Predicting the invasion degree of subsolid nodule lung adenocarcinoma by artificial intelligence quantitative parameters combined with imaging signs
10.3969/j.issn.1002-1671.2025.08.010
- VernacularTitle:人工智能定量参数联合影像征象预测亚实性结节肺腺癌的浸润程度
- Author:
Kejia NING
1
;
Rui WU
1
;
Jinfeng GU
1
;
Junbo SONG
1
;
Lei MA
1
;
Huiping CAO
1
Author Information
1. 阜阳市第二人民医院CT室,安徽 阜阳 236000
- Publication Type:Journal Article
- Keywords:
subsolid nodule;
lung adenocarcinoma;
artificial intelligence;
imaging signs;
computed tomography
- From:
Journal of Practical Radiology
2025;41(8):1299-1303
- CountryChina
- Language:Chinese
-
Abstract:
Objective To predict the invasion degree of subsolid nodule(SSN)lung adenocarcinoma using a combined model incorporating artificial intelligence(AI)quantitative parameters and imaging signs,and to validate the predictive efficacy of this model.Methods A total of 281 SSN lung adenocarcinoma CT images in 243 patients were retrospectively collected and randomly divided into training set(224 cases)and validation set(57 cases)in an 8∶2 ratio,with atypical adenomatous hyperplasia(A AH)+adenocarcinoma in situ(AIS)+minimally invasive adenocarcinoma(MIA)(191 cases)as the non-invasive adenocarcinoma(I AC)group and I AC(90 cases)as the IAC group.Multivariate logistic regression analysis was performed based on the AI quantitative parameters and CT signs in the training set to obtain independent predictors of IAC.A combined model and nomogram were then constructed and validated.The diagnostic efficacy and clinical applicability of the model were evaluated by the receiver operating characteristic(ROC)curve,calibration curve,and clinical decision curve analysis(DCA).Results Multivariate logistic regression analysis of the training set showed nodule type,spicule sign,vascular abnormality,long diameter＞11.5 mm,median CT value＞—426.25 HU,and mass＞391.5 mg were independent predictors of IAC(P＜0.05).The area under the curve(AUC)of the training set model,based on these independent predictive factors,was 0.915[95％confidence interval(CI)0.875-0.954],and the AUC of the validation set model was 0.903(95％CI 0.824-0.982),indicating both the training set and validation set models had high efficacy in distinguishing IAC.The nomogram model,which quantified these independent factors,demonstrated enhanced predictive power for IAC.The calibration curve indicated good fit of the prediction model,and the clinical DCA showed the model had good clinical applicability.Conclusion The model combining AI quantitative parameters and imaging signs has a higher ability to predict the risk of IAC,compared to a single indicator.It helps clinicians in determining the appropriate surgical timing,formulating surgical methods,and reducing overtreatment.
