CT radiomics-based cluster analysis for predicting invasiveness of stage I lung adenocarcinoma.

[OBJECTIVE] To explore a noninvasive predicting model for identifying patients with stage I lung invasive adenocarcinoma (IAC).

[METHODS] This study enrolled 289 patients from two medical centers, with 227 and 62 patients in the training and validation sets, respectively. Patients' chest computed tomography (CT) images were used. The K-means cluster algorithm was employed to group patients into new clusters based on radiomics features. In addition, logistic regression was used to develop prediction models. Diagnostic efficiency was assessed using the area under the receiver operating characteristic curve, along with calibration and decision curve analysis.

[RESULTS] The K-means cluster algorithm classified patients into cluster 1 (training: 143; validation: 35) and cluster 2 (training: 84; validation: 27). Cluster 2 had a higher proportion of patients with IAC. The optimal model incorporating tumor diameter, tumor type, and cluster labels achieved the best discriminatory performance, with area under the receiver operating characteristic curve values of 0.848 (95% confidence interval: 0.799-0.898) in the training set and 0.744 (95% confidence interval: 0.583-0.905) in the validation set.

[CONCLUSION] This study proposes a radiomics model that accurately identifies patients with IAC. This prediction tool could aid in personalized risk classification and treatment planning.