Rationale and objectives: This study aimed to develop a diagnostic model based on clinical and CT features for identifying clear cell renal cell carcinoma (ccRCC) in small renal masses (SRMs).
Material and methods: This retrospective multi-centre study enroled patients with pathologically confirmed SRMs. Data from three centres were used as training set (n = 229), with data from one centre serving as an independent test set (n = 81). Univariate and multivariate logistic regression analyses were utilised to screen independent risk factors for ccRCC and build the classification and regression tree (CART) diagnostic model. The area under the curve (AUC) was used to evaluate the performance of the model. To demonstrate the clinical utility of the model, three radiologists were asked to diagnose the SRMs in the test set based on professional experience and re-evaluated with the aid of the CART model.
Results: There were 310 SRMs in 309 patients and 71% (220/310) were ccRCC. In the testing cohort, the AUC of the CART model was 0.90 (95% CI: 0.81, 0.97). For the radiologists' assessment, the AUC of the three radiologists based on the clinical experience were 0.78 (95% CI:0.66,0.89), 0.65 (95% CI:0.53,0.76), and 0.68 (95% CI:0.57,0.79). With the CART model support, the AUC of the three radiologists were 0.93 (95% CI:0.86,0.97), 0.87 (95% CI:0.78,0.95) and 0.87 (95% CI:0.78,0.95). Interobserver agreement was improved with the CART model aids (0.323 vs 0.654, P < 0.001).
Conclusion: The CART model can identify ccRCC with better diagnostic efficacy than that of experienced radiologists and improve diagnostic performance, potentially reducing the number of unnecessary biopsies.
Keywords: Clear Cell Renal Cell Carcinoma; Kidney Neoplasms; Machine Learning; Multidetector Computed Tomography.
Copyright © 2024 The Association of University Radiologists. Published by Elsevier Inc. All rights reserved.