低剂量CT特征及组织NRF2、Ki-67表达对低分化浸润性非黏液肺腺癌的预测价值OA

Objective To construct a machine learning prediction model for poorly differentiated invasive non-mucinous lung adenocarcinoma(INMA)based on the quantitative and qualitative features of low-dose CT(LDCT)and the expression of NRF2 and Ki-67 in tissues,and to verify its efficacy.Methods The data of 196 INMA patients from January 2023 to January 2025 in the 73rd Army Group Hospital of PLA Army were retrospectively collected as the training set.Another 85 INMA patients'data from February 2025 to July 2025 were selected as the test set at a ratio of 7∶3.The training set was divided into high/moderately differentiated(n=146)and poorly differentiated(n=50)groups according to the differentiation degree of the lesions.Univariate analysis,LASSO re-gression,and multivariate logistic regression were conducted on the variables,and the independent predictive factors were selected to establish machine learning prediction models of logistic regression(LR),random forest(RF),extreme gradient boosting(XGBoost),and support vector machine(SVM).The predictive value of the models constructed by the four methods was compared using the re-ceiver operating characteristic(ROC)curve.Results The independent predictors selected by univariate,LASSO regression and multi-variate Logistic included mean nodule diameter,nodule volume,mean nodule density,NRF2 expression,and Ki-67 index.The areas under the curve(AUC)of the LR,SVM,RF,and XGBoost machine learning models in the training set were 0.958,0.952,0.977,and 0.987,respectively.The XGBoost model demonstrated the highest accuracy 94.39%,sensitivity 96.00%,and AUC,outperfor-ming other models.Conclusions The XGBoost model based on the quantitative and qualitative features of low-dose CT and the expres-sion of NRF2 and Ki-67 in tissues can accurately predict the risk of developing poorly differentiated INMA,providing evidence-based basis for early identification and targeted intervention in high-risk patients.