Application value of machine learning models based on CT radiomics for assessing split renal function
10.13491/j.issn.1004-714X.2025.01.018
- VernacularTitle:基于CT影像组学的机器学习模型评估分肾功能的应用价值
- Author:
Junjie ZOU
1
;
Ruidong LI
1
;
Hu SONG
2
;
Feng WANG
2
;
Ning DING
3
;
Kongyuan ZHANG
4
Author Information
1. School of Medical Imaging, Shandong Second Medical University, Weifang 261053 China.
2. Department of Radiology, Weifang No.2 People's Hospital, Weifang 261041 China.
3. Medical Imaging Center, Affiliated Hospital of Shandong Second Medical University, Weifang 261031 China.
4. Interventional Catheterization Center, Weifang People's Hospital, Weifang 261041 China.
- Publication Type:OriginalArticles
- Keywords:
Split renal function;
Radiomics;
Machine learning;
Computed tomography;
Glomerular filtration rate
- From:
Chinese Journal of Radiological Health
2025;34(1):108-113
- CountryChina
- Language:Chinese
-
Abstract:
Objective Based on the radiomics features extracted from the unenhanced CT images of the lower abdomen, a variety of machine learning models were constructed to explore their application value in the assessment of split renal function. Methods A retrospective analysis was conducted on the unenhanced CT images from 240 single kidneys in patients with clinically suspected renal dysfunction. Based on the results of single-photon emission computed tomography renal dynamic imaging, the cases were classified into the normal glomerular filtration rate group (n=118) and the decreased glomerular filtration rate group (n=122). The region of interest was outlined on the unenhanced CT images and the radiomics features were extracted. The features were selected by correlation analysis and least absolute shrinkage and selection operator, and the machine learning models were constructed based on the algorithms of decision tree, support vector machine, random forest, logistic regression, and extreme gradient boosting. Area under the receiver operating characteristic curve, accuracy, sensitivity, and specificity were calculated to compare the performance of different models. Results Sixteen radiomics features were selected for constructing the machine learning models. The support vector machine model showed relatively high performance for the assessment of split renal function on the test set, with an area under the receiver operating characteristic curve value of 0.883 (95% confidence interval: 0.804-0.961), an accuracy of 0.778, a sensitivity of 0.811, and a specificity of 0.743. Conclusion The machine learning models constructed based on unenhanced CT radiomics can be used to preliminarily assess split renal function, which provides an innovative, convenient, and safe method for clinical diagnosis and has positive significance for treatment.
