Radiomics-Based Support Vector Machine Distinguishes Molecular Events Driving the Progression of Lung Adenocarcinoma
Hongji Li, Zhen‐Bin Qiu, Meng-Min Wang, Chao Zhang, Hui-Zhao Hong, Rui Fu, Lishan Peng, Chen Huang, Qian Cui, Jia-Tao Zhang, Jingyun Ren, Lei Jiang, Yi‐Long Wu, Wen-Zhao Zhong
Abstract
INTRODUCTION: An increasing number of early-stage lung adenocarcinomas (LUAD) are detected as lung nodules. The radiological features related to LUAD progression warrant further investigation. Exploration is required to bridge the gap between radiomics-based features and molecular characteristics of lung nodules. METHODS: Consensus clustering was applied to the radiomic features of 1212 patients to establish stable clustering. Clusters were illustrated using clinicopathological and next-generation sequencing. A classifier was constructed to further investigate the molecular characteristics in patients with paired computed tomography and RNA sequencing data. RESULTS: Patients were clustered into four clusters. Cluster 1 was associated with a low consolidation-to-tumor ratio, preinvasion, grade I disease, and good prognosis. Clusters 2 and 3 reported increasing malignancy with a higher consolidation-to-tumor ratio, higher pathologic grade, and poor prognosis. Cluster 2 possessed more spread through air spaces and cluster 3 reported a higher proportion of pleural invasion. Cluster 4 had similar clinicopathological features as cluster 1 except but a proportion of grade II disease. RNA sequencing indicated that cluster 1 represented nodules with indolent growth and good differentiation, whereas cluster 4 reported progression in cell development but still had low proliferative activity. Nodules with high proliferation were classified into clusters 2 and 3. In addition, the radiomics classifier distinguished cluster 2 as nodules harboring an activated immune environment, whereas cluster 3 represented nodules with a suppressive immune environment. Furthermore, signatures associated with the prognosis of early-stage LUAD were validated in external datasets. CONCLUSIONS: Radiomics features can manifest molecular events driving the progression of LUAD. Our study provides molecular insight into radiomics features and assists in the diagnosis and treatment of early-stage LUAD.