CT-based Radiomics Predicting The Good Pathological Response For Neoadjuvant Immunotherapy In Non-small Cell Lung Cancer

OBJECTIVE: In radiomics,high-throughput algorithms are used to extract objective quantitative features from medical images.In this study,we evaluated CT-based radiomics features,clinical features,deep learning features,and a combination of features for predicting good pathological response(GPR)to neoadjuvant immunotherapy in patients with non-small cell lung cancer(NSCLC).Methods and materials: We reviewed 62 patients with NSCLC who received surgery after neoadjuvant immunotherapy and collected clinicopathological data and CT images both before and after neoadjuvant immunotherapy.A series of image preprocessing was carried out on CT scanning images,namely tumor segmentation,traditional radiomics feature extraction,deep learning feature extraction,and normalization.Spearman correlation coefficient(Spearman),Principal Component Analysis(PCA),Least absolute shrinkage,and selection operator(Lasso)were used to screen features.According to the data collected before treatment,the pre-treatment traditional radiomics combined with clinical characteristics(before＿rad＿cil)model and pre-treatment deep learning characteristics(before＿dl)model were constructed;The data collected after treatment constructed the after＿rad＿cil model and after＿dl model;The total model jointly constructed by all clinical features,traditional radiomics features and deep learning features before and after treatment.Finally,according to the data obtained before and after treatment,the before＿nomogram and after＿nomogram were constructed.Results:(1).In before＿rad＿cil model,four traditional radiomics features("original＿shape＿flatness","wavelet hhl＿firer＿skewness","wavelet hlh＿firer＿skewness","wavelet lll＿glcm＿correlation")and two clinical features("gender","N stage")were screened out to predict GPR.The average prediction accuracy(ACC)after modeling with k-Nearest Neighbor(KNN)was 0.707.(2).In after＿rad＿cil model,nine features predictive of GPR were obtained after feature screening,among which seven were traditional radiomics features: "exponential＿firer＿skewness","exponential＿glrlm＿runentropy","log-sigma-5-0-mm-3d＿firer＿kurtosis","logarithm＿skewness","original＿shape＿elongation","original＿shape＿brilli ance","wavelet＿llh＿glcm＿clustershade";Two were clinical features: "after＿crp","after lymphocyte percentage";The ACC after modeling with Support Vector Machine(SVM)was 0.682.(3).Before＿dl model and after＿dl model were modeled by SVM,and the ACC was 0.629 and 0.603 respectively.(4).After feature screening,the total model was constructed by Multilayer Perceptron(MLP),and the average prediction accuracy of GPR was the highest,which was 0.805.(5).The Calibration curve showed that the predictions of GPR by before＿nomogram and after＿nomogram were in good agreement with the actual GPR.Conclusion:(1).CT-based radiomics has a good predictive ability of the GPR for neoadjuvant immunotherapy in non-small cell lung cancer.(2).Among the radiomics features combined clinicopathological information model,deep learning features model,and the total model,the total model has the highest prediction efficiency.(3).Nomogram based on the radiomics label is helpful to the individualized pathological response evaluation for neoadjuvant immunotherapy in NSCLC.