Prediction Of Micropapillary Components In Lung Invasive Adenocarcinoma Based On CT Radiomics

IntroductionLung cancer is the most common cancer in the world,the highest cause of cancer death.Studies on the pathological types of lung cancer show that adenocarcinoma has replaced squamous cell carcinoma as the most common pathological subtype in the world.Lung adenocarcinoma with micropapillary components has been reported to have a poor prognosis,and preoperative knowledge of the presence of micropapillary components may be important for the development of personalized surgical plans for patients.Few studies are based on CT Radiomics methods to identify micropapillary components.We aim to explore an effective method for predicting micropapillary components in lung adenocarcinoma before operation.MethodsWe enrolled 236 patients pathologically confirmed with lung invasive adenocarcinoma,including 113 patients with micropapillary components,who underwent surgical resection in our hospital between January 2017 and May 2020.The DICOM images of thin slice CT plain scan of each patient’s lung were imported into 3D Slicer 4.10.2 software,and the region of interest(ROI)of the tumor was manually delineated by the physician for 3D reconstruction.Radiomics was used to extract the gray level co-occurrence matrix,shape,first-order statistics and other omics features for quantitative CT analysis.We use the Least Absolute Shrinkage and Selection Operator(Lasso)regression model and the multivariate logistic regression equation to develop the prediction model—nomogram.The evaluation of the model used the c-index,calibration curve,receiver operating characteristic(ROC)to know its clinical validity and practicability,the accuracy of the prediction model was further analyzed by internal validation.ResultsIn this study,the model has established a total of five characteristics which namely difference variance(OR:1.223,95% CI:1.125～1.329)and difference entropy value(OR:0.046,95% CI:0.011～0.198)based on GLCM,maximum diameter(OR: 3.197,95% CI: 1.088 ～ 9.397)and surface volume ratio based on shape(OR: 0.103,95% CI: 0.023 ～ 0.473),and minimum of whole pixel value based on first-order statistics(OR:0.997,95% CI:0.996～0.999),and they were determined to be predictive of the presence of micropapillary components in lung invasive adenocarcinoma.Based on the results,we further constructed the nomogram and successfully presented a visual prediction model.The model shows excellent accuracy and calibration performance.The area under curve(AUC)is up to 0.829(95% CI: 0.775～0.882),which can still reach the high C-index value of0.814 in internal verification.The calibration curve shows good agreement between the actual observations and the predicted results.ConclusionRadiomics can be used to characterize the entire tumor noninvasively.We used 3D Slicer 4.10.2 software to segment and reconstruct tumors,extracted image features with Radiomics analysis,and successfully constructed a nomogram model to predict the existence of micropapillary components in lung adenocarcinoma,which can influence the formulation of preoperative surgical plan and guide physicians to make better decisions in clinical practice.