Construction and validation of a model for differentiating adrenal metastatic nodules from lung cancer based on plain CT radiomics combined with clinicopathological features

ZHANG Dejiang; CAO Lixiu; LIN Tao

doi:10.3969/j.issn.1009-0460.2026.01.005

Abstract：Objective To develop and validate a combined model integrating clinical,pathological,and plain CT radiomics features for differentiating adrenal metastatic nodules in lung cancer patients.Methods A total of 364 patients admitted to Tangshan People's Hospital from February 2013 to December 2023 were enrolled.Using a 7∶3 random allocation method,they were divided into a training set(254 cases,comprising 124 benign nodules and 130 metastatic nodules)and a test set(110 cases,comprising 54 benign nodules and 56 metastatic nodules).A comparative analysis of clinicopathological characteristics,including sex,age,clinical stage of lung cancer,and histological type,was conducted between patients in the training and test sets.Based on non-enhanced CT images,1316 radiomics features of adrenal indeterminate nodules were extracted.The least absolute shrinkage and selection operator(LASSO)was used to screen statistically significant features,which were then employed to construct a radiomics score(Rad-score).Subsequently,multivariate Logistic regression analysis was further applied to identify independent risk factors for adrenal metastatic nodules in lung cancer.By integrating clinical,pathological,and radiomics information,a combined diagnostic model was developed and visualized as a nomogram.The differential diagnostic performance and calibration accuracy of this nomogram were evaluated using the area under the receiver operating characteristic curve(AUC)and calibration curves,respectively.Results The clinicopathological analysis revealed statistically significant differences in gender,clinical stage of lung cancer,and histologic type distribution between metastatic and benign nodules(all P＜0.05),whereas no statistically significant difference was observed in age(P＞0.05).Specifically,metastatic nodules showed significantly higher proportions in male gender(training set:61.54％vs.43.55％;test set:64.29％vs.37.04％),advanced stage(Ⅲ/Ⅳ)(training set:92.31％vs.51.61％;test set:92.86％vs.40.74％),and small cell lung cancer subtype(training set:34.62％vs.9.68％;test set:32.14％vs.11.11％)compared to benign nodules.Following LASSO screening and collinearity exclusion,four independent features,including original_shape_Flatness,wavelet-LLH_glcm_Imc1,wavelet-LLH_glcm_InverseVariance,and wavelet-LLL_firstorder_Maximum,were incorporated to construct the Rad-score.The Rad-score was significantly higher in the metastatic nodule group than in the benign nodule group(training set:0.746 vs.0.233;test set:0.796 vs.0.245,both P＜0.00 1).In univariate analysis,Rad-score demonstrated the best discriminative performance with an AUC of 0.897.The combined diagnostic model,constructed based on gender,clinical stage of lung cancer,and Rad-score,achieved AUCs of 0.918(95％CI:0.882-0.954)in the training set and 0.910(95％CI:0.857-0.963)in the test set.Sensitivity,specificity,and accuracy were all superior to those of any single variable(all P＜0.05).Using a nomogram cutoff score of 96.30 points,the model exhibited well-fitted calibration curves in both cohorts(P=0.240,0.563),demonstrating good agreement between the predicted results and pathological outcomes.Conclusion The combined clinical-pathological-radiomics diagnostic model,constructed based on Rad-score from non-contrast CT,gender,and clinical stage of lung cancer,can effectively differentiate adrenal metastatic nodules,contributing to precise staging and individualized treatment.