cohort:n= 126; internal validation cohort:n= 39; external validation cohort:n= 36) with ccRCC were retrospectively enrolled between January 2013 and June 2019. In the training cohort, the optimal MRI radiomics features were selected and combined to calculate the radiomics score (Rad-score). Incorporating Rad-score and SDM-related clinicoradiologic characteristics, the radiomics-based nomogram was established by multivariable logistic regression analysis, then the performance of the nomogram (discrimination and clinical usefulness) was evaluated and validated subsequently. Moreover, the prediction efficacy for SDM in ccRCC subgroups of different sizes was also assessed. Results Incorporating Rad-score derived from 9 optimal MR radiomics features (age, pseudocapsule and regional lymph node), the radiomics-based nomogram was capable of predicting SDM in the training cohort (area under the ROC curve (AUC) = 0.914) and validated in both the internal and external cohorts (AUC = 0.854 and 0.816, respectively) and also showed a convincing predictive power in ccRCC subgroups of different sizes (<= 4 cm, AUC = 0.875; 4-7 cm, AUC = 0.891; 7-10 cm, 0.908; > 10 cm, AUC = 0.881). Decision curve analysis indicated that the radiomics-based nomogram is of clinical usefulness. Conclusions The multiparametric MRI radiomics-based nomogram could achieve precise individualised prediction of SDM in patients with ccRCC, potentially improving the management of ccRCC.

Automated summary

The study evaluated the performance of a multiparametric MRI radiomics-based nomogram for individualized prediction of SDM in ccRCC patients, showing good predictive ability in the training cohort and internal and external validation cohorts, as well as effective prediction in different sizes of ccRCC subgroups.