Multiparameter computed tomography (CT) radiomics signature fusion-based model for the preoperative prediction of clear cell renal cell carcinoma nuclear grade: a multicenter development and external validation study

Abstract

Background: The preoperative prediction of the pathological nuclear grade of clear cell renal cell carcinoma (CCRCC) is crucial for clinical decision making. However, radiomics features from one or two computed tomography (CT) phases are required to predict the CCRCC grade, which reduces the predictive performance and generalizability of this method. We aimed to develop and externally validate a multiparameter CT radiomics-based model for predicting the World Health Organization/International Society of Urological Pathology (WHO/ISUP) grade of CCRCC.

Methods: A total of 500 CCRCC patients at The First, Second, and Yongchuan Hospitals of Chongqing Medical University between January 2016 and May 2022 were retrospectively enrolled in this study. The patients were divided into the training set (n=268), internal testing set (n=115), and two external testing sets (testing set 1, n=62; testing set 2, n=55). Radiomics features were extracted from multi-phase CT images, and radiomics signatures (RSs) were created by least absolute shrinkage and selection operator (LASSO) regression. In addition, a clinical model was developed. A combined model was also established that integrated the RSs with the clinical factors, and was visualized via a nomogram. The performance of the established model was assessed using area under the curve (AUC) values, a calibration curve analysis, and a decision curve analysis (DCA).

Results: Among the four RSs and the clinical model, the RS-Triphasic had the best predictive performance with AUCs of 0.88 [95% confidence interval (CI): 0.85-0.91] and 0.84 (95% CI: 0.74-0.95) in the training and testing sets, respectively, and 0.82 (95% CI: 0.72-0.93) and 0.82 (95% CI: 0.71-0.93) in external testing sets 1 and 2. Integrating the RS-Triphasic, RS-corticomedullary phase (CMP), RS-nephrographic phase (NP), RS-non-contrast phase (NCP) with the clinical risk factors, a combined model was established with AUCs of 0.92 (95% CI: 0.89-0.94), 0.86 (95% CI: 0.76-0.95), 0.84 (95% CI: 0.73-0.95), and 0.82 (95% CI: 0.70-0.94) for the training, internal testing, and external testing sets 1 and 2, respectively. The DCA indicated that the nomogram had a greater overall net benefit than the clinical and radiomics models.

Conclusions: The multiparameter CT RS fusion-based model had high accuracy in differentiating between high- and low-grade CCRCC preoperatively. Thus, it has great potential as a useful tool for personalized treatment planning and clinical decision making for CCRCC patients.

Keywords: Computed tomography (CT); clear cell renal cell carcinoma (CCRCC); multiparameter fusion radiomics model; pathological nuclear grade; radiomics.

Figure 1

Flowchart of patient recruitment. Center 1 (FAHCQMU), The First Affiliated Hospital of Chongqing Medical University; Center 2 (SAHCQMU), The Second Affiliated Hospital of Chongqing Medical University; Center 3 (YCHCQMU), Yongchuan Hospital of Chongqing Medical University; CCRCC, clear cell renal cell carcinoma; CT, computed tomography.

Figure 2

The overall workflow of the feature selection and model building procedures. NCP, non-contrast phase; NP, nephrographic phase; CMP, corticomedullary phase; CT, computed tomography; ROIs, regions of interest; 2D, two-dimensional; 3D, three-dimensional; RS, radiomics signature; BMI, body mass index; WHO/ISUP, World Health Organization/International Society of Urological Pathology; ROC, receiver operator characteristic; DCA, decision curve analysis.

Figure 3

The radiomics feature selection results using the LASSO algorithm. Based on minimum criteria, LASSO regression selected 16, 14, 11, and 25 radiomics features from the (A) CMP, (D) NP, (G) NCP, and (J) Triphasic CT images, respectively, with Tuning parameter (λ) values of 0.0309, 0.0362, 0.0489, and 0.0309, respectively. The coefficient profile plots of the identified non-zero coefficients for (B) CMP, (E) NP, (H) NCP, and (K) Triphasic radiomics features were generated against the selected log λ values. The names and corresponding weighting coefficients of the selected (C) CMP, (F) NP, (I) NCP, and (L) Triphasic radiomics features. LASSO, least absolute shrinkage and selection operator; CMP, corticomedullary phase; NP, nephrographic phase; NCP, non-contrast phase; CT, computed tomography.

Figure 4

The ROC curves of the (A) RS-Triphasic, (B) RS-CMP, (C) RS-NP, and (D) RS-NCP in the training, internal testing, and external testing sets 1 and 2. AUC, area under the curve; ROC, receiver operator characteristic; RS, radiomics signature; CMP, corticomedullary phase; NP, nephrographic phase; NCP, non-contrast phase.

Figure 5

The establishment and performance evaluation of the nomogram. (A) The multivariate logistic regression indicated that RS-Triphasic, RS-CMP, RS-NP, RS-NCP, age, and intratumoral necrosis were independent risk factors for patients with high-grade CCRCC. A combined model was then constructed and visualized via a nomogram. (B) The ROC analysis of the combined model. (C) The DCA result showed that the combined model provides more benefits to patients with CCRCC across most thresholds. The calibration curve analysis of the combined model in the (D) training, (E) internal testing, and external testing sets (F) 1 and (G) 2. AUC, area under the curve; RS, radiomics signature; CMP, corticomedullary phase; NP, nephrographic phase; NCP, non-contrast phase; CCRCC, clear cell renal cell carcinoma; ROC, receiver operator characteristic; DCA, decision curve analysis.