Abstract: SA-PO0298
Construction of a Predictive Model for AKI After Vascular Interventional Procedures Based on Imaging-Omics
Session Information
- AKI: Epidemiology and Risk Factors
October 24, 2026 | Location: Exhibit Hall A, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Acute Kidney Injury
- 101 AKI: Epidemiology, Risk Factors, and Prevention
Authors
- Guo, Zhenyu, Ningbo No.2 Hospital, Ningbo, China
- Zhou, Fangfang, Ningbo No.2 Hospital, Ningbo, China
- Luo, Qun, Ningbo No.2 Hospital, Ningbo, China
Background
This study aimed to establish and validate a radiomics-based model for predicting contrast-induced acute kidney injury (CI-AKI) after vascular intervention to achieve early risk stratification.
Methods
A single-center retrospective case-control study was performed at Ningbo No.2 Hospital from March 2020 to June 2024. Seventy-five patients with CI-AKI were enrolled as the case group, and 75 non-AKI patients were matched 1:1 by age and sex as the control group. Patients with stage 5 chronic kidney disease, renal transplantation, or malignant tumors were excluded. Preoperative clinical data and renal computed tomography images were collected and randomly divided into training and validation sets at a 7:3 ratio. Renal regions of interest (ROI) were delineated and radiomics features were extracted using software developed by Hangzhou Smart Ai Co., Ltd. Fourteen machine-learning models were constructed by combining clinical and radiomics features, and their predictive performance was assessed using the area under the receiver operating characteristic curve (AUC).
Results
(1)The AKI and non-AKI groups were comparable in demographics and baseline renal function (P>0.05) but differed significantly in smoking, medication use, and laboratory parameters (P<0.05). (2)Logistic regression showed statin use, smoking, elevated PLT, AST, and ALT were risk factors for PCI-AKI (OR 1.427-3.554, P<0.05), while spironolactone, Hb, RBC, and ALB were protective (OR 0.519-0.670, P<0.05). (3)The MNB model performed best among 14 models, with AUC 0.809 (radiomics), 0.711 (clinical), 0.796 (combined), and specificity 0.800.(4) Key radiomics features were 2D wavelet gray-level matrices; feature importance verified these and key clinical variables as main predictors.
Conclusion
The imaging-based predictive model facilitates preoperative prediction of CI-AKI occurrence post-intervention, identifies high-risk AKI patients, enables timely implementation of preventive measures, and improves patient prognosis.