A New Classification Method for Pulmonary Ground-Glass Nodules Using Radiomics Approach

Abstract

Abstract Purpose To create new method utilizes radiomics to classify ground-glass nodules (GGNs). Methods A total of 855 patients with lung adenocarcinoma, presenting GGNs of size ≤ 3cm, were included in the study. The radiomics features were dimensionally reduced using the t-Distributed Stochastic Neighbor Embedding (t-SNE) algorithm and clustered with the K-Means algorithm. Single-factor analysis was conducted to compare patient conditions across different clusters. Finally, the new classification method was compared with the method used two-dimensional (2D) computed tomography (CT) features to verify the efficacy of the novel approach. Results The nodules were clustered into two groups, A and B. Single-factor analysis revealed significant statistical differences between the two groups in terms of age, smoking history, nodule diameter, solid component diameter, and the consolidation tumor ratio (CTR). Group A primarily comprised non-invasive adenocarcinoma (non-IAC) (81.2%) and low-risk nodules (75.9%), while group B primarily comprised invasive adenocarcinoma (IAC) (85.8%) and medium-high risk nodules (77.4%). In terms of predicting IAC and medium-high risk nodules, the new method performed better. Conclusion The new classification method effectively utilizes radiomics information and offers significant guidance for the management of various GGNs categories, exhibiting notable advantages over traditional methods.