A Validation of Immune Pathology-Based Radiomics Signature With Stage I NSCLC Treated With SABR

Abstract

Radiomics-based selection might aid in stratifying patients for treatments such as stereotactic ablative radiotherapy (SABR). An immune pathology-based radiomics signature was previously created based on a NSCLC population treated with surgical resection, showing its significant association with overall survival (OS). This study was to conduct a model validation in an independent NSCLC cohort treated with SABR alone. Pretreatment CT images of 295 stage I NSCLC patients treated with SABR between February 2005 and October 2013 were analyzed. We utilized IBEX to quantify the four radiomics features that constituted the model: CT intensity global mean, standard deviation, uniformity and gray level co-occurrence matrix [GLCM] homogeneity (Tang et al. ASTRO 2016). One of the four previously identified pathology-based radiomics clusters [Cluster A (high intensity and low heterogeneity), B (high intensity and high heterogeneity), C (low intensity and low heterogeneity) and D (low intensity and high heterogeneity)] was assigned to each SABR patient through evaluation of distributions of pairwise similarity measures obtained from consensus clustering. It is noted that Cluster D was characterized by lower PDL1 expression and higher count of tumor infiltrating lymphocytes (TIL) as well as better survival in our previous study. Univariate and multivariate analyses were performed to evaluate the radiomics-derived subtypes for association with survival in the SABR cohort using Cox proportional hazards regression. The 295 patients were predominantly T1N0 (85.8%) adenocarcinomas (54.6%), with median OS of 61.2 months after receiving SABR of BED ≥ 100 Gy (98.6%). Patients were assigned into four clusters: Cluster A/B/C/D (56/79/69/91). Cluster D exhibited evidence of prolonged OS (median of 73.7 months) when compared to the others (57.0/50.4/63.2 months for Cluster A/B/C, respectively). Statistically significant association was demonstrated in univariate analysis (p<0.05), for which the partial effects approached marginal significance in multiple regression adjusted for age and volume (p=0.05). The prognostic performance of the radiomics signature was validated on univariate analysis, with a strong trend toward significance on multivariate analysis in this stage I NSCLC cohort treated with SABR. Based on the association with immune pathology marker expression, the model could potentially be used in early detection of patients likely or unlikely to benefit from the combination of SABR and immunotherapy.