Establishing a predictive model for tumor mutation burden status based on CT radiomics and clinical features of non-small cell lung cancer patients

Jihua Yang,Wenjia Shi,Zhen Yang,Hang Yu,Miaoyu Wang,Yuanhui Wei,Juyi Wen,Wei Zheng,Peng Zhang,Wei Zhao,Liang'an Chen

DOI: https://doi.org/10.21037/tlcr-23-171

2023-04-28

Abstract:Background: Tumor mutation burden (TMB) is one of the biomarkers for efficacy of immune checkpoint inhibitors (ICIs) in non-small cell lung cancer (NSCLC). Due to the potential of radiomic signatures to identify microscopic genetic and molecular differences, thus radiomics is considered a suitable tool for judging the TMB status probably. In this paper, the radiomics method was applied to analyze the TMB status of NSCLC patients, so as to construct a prediction model for distinguishing between TMB-high and TMB-low status. Methods: A total of 189 NSCLC patients with TMB detection result were retrospectively included between 30 November 2016 and 1 January 2021, and were divided into two groups: TMB-high (≥10/Mb, 46 patients) and TMB-low (<10/Mb, 143 patients). Some clinical features related to TMB status were screened out in 14 clinical features and 2,446 radiomic features were extracted. All patients were randomly divided into a training set (n=132) and a validation set (n=57). Univariate analysis and least absolute shrinkage and selection operator (LASSO) were used for radiomics feature screening. A clinical model, radiomics model, and nomogram were constructed with the above screened features and compared. Decision curve analysis (DCA) was used to evaluate the clinical value of the established models. Results: Two clinical features (smoking history, pathological type) and 10 radiomics features were significantly correlated with the TMB status. The prediction efficiency of the intra-tumoral model was better than that of the peritumoral model (AUC: 0.819 vs. 0.816; accuracy: 0.773 vs. 0.632, specificity: 0.767 vs. 0.558). The efficacy of the prediction model based on radiomic features was significantly better than that of the clinical model (AUC: 0.822 vs. 0.683; specificity: 0.786 vs. 0.643). The nomogram, established by combining smoking history, pathologic type, and rad-score, showed the best diagnostic efficacy (AUC =0.844) and had potential clinical value in assessing the TMB status of NSCLC. Conclusions: The radiomics model based on CT images of NSCLC patients performed well in distinguishing the status of TMB-high and TMB-low, and the nomogram could provide additional information on the timing and regimen of immunotherapy.