Integrating Imaging, Histologic, and Genetic Features to Predict Tumor Mutation Burden of Non-Small-Cell Lung Cancer

Our investigation of the relationship between the tumor mutation burden (TMB) and the imaging, histologic, and genetic features in non-small-cell lung cancer showed that specific computed tomography features are associated with TMB in non small-cell lung cancer. Thus, integration of imaging findings, histologic type, and TP53 genotype might serve as a clinically practical assay for TMB. The composite model has the potential to identify patients most likely to benefit from cancer immunotherapy. Background: Immune checkpoint inhibitors have dramatically changed the landscape of therapeutic management of non-small-cell lung cancer (NSCLC). Tumor mutation burden (TMB) is an important biomarker of the response to cancer immunotherapy. We investigated the relationship between TMB and the imaging, histologic, and genetic features in NSCLC. Materials and Methods: We evaluated the associations between the semantic imaging features (7 quantitative or semiquantitative imaging features and 13 qualitative features that reflect the tumor characteristics) and TMB and built an imaging signature for TMB using logistic regression. Finally, we integrated the imaging signature, histologic type, and TP53 genotype into a composite model. Results: Among 89 patients, 37 (41.6%) had low TMB and 52 (58.4%) had high TMB. Tumors with high TMB were more prevalent in squamous cell carcinoma (P = .017) and those with a TP53 mutation (P < .0001). The absence of concavity was significantly associated with higher TMB (P =.008). An imaging signature containing 5 features, including concavity, border definition, spiculation, thickened adjacent bronchovascular bundle and size, achieved good discrimination between tumors with low and high TMB (area under the curve [AUG], 0.79; 95% confidence interval [CI], 0.69-0.89). The composite model integrating the imaging signature, histologic type, and TP53 genotype improved the classification (AUC, 0.89; 95% CI, 0.82-0.95) compared with the imaging signature alone using the DeLong test (P = .012). The composite model achieved a high sensitivity of 95% and a specificity of 62%. Conclusion: Specific computed tomography features are associated with TMB in NSCLC, and the integration of imaging, histologic, and genetic information might allow for accurate prediction of TMB. (C) 2019 Elsevier Inc. All rights reserved.