Deep learning model for non-contrast CT predicting contrast medium extravasation in patients with tumors prior to contrast-enhanced CT
10.3969/j.issn.1002-1671.2025.10.028
- VernacularTitle:平扫CT深度学习模型预测肿瘤患者增强CT检查前对比剂外渗
- Author:
Lili HU
1
;
Xiaofei WU
;
Ying ZHANG
;
Shudong HU
;
Ling HANG
;
Yuxi GE
Author Information
1. 江南大学附属医院放射科,江苏 无锡 214122
- Publication Type:Journal Article
- Keywords:
contrast medium extravasation;
deep learning;
computed tomography;
vascular segmentation;
feature extraction
- From:
Journal of Practical Radiology
2025;41(10):1723-1728
- CountryChina
- Language:Chinese
-
Abstract:
Objective To investigate the potential value of a deep learning(DL)model based on non-contrast CT images in predicting contrast medium extravasation in contrast-enhanced CT scans of tumor patients.Methods A total of 298 tumor patients were retrospectively selected,including 90 patients with extravasation and 208 without extravasation,and divided into training set(207 patients),validation set(46 patients),and external test set(45 patients)in a ratio of 7︰1.5︰1.5.U-Net was employed to segment the right common carotid artery/internal jugular vein and right subclavian artery/vein in non-contrast CT images,and ResNet50 was utilized to extract imaging features to construct the DL model,which was subsequently integrated with independent clinical predictors to establish the combined model.The segmentation performance of the DL model was evaluated using Dice similarity coefficient(DSC)and Intersection over Union(IoU),while the area under the curve(AUC),accuracy,sensitivity,and specificity of the model were calculated.Results The DL model demonstrated superior vascular segmentation(DSC 0.81-0.95,IoU 0.79-0.90).The combined model achieved optimal predictive performance,with AUC of 0.961[95%confidence interval(CI)0.924-0.983],0.949(95%CI 0.840-0.992),and 0.891(95%CI 0.762-0.964)in the training,validation,and external test sets,respectively.Its accuracy,sensitivity,and specificity were consistently higher than those of the standalone clinical model.Conclusion The DL model based on non-contrast CT images shows significant potential value in predicting contrast medium extravasation risk in tumor patients,providing an objective and intelligent tool for clinical risk assessment.
