Radiomics of Tumor Heterogeneity in (18)F-FDG-PET-CT for Predicting Response to Immune Checkpoint Inhibition in Therapy-Naïve Patients with Advanced Non-Small-Cell Lung Cancer

SIMPLE SUMMARY: Biomarkers reliably predicting treatment response to immune checkpoint inhibition (CKI) therapy in advanced non-small cell lung cancer (NSCLC) are warranted. Baseline (18)F-FDG-PET-CT (PET-CT) is an integral part of the diagnostic algorithm of NSCLC. However, there is poor evidence on the predictive and prognostic value of initial PET-CT imaging in these patients. The use of Radiomics has gained prominence in the last decade allowing for the extraction and artificial intelligence-based analysis of additional imaging parameters, so-called radiomic features (RFs). We aimed to find RFs predicting treatment response for CKI-based first-line therapy in advanced NSCLC patients, out of whole-body metabolic PET and morphological CT imaging. PET RFs might additionally be predictive and prognostic and could thus provide important information for future therapy monitoring and guidance. ABSTRACT: We aimed to evaluate the predictive and prognostic value of baseline (18)F-FDG-PET-CT (PET-CT) radiomic features (RFs) for immune checkpoint-inhibitor (CKI)-based first-line therapy in advanced non-small-cell lung cancer (NSCLC) patients. In this retrospective study 44 patients were included. Patients were treated with either CKI-monotherapy or combined CKI-based immunotherapy–chemotherapy as first-line treatment. Treatment response was assessed by the Response Evaluation Criteria in Solid Tumors (RECIST). After a median follow-up of 6.4 months patients were stratified into “responder” (n = 33) and “non-responder” (n = 11). RFs were extracted from baseline PET and CT data after segmenting PET-positive tumor volume of all lesions. A Radiomics-based model was developed based on a Radiomics signature consisting of reliable RFs that allow classification of response and overall progression using multivariate logistic regression. These RF were additionally tested for their prognostic value in all patients by applying a model-derived threshold. Two independent PET-based RFs differentiated well between responders and non-responders. For predicting response, the area under the curve (AUC) was 0.69 for “PET-Skewness” and 0.75 predicting overall progression for “PET-Median”. In terms of progression-free survival analysis, patients with a lower value of PET-Skewness (threshold < 0.2014; hazard ratio (HR) 0.17, 95% CI 0.06–0.46; p < 0.001) and higher value of PET-Median (threshold > 0.5233; HR 0.23, 95% CI 0.11–0.49; p < 0.001) had a significantly lower probability of disease progression or death. Our Radiomics-based model might be able to predict response in advanced NSCLC patients treated with CKI-based first-line therapy.