Long-Term (5-Year) Growth Risk Stratification of Pulmonary Subsolid Nodules by Integrating Clinical, Radiomics, and Deep Learning Features.

RATIONALE AND OBJECTIVES: Pulmonary subsolid nodules (SSN) often exhibit indolent growth, necessitating extended follow-up to assess progression. We aimed to develop and externally validate a long-term (>5-year) growth prediction model for SSN by integrating clinical, radiomics, and deep learning (DL) features. MATERIALS AND METHODS: Patients with SSN who had a follow-up period of at least 5 years between 2012 and 2022 in two hospitals were retrospectively included. Clinical characteristics selected by multivariate regression analysis were used to construct the clinical model. The variational autoencoder (VAE)-structured DL network was trained on lesion-derived radiomics and DL features to output a Radscore (RS). The previous features were analyzed by multivariate regression to construct combined models. The performance of these models was evaluated by the receiver operating characteristic (ROC) curve and calibration curve. RESULTS: Ultimately, 574 patients were included, and six prediction models were constructed. After multivariate regression analysis, five clinical characteristics, as well as RS_Radiomics and RS_DL, were used to construct the Clinical_Radiomics_DL combined model. A nomogram was created to predict SSN growth probability. In external validation, the combined model showed the best performance, achieving an AUC of 0.81 (95% CI: 0.70-0.92) and the lowest IPCW Brier score of 0.074. Kaplan-Meier analysis showed significantly different SSN stability rates between high- and low-risk groups. Calibration curve indicated high consistency between the model's predictions and actual outcomes. CONCLUSION: We developed a long-term SSN growth prediction model integrating clinical, radiomics, and DL features, which provides a valuable reference for the precise management of SSN in clinical practice.