基于治疗前PET/CT影像组学特征联合临床指标的肺腺癌发生形态正常代谢增高淋巴结转移预测模型OA

Objective To construct and evaluate a prediction model for morphologically normal but hypermetabolic lymph node metastasis in lung adenocarcinoma based on pre-treatment PET/CT radiomics features combined with clinical indicators.Methods A total of 116 patients with lung adenocarcinoma who had pathologically confirmed morphologically normal but metabolically increased hilar or ipsilateral mediastinal lymph nodes were included.Patients were divided into the training set(82 cases)and test set(34 cases)using stratified random sampling in a 7∶3 ratio.Based on postoperative pathological results of the lymph nodes,the 82 patients in the training set were classified into the metastasis group(39 ca-ses)and non-metastasis group(43 cases).Radiomics features were extracted from positron emission tomography/computed tomography(PET/CT)images of the primary tumor.LASSO regression was used to select key features,and a radiomics score(Rad-score)formula was established to construct a radiomics prediction model.Univariate and multivariate Logistic regression analyses were performed on clinical variables to identify independent influencing factors and to construct a clini-cal prediction model.Independent clinical variables and the Rad-score were included as independent variables in a multiva-riate binary Logistic regression model to construct a combined prediction model.Model performance was evaluated using re-ceiver operating characteristic(ROC)curves,calibration curves,and decision curve analysis.Results The radiomics prediction model was formulated as:Logit(Pradiomics)=-0.537+1.519×Rad-score.The clinical prediction model was formulated as:Logit(Pclinical)=-1.094+1.209×maximum tumor diameter(≥3 cm)+1.270×CEA(≥5 ng/mL).The combined prediction model was formulated as:Logit(Pcombined)=-0.688+1.487×Rad-score+0.601×maximum tumor diameter(≥3 cm)+0.735×CEA(≥5 ng/mL).Both maximum tumor diameter and CEA were binary variables(yes=1,no=0).The area under the curve(AUC)values of the radiomics prediction model and the combined prediction model were significantly higher than those of the clinical prediction model in both the training set and the test set(all P<0.05).The calibration curve of the combined prediction model showed good agreement between predicted probabilities and actual observed probabilities.The decision curve of the combined prediction model indicated a high net clinical benefit.Conclusion A prediction model for morphologically normal but hypermetabolic lymph node metastasis in lung adenocarci-noma was successfully constructed based on pre-treatment PET/CT radiomics features and clinical indicators,demonstrating satisfactory predictive performance.