Preoperative CT-based Radiomic Features Combined With Clinical Features Prediction For Tumor Spread Through Air Spaces In Clinical Stage Ⅰ Lung Adenocarcinoma

Objective: Tumor spread through air spaces(STAS)was a novel pattern of invasion in lung cancer,defined as micropapillary clusters,solid nests,or single cells spreading within air spaces in the surrounding lung parenchyma,beyond the edge of the main tumor.In lung adenocarcinoma(LUAD),14.8%～56.4% has been found STAS and patients with STAS-positive has been proven a worse recurrence-free survival(RFS)and overall survival(OS).As a quantitative analysis tool,radiomics can extract and analyze the features from radiographic images.A number of studies have already demonstrated that radiomics were promise in predicting pathological response,lymph node metastasis,and disease diagnosis.However,most researches have ignored the peritumoral microenvironment and only assessed the radiomic feature of the primary tumor.Several studies have already demonstrated that peritumoral pulmonary parenchyma can provide significant value for clinical assessment of tumor metastasis.But there still remains unclear that peritumoral radiomic features are useful tools for prediction of STAS status.This study intends to develop and validate a noninvasive model combined with the preoperative CT-based radiomic and clinical signatures to predict STAS in clinical stage I lung adenocarcinomas.Methods: In this retrospective,diagnostic study,we reviewed the patients with pathologically-confirmed invasive LUAD,who accepted surgical resection from June 2018 to December 2019 in Guangdong Provincial People’s Hospital.‘Py Radiomics’ package were used to extracted the features from tumor and incremental distances of 5,10,15,and 20 mm outside the tumor.The least absolute shrinkage and selection operator algorithm and logistic regression analysis were used to select the signatures and constructed into models.The performance of each model was measured by the receiver operating curve analysis.Results: The study comprised 256 individuals with 85(33.2%)participants were STAS positive status and 171(66.8%)participants were STAS negative status.Our univariable and multivariable logistic regression analysis show that,the CEA and Boundary were independent risk factors associated with STAS positive.Postoperative pathology showed the STAS-positive group had more lymphatic metastasis and pleural Invasion,while the results of genetic testing also indicated that epidermal growth factor receptor(EGFR)wild type and Anaplastic lymphoma kinase(ALK)mutation were more frequently identified in STAS-positive tumors.The AUC for the CRS,TRS and each PRS were 0.665(95% CI,0.522-0.808),0.796(95% CI,0.686-0.906),0.709(95% CI,0.583-0.835),0.740(95% CI,0.618-0.863),0.831(95% CI,0.730-0.932),0.784(95%CI,0.673-0.894),and the AUC of combined radiomic signature selected from the tumor area and the peritumoral area of 15 mm(GRS)was 0.870(95% CI,0.781-0.958).The Nomogram combining the clinical and gross radiomic signatures can accurately predict STAS in clinical stage I lung adenocarcinoma of AUC,0.869(95% CI,0.776-0.961).Conclusion: Our study show that the CEA and Boundary were independent risk factors of STAS positive.Postoperative pathology showed the STAS-positive group had more lymphatic metastasis,pleural Invasion,epidermal growth factor receptor(EGFR)wild type and Anaplastic lymphoma kinase(ALK)mutation.And our study also identified the tumor and peritumoral features were significantly associated with STAS status.Compared with the clinical,tumor and each peritumoral radiomic model alone,the PRS-15 mm have the best predictive features to predicted STAS.The combine of tumor and pertumoral radiomic model(GRS)can achieve the better performance of the prediction of STAS;This study demonstrated Nomogram that unit of radiomic signature and clinical signature could have a better performance of prediction of STAS in clinical stage I lung adenocarcinoma.