Objectives: The measurement of serum creatinine and estimated glomerular filtration rate (GFR) is a gold standard for the assessment of renal function in clinical practice. However, in usual clinical settings, serum creatinine is measured in non-steady-state conditions. Physicians often depend on limited information such as kidney size and past medical history in the differential diagnosis of acute kidney injury and chronic kidney injury. Radiomics is a promising potential approach for quantitative analysis of various medical images for the diagnosis and prognosis prediction of diseases, combined with recent machine learning and deep learning algorithms. In this study, we aimed quantitative analysis of CT-based radiomic feature analysis to assess baseline renal function.  Methods: A total of 2,961 patients who underwent a noncontrast CT scan of the abdomen from 2015 to 2018 were initially screened from the EHR data. After excluding low-quality CT images or individuals without baseline creatinine values, 489 patients were included in the main analysis. Kidney segmentation was performed using semi-automated tools. A hundred types of radiomics features were extracted with segmented CT images using the PyRadiomics package.  Results: Voxel volume, first-order energy, and grayscale non-uniformity were the most highly correlated radiomic features with GFR (correlation coefficient >0.6), and surface volume ratio was the most negatively correlated variable (correlation coefficient <-0.67). The Prediction for GFR <60ml/min/1.73m2 with selected radiomic features showed relatively high accuracy (0.88 in the XGBoost model).  Conclusions: Our findings suggest the potential utility of radiomic feature analysis to assess baseline renal function and CKD stages.