Incremental value of radiomics with machine learning to the existing prognostic models for predicting outcome in renal cell carcinoma

Jiajun Xing; Yiyang Liu; Zhongyuan Wang; Aiming Xu; Shifeng Su; Sipeng Shen; Zengjun Wang

doi:10.3389/fonc.2023.1036734

Incremental value of radiomics with machine learning to the existing prognostic models for predicting outcome in renal cell carcinoma

Front Oncol. 2023 Apr 28:13:1036734. doi: 10.3389/fonc.2023.1036734. eCollection 2023.

Authors

Jiajun Xing¹, Yiyang Liu¹, Zhongyuan Wang¹, Aiming Xu¹, Shifeng Su¹, Sipeng Shen², Zengjun Wang¹

Affiliations

¹ Department of Urology, The First Affiliated Hospital of Nanjing Medical University, Nanjing, China.
² Department of Biostatistics, School of Public Health, Nanjing Medical University, Nanjing, Jiangsu, China.

Abstract

Purpose: To systematically evaluate the potential of radiomics coupled with machine-learning algorithms to improve the predictive power for overall survival (OS) of renal cell carcinoma (RCC).

Methods: A total of 689 RCC patients (281 in the training cohort, 225 in the validation cohort 1 and 183 in the validation cohort 2) who underwent preoperative contrast-enhanced CT and surgical treatment were recruited from three independent databases and one institution. 851 radiomics features were screened using machine-learning algorithm, including Random Forest and Lasso-COX Regression, to establish radiomics signature. The clinical and radiomics nomogram were built by multivariate COX regression. The models were further assessed by Time-dependent receiver operator characteristic, concordance index, calibration curve, clinical impact curve and decision curve analysis.

Result: The radiomics signature comprised 11 prognosis-related features and was significantly correlated with OS in the training and two validation cohorts (Hazard Ratios: 2.718 (2.246,3.291)). Based on radiomics signature, WHOISUP, SSIGN, TNM Stage and clinical score, the radiomics nomogram has been developed. Compared with the existing prognostic models, the AUCs of 5 years OS prediction of the radiomics nomogram were superior to the TNM, WHOISUP and SSIGN model in the training cohort (0.841 vs 0.734, 0.707, 0.644) and validation cohort2 (0.917 vs 0.707, 0.773, 0.771). Stratification analysis suggested that the sensitivity of some drugs and pathways in cancer were observed different for RCC patients with high-and low-radiomics scores.

Conclusion: This study showed the application of contrast-enhanced CT-based radiomics in RCC patients, creating novel radiomics nomogram that could be used to predict OS. Radiomics provided incremental prognostic value to the existing models and significantly improved the predictive power. The radiomics nomogram might be helpful for clinicians to evaluate the benefit of surgery or adjuvant therapy and make individualized therapeutic regimens for patients with renal cell carcinoma.

Keywords: computed tomography running title; incremental value estimation; machine learning; prognostic model; radiomics; renal cell carcinoma.

Grants and funding

This study was supported by by National Natural Science Foundation of China (No. 81771640 to ZJW, NO.82103946 to SS) and Natural Science Foundation of the Jiangsu Higher Education Institutions of China (NO.21KJB330004 to SS).