Abstract
Thymic involution, characterized by adipose replacement of functional thymic tissue, is a broadly recognized feature of age-related immunosenescence. Currently, no established, non-invasive method measures residual thymus in adults, its relationship with host immunity, or its influence on tumor immunosurveillance, cancer outcomes, and treatment-related toxicities. In a multi-institutional cohort of patients with locally advanced non-small cell lung cancer (NSCLC), we created a new CT-based metric to estimate residual functional thymic tissue and examined its association with toxicities and clinical outcomes. We defined a thymic region atlas and employed a novel segmentation model that integrates Transformer-based attention with convolutional detail preservation to automatically detect and delineate the thymic region on chest CT images. We then computed a new radiographic parameter, capturing the proportion of persistent thymic tissue or "percent thymic tissue" (pTT), using Gaussian mixture modeling of the components of the delineated primal thymic space. We applied our framework to a publicly available NSCLC radiotherapy dataset (MAASTRO) and an institutional cohort at Memorial Sloan Kettering Cancer Center (MSKCC), assessing associations with age and sex. Kaplan-Meier estimates and Cox proportional hazards modeling were used to evaluate the association between pTT and distant metastasis-free survival (DMFS), locoregional failure (LRF), progression-free survival (PFS), and overall survival (OS). A total of 464 patients with stage III NSCLC, pre-treatment chest CT scans, curated clinical data, and long-term follow-up were analyzed. Of these, 277 were from MAASTRO and 187 from MSKCC. Our pTT method allowed tissue decomposition without manual thresholding or shape priors and was inherently robust to segmentation variability. pTT decreased with age at a rate of 0.36 percentage points per year (p < 0.0001) in the MAASTRO cohort and 0.40 percentage points per year in the MSKCC group. pTT was significantly lower in males (MAASTRO: 9.75% vs. 17.0%, p<0.001; MSKCC: 5.97% vs. 12.3%, p<0.001). On multivariable Cox regression adjusting for age, sex, and histology, pTT was an independent predictor of survival (MAASTRO: aHR = 0.71, 95% CI: 0.50-0.99; p = 0.049; MSKCC: aHR = 0.51, 95% CI: 0.27-0.95; p = 0.058). Notably, pTT was also associated with DMFS (aHR = 0.51, 95% CI: 0.27-0.95; p = 0.03), LRF (aHR, 0.96; 95% CI, 0.93-0.99; p = 0.014), and PFS (aHR = 0.57, 95% CI: 0.32-1.01; p = 0.052) in the MSKCC cohort. There was a link between pTT and 12- and 24-month estimates for all end-points. pTT remained a significant predictor of improved DMFS even after adjusting for thymic radiation dose (aHR = 0.97; 95% CI, 0.94-0.99; p = 0.037), which itself was independently associated with inferior DMFS (aHR = 2.01; 95% CI, 1.02-3.99; p = 0.045). The LRF association also persisted after adjusting for thymus V20 (aHR, 0.96; 95% CI, 0.92-0.99; p = 0.008). Severe pneumonitis (grade ≥3) occurred more frequently in patients with low pTT values (13.9% vs. 5.2%; p = 0.04). Patients with both high lung V20 and low pTT represented the subgroup at greatest pneumonitis risk (20.8%, n = 11), whereas those with low lung V20 and high pTT had few observed events (1.9%, n =1). pTT also predicted durvalumab discontinuation due to treatment-related adverse events, particularly among patients with unresectable stage IIIC NSCLC (AUC = 0.8). Thus, the parameter pTT offers a reliable and interpretable non-invasive quantitative measure of residual functional thymus in adults, reflects age-related thymic involution, and independently predicts survival and treatment-related toxicity in stage III NSCLC. These findings support pTT as a structural imaging indicator of thymic function and suggest its potential for studying how cancer therapies impact host immunity, which may, in turn, influence long-term lung cancer treatment outcomes.