Validation of High-Risk Computed Tomography Features for Detection of Local Recurrence After Stereotactic Body Radiation Therapy for Early-Stage Non-Small Cell Lung Cancer

Purpose Fibrotic changes after stereotactic body radiation therapy (SBRT) for stage I non-small cell lung cancer (NSCLC) are difficult to distinguish from local recurrences (LR), hampering proper patient selection for salvage therapy. This study validates previously reported high-risk computed tomography (CT) features (HRFs) for detection of LR in an independent patient cohort. Methods and Materials From a multicenter database, 13 patients with biopsy-proven LR were matched 1:2 to 26 non-LR control patients based on dose, planning target volume (PTV), follow-up time, and lung lobe. Tested HRFs were enlarging opacity, sequential enlarging opacity, enlarging opacity after 12 months, bulging margin, linear margin disappearance, loss of air bronchogram, and craniocaudal growth. Additionally, 2 new features were analyzed: the occurrence of new unilateral pleural effusion, and growth based on relative volume, assessed by manual delineation. Results All HRFs were significantly associated with LR except for loss of air bronchogram. The best performing HRFs were bulging margin, linear margin disappearance, and craniocaudal growth. Receiver operating characteristic analysis of the number of HRFs to detect LR had an area under the curve (AUC) of 0.97 (95% confidence interval [CI] 0.9-1.0), which was identical to the performance described in the original report. The best compromise (closest to 100% sensitivity and specificity) was found at ≥4 HRFs, with a sensitivity of 92% and a specificity of 85%. A model consisting of only 2 HRFs, bulging margin and craniocaudal growth, resulted in a sensitivity of 85% and a specificity of 100%, with an AUC of 0.96 (95% CI 0.9-1.0) (HRFs ≥2). Pleural effusion and relative growth did not significantly improve the model. Conclusion We successfully validated CT-based HRFs for detection of LR after SBRT for early-stage NSCLC. As an alternative to number of HRFs, we propose a simplified model with the combination of the 2 best HRFs: bulging margin and craniocaudal growth, although validation is warranted.