Published in:
01-10-2020 | Computed Tomography | Urologic Oncology
Development and External Validation of Radiomics Approach for Nuclear Grading in Clear Cell Renal Cell Carcinoma
Authors:
Hongyu Zhou, MS, Haixia Mao, MD, Di Dong, PhD, Mengjie Fang, MS, Dongsheng Gu, MS, Xueling Liu, MD, Min Xu, MD, Shudong Yang, MD, Jian Zou, PhD, Ruohan Yin, MD, Hairong Zheng, PhD, Jie Tian, PhD, Changjie Pan, MD, Xiangming Fang, MD
Published in:
Annals of Surgical Oncology
|
Issue 10/2020
Login to get access
Abstract
Background and Purpose
Nuclear grades of clear cell renal cell carcinoma (ccRCC) are usually confirmed by invasive methods. Radiomics is a quantitative tool that uses non-invasive medical imaging for tumor diagnosis and prognosis. In this study, a radiomics approach was proposed to analyze the association between preoperative computed tomography (CT) images and nuclear grades of ccRCC.
Methods
Our dataset included 320 ccRCC patients from two centers and was divided into a training set (n = 124), an internal test set (n = 123), and an external test set (n = 73). A radiomic feature set was extracted from unenhanced, corticomedullary phase, and nephrographic phase CT images. The maximizing independent classification information criteria function and recursive feature elimination with cross-validation were used to select effective features. Random forests were used to build a final model for predicting nuclear grades, and area under the receiver operating characteristic curve (AUC) was used to evaluate the performance of radiomic features and models.
Results
The radiomic features from the three CT phases could effectively distinguished the four nuclear grades. A combined model, merging radiomic features and clinical characteristics, obtained good predictive performances in the internal test set (AUC 0.77, 0.75, 0.79, and 0.85 for the four grades, respectively), and performance was further confirmed in the external test set, with AUCs of 0.75, 0.68, and 0.73 (no fourth-level data).
Conclusion
The combination of CT radiomic features and clinical characteristics could discriminate the nuclear grades in ccRCC, which may help in assisting treatment decision making.