Value Of ¹⁸F-FDG PET/CT Radiomics In Predicting The Prognosis Of Non-small Cell Lung Cancer

Objective:With the popularization of clinical application of ¹⁸F-FDG PET/CT in non-small cell lung cancer（NSCLC）,the research on radiomics features based on PET/CT has gradually increased,but at present there are few reports on predicting the survival and prognosis of patients with non-small cell lung cancer at home and abroad.The aim of this study is to investigate the value of baseline ¹⁸F-FDG PET/CT radiomics features in predicting the prognosis of NSCLC before treatment,to construct and verify radiomics model,clinical model and complex model which integrated the two,and further develop nomogram based on the best prediction model.Thus,the radiomics features of pre-treatment ¹⁸F-FDG PET/CT imaging were used to judge the prognosis of patients with non-small cell lung cancer.Methods:The clinical data of 300 patients with solitary non-small cell lung cancer who underwent ¹⁸F-FDG PET/CT imaging before treatment in the fourth Hospital of Hebei Medical University between January 2016 and August 2018 were retrospectively analyzed.The primary lesions were confirmed by pathology.According to the proportion of 7:3,they were randomly divided into training group and verification group.Using the imaging software LIFEx 6.20,using 40%SUVmax of the region of interest（ROI）of the primary lung lesions as the threshold,was delineated layer by layer on PET/CT images,and the radiomics feautures of PET and CT were extracted respectively.In the training group,using the Least Absolute Shrinkage and Selection Operator（LASSO）algorithm,combined with Cox proportional hazard regression,the extracted radiomics features and included clinical features were selected out the best features.According to the weight coefficient,the formula was established and the imaging score of each patient was calculated,which was used to construct the radiomics and clinical model,as well as the complex model of the combination of the two.The accuracy and predictive efficiency of the three models were evaluated by the concordance index（C-index）of the training group and the verification group,and a nomogram was developed based on the best prediction model to judge the prognosis of patients with non-small cell lung cancer.Results:1.A total of 300 patients with NSCLC were included in the study through follow-up（210 cases in training group and 90 cases in verification group）,including 189 males and 111 females with a mean age of 62.3±9.0years and a median age of 63 years.The average follow-up time was 39.4±9.1months,and the median follow-up time was 37.5 months.164 cases（54.7%）developed disease progression and 110 cases（36.7%）died.The median OS was 31 months,the 1-,2-and 3-year OS rates were 88.3%,72.3%and 65.3%respectively,the median PFS was 28 months,and the 1-,2-and 3-year PFS rates were 68.7%,52.7%and 46.7%,respectively.Kaplan-Meier univariate analysis showed that gender,smoking history,TNM stage,tumor diameter,lymph node metastasis,distant metastasis,pathological classification and treatment were common influencing factors of OS and PFS in NSCLC patients,while Cox multivariate analysis showed that gender,TNM stage and treatment mode were independent prognostic factors of OS,while smoking history,TNM stage and treatment mode were independent prognostic factors of PFS.2.96 radiomics feature parameters,including 49 PET features and 47 CT features,were extracted from the primary tumor of each lung cancer patient by LIFEx software.Mann-Whitney U test screened 86 features（including 46 PET and 40 CT features）significantly related to OS and 90 features（including 47PET and 43 CT features）significantly related to PFS from the training group.LASSO-Cox algorithm combined with 10-fold cross-validation further screened 10 features（including 6 PET and 4 CT features）that were significantly related to OS and 6 features（including 3 PET and 3 CT features）that were significantly related to PFS,and calculated the radiomics score（Rad-score）of each patient.The concordance index（C-index）of OS of NSCLC patients predicted by radiology model was 0.762（95%CI:0.737～0.788）,the C-index of verification group was 0.762（95%CI:0.719～0.804）,the C-index of predicted PFS of patients was 0.724（95%CI:0.700～0.749）,and the C-index of verification group was 0.689（95%CI:0.655～0.723）.3.In this study,there are 9 clinical features that may be related to the prognosis of NSCLC:sex,age,smoking history,TNM stage,tumor diameter,lymph node metastasis,distant metastasis,pathological classification and treatment.LASSO-Cox algorithm and 10-fold cross-validation were used to select the best predictive features in the training group,and the clinical model score（clinical-score）of each patient was calculated.5 features significantly related to OS and 6 features significantly related to PFS were screened.The five best clinical features based on OS were gender,smoking history,TNM stage,tumor diameter and treatment,while the six best clinical features based on PFS were gender,smoking history,TNM stage,tumor diameter,lymph node metastasis and treatment.The concordance index（C-index）of OS of NSCLC patients predicted by clinical model was 0.834（95%CI:0.814～0.854）,the C-index of verification group was 0.770（95%CI:0.729～0.810）,the C-index of predicted PFS of patients was 0.780（95%CI:0.760～0.801）,and the C-index of verification group was 0.777（95%CI:0.747～0.807）.4.The complex model for predicting OS and PFS was obtained by combining the radiomics score with the clinical factors screened by LASSO,and the score of each patient（Complex-score）was calculated.The concordance index C-index of the complex model for predicting the OS of NSCLC patients was 0.842（95%CI:0.822～0.861）,the C-index of the verification group was 0.778（95%CI:0.738～0.818）,the C-index of the predicted patient PFS was 0.787（95%CI:0.767～0.806）,and the C-index of the verification group was 0.775（95%CI:0.745～0.805）.The results of Cox multivariate analysis showed that both Rad-score and Complex-score were independent prognostic factors affecting OS（HR:1.804,9.996,P=0.042、P=0.000）and PFS（HR:1.771,5.627,P=0.011,P=0.000）.That is,radiomics model and complex model have good predictive effect on the prognosis of patients with NSCLC.5.The complex model has the highest accuracy and efficiency in predicting the prognosis of patients with NSCLC.Based on the complex model,a nomogram is developed to predict the survival probability of OS and PFS.The concordance index C-index of OS predicted by nomogram was0.855,and the C-index of predicted PFS was 0.797.The calibration curve showed a good coincidence with the actual survival probability,and it was also true in the verification group.Conclusions:1.Gender,TNM stage and treatment mode were independent prognostic factors for OS,while smoking history,TNM stage and treatment mode were independent prognostic factors for PFS.2.Before treatment,¹⁸F-FDG PET/CT radiomics could predict the consistency index of OS and PFS,radiomics in patients with NSCLC,the consistency index of OS and PFS was 0.762 and 0.724,respectively,and the predictive efficiency was higher,and the radiomics score was an independent prognostic factor affecting OS and PFS.The prognosis of patients in high risk group（greater than the median of radiomics score）was poor.3.The clinical characteristic model of patients with non-small cell lung cancer can also predict that the concordance index of OS and PFS,for predicting OS and PFS is 0.834 and 0.780 respectively.The best clinical features based on OS are gender,smoking history,TNM stage,tumor diameter and treatment mode,while PFS-based correlation is gender,smoking history,TNM stage,tumor diameter,lymph node metastasis and treatment.4.The complex model based on the combination of radiomics and clinical model is the most effective in predicting the prognosis of patients with non-small cell lung cancer,and the resulting nomogram is simple and convenient to assist clinical decision-making.Radiomics model can be used as a powerful auxiliary tool in predicting the survival and prognosis of non-small cell lung cancer.Clinical factors combined with radiomics score can more accurately predict the overall survival time and progression-free survival time of patients.