e20613 Background: Accurate prediction of how patients with non-small cell lung cancer (NSCLC) will respond to immunotherapy is essential for customizing individual treatment plans. This research explores the use of radiomics analysis from pre-treatment CT scans as a method to predict the effectiveness of immunotherapy in individuals with NSCLC. Methods: This IRB-approved retrospective study investigated radiologic and clinical data from 159 patients with stage III-IV NSCLC treated with immunotherapy. Imaging responses were assessed according to RECIST 1.1 and immune-related RECIST (irRECIST) criteria at least twice during a 24-week period. Responses were categorized as complete response (CR), partial response (PR), stable disease (SD), and progressive disease (PD). 3D radiomics features were extracted from both the tumor and a 1 cm thick peritumoral region. A linear mixed-effect harmonization model was employed to correct for scanner-associated variations. A Random Forest algorithm was used to develop a classification model to predict response according to RECIST 1.1 or irRECIST. The dataset was split into training (70%) and test (30%) sets. The accuracy of predictions was evaluated using confusion matrix statistics and bootstrapping with 1,000 iterations for median and 95% confidence interval (CI). Results: Using RECIST 1.1 criteria, 41 patients (25.2%) achieved PR and CR(0.6), 62 (39.0%) achieved SD, and 56 (35.2%) achieved PD. IrRECIST criteria yielded slightly higher CR/PR rates with 52 patients (32.7%), while SD decreased to 65 (41.0%) and PD remained at 42 (26.4%). For predicting CR/PR and SD responses, the models achieved a sensitivity of 0.45 (95% CI: 0.38-0.50) and 0.64 (0.60-0.68) for irRECIST, compared to 0.34 (0.28-0.38) and 0.55 (0.52-0.60) for RECIST 1.1. The corresponding specificity was 0.85 (0.83-0.87) and 0.63 (0.59-0.67) for irRECIST, compared to 0.91 (0.89-0.93) and 0.67 (0.63-0.72) for RECIST 1.1. For predicting PD response, the models achieved a sensitivity of 0.49 (0.45-0.56) for irRECIST and 0.57 (0.52-0.62) for RECIST 1.1. The models' balanced accuracy for predicting CR/PR, SD, and PD were 0.65 (0.61-0.68), 0.64 (0.61-0.67), and 0.65 (0.63-0.69) for irRECIST, compared to 0.62 (0.59-0.64), 0.61 (0.59-0.64), and 0.62 (0.59-0.64) for RECIST 1.1, respectively. Conclusions: This study demonstrates the potential of pre-treatment CT radiomics to predict immunotherapy responses in NSCLC patients, offering insights into personalized treatment approaches. However, larger studies are needed to validate our findings.