.png){kind=icon}
.png){kind=icon}
Return
Delta radiomics for predicting early radiation-induced lung injury after thoracic radiotherapy: a retrospective paired-CT study
.png){kind=icon}
OA
AI
N
X
Y
X
T
H
N
L
DOI：10.1007/s00066-026-02562-9.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Early radiation-induced lung injury remains a clinically relevant complication after thoracic radiotherapy. We compared pretreatment, posttreatment, delta radiomics, and combined models based on paired CT scans for early prediction of lung injury. This retrospective study included 82 patients with paired CT scans acquired before and after thoracic radiotherapy. The cohort was divided into a training set (n = 57) and an independent test set (n = 25). The endpoint was grade 1 or higher radiation-induced lung injury within 3 months after radiotherapy according to Common Terminology Criteria for Adverse Events Version 5.0 (CTCAE V5.0). Delta radiomics features were defined as posttreatment minus pretreatment values. Five signatures were constructed: clinical, pre, post, delta, and combined. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration analysis, DeLong testing, and decision curve analysis (DCA). Only nodal (N) classification remained significant in multivariable analysis (OR 1.189, 95% CI 1.044–1.354, P = 0.030). In the test cohort, the post signature outperformed the pre signature (AUC 0.844 vs. 0.681). The delta signature achieved the best performance among the radiomics-only models, with a test AUC of 0.882, accuracy of 0.840, sensitivity of 0.889, and specificity of 0.812. The combined signature achieved the highest overall test AUC at 0.889, but the gain over the delta model was limited. Posttreatment and delta radiomics outperformed pretreatment radiomics for predicting early radiation-induced lung injury after thoracic radiotherapy. The temporal imaging change captured by paired CT may provide the most informative signal for early risk stratification.
Keywords:
Radiation pneumonitis
Machine learning
Pulmonary toxicity
Temporal imaging
Model validation
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
2.5
Papers:
270
Citations:
3.4K
Organization
Cited Papers