Ablation versus sublobar resection for stage IA non-small cell lung cancer: a multicenter retrospective cohort study using a deep learning model in the ablation group.
Authors
Affiliations (4)
Affiliations (4)
- State Key Laboratory of Oncology in South China, Sun Yat-sen University Cancer Center, Department of Minimally Invasive Intervention, Guangdong Provincial Clinical Research Center for Cancer, Collaborative Innovation Center for Cancer Medicine, Guangzhou, P.R. China.
- Department of Interventional Radiology, Xinjiang Key Laboratory of Translational Biomedical Engineering Research, Tumor Hospital Affiliated to Xinjiang Medical University, The Third Clinical Medical College of Xinjiang Medical University, Urumqi, China.
- GuangZhou FuDa Cancer Hospital, Guangzhou, P.R. China.
- State Key Laboratory of Oncology in South China, Sun Yat-sen University Cancer Center, Department of Thoracic Surgery, Guangdong Provincial Clinical Research Center for Cancer, Collaborative Innovation Center for Cancer Medicine, Guangzhou, P.R. China.
Abstract
Image-guided thermal ablation (IGTA) has increasingly been used for stage IA non-small cell lung cancer (NSCLC), yet long-term outcomes comparing IGTA with sublobar resection (SLR) without surgical contraindications remain unclear. Identifying patients who may benefit from ablation is also technically challenging. A total of 2,145 patients treated with IGTA or SLR from 2012 to 2023 were analyzed. Propensity score matching was performed. Progression-free survival (PFS) and cancer-specific survival (CSS) rates were estimated using the Kaplan-Meier method. A deep learning model based on pre-treatment computed tomography images was developed to predict disease-free survival after IGTA and identify patients most appropriate for ablation. After matching, the median follow-up was 39.3 months for the SLR group and 46.7 months for the IGTA group. No significant differences were observed in PFS (hazard ratio [HR], 2.02; 95% confidence interval [CI], 0.83-4.89; P = 0.120) or CSS (HR, 3.09; 95% CI, 0.87-11.04; P = 0.082). The area under the curve (AUC) of the Vision Transformer (ViT) model was 0.826 and 0.814 for the training and external validation cohorts, respectively. Higher ViT scores predicted worse survival (HR, 8.61; P < 0.001). The Multimodal XGBoost model further improved the performance. In this study, long-term outcomes after IGTA were not significantly different from those after SLR in patients with stage IA NSCLC. The ViT model accurately predicted outcomes after ablation. These results indicate that ablation may be a feasible alternative to conventional surgery in appropriately selected patients.