{
  "abstract": "Background Immunotherapy has transformed the treatment landscape of advanced non-small cell lung cancer (NSCLC), offering prolonged survival and durable responses in select patients. However, current biomarkers such as programmed cell death ligand 1 (PD-L1) expression lack sufficient predictive accuracy, with up to 70% of patients with advanced NSCLC failing to derive long-term benefit from immune checkpoint inhibitors (ICIs). Ranging from molecular alterations to fibroblast barriers, immune evasion mechanisms are likely to manifest in radiologic patterns. In this study, we aimed to develop a predictive model using baseline Computed Tomography (CT) scans to identify patients likely to achieve durable clinical benefit (DCB) from ICI.Methods A total of 101 eligible patients with stage IIIB and IV NSCLC who have been treated with ICIs were analyzed in this study. DCB was defined as response or stable disease lasting at least 24 weeks according to the RECIST criteria. Tumor features were extracted and quantified from pre-ICI CT scans via computational image analysis. A total of 185 radiomic features were extracted, quantifying tumor shape, intensity, and texture. A 3D-CNN was then trained on the segmented tumor volumes, and its output was fused with an XGBoost model trained on the top 15 radiomic features selected by recursive feature elimination. The final prediction was generated by a logistic regression meta-learner. Patients were also stratified into high- and low- risk groups for survival analysis using Kaplan-Meier curves.Results Among the included patients (median age 66 years, IQR 59.5-73.5), fifty-eight (57.4%) patients achieved DCB, while forty-three (42.6%) did not. The most frequently administered regimens were ipilimumab, nivolumab, and pembrolizumab, with or without platinum-based chemotherapy. The integrated radiomics-CNN model demonstrated robust performance for predicting DCB, with a mean Area Under the Curve (AUC) of 0.72 (95% CI: 0.67-0.76) in 5-fold cross-validation. Key predictive features included radiomic descriptors of tumor heterogeneity, such as Gray Level Run Length Matrix (GLRLM), suggesting that more texturally complex tumors are associated with a lack of durable response. Risk stratification identified two groups, with the low-risk group showing a significantly higher likelihood of response to immunotherapy (Relative Benefit: 1.72; 95% CI: 1.08-2.78; p < 0.02).Conclusions An AI-driven model integrating deep learning with quantitative radiomic features from routine CT scans can be utilized to predict durable immunotherapy benefit in patients with advanced NSCLC, enhancing patient selection and personalizing treatment strategies. Greater tumor textural complexity was associated with reduced response, potentially reflecting tissue-level immune evasion.Ethics Approval This study was approved by the Institutional Review Board Committee Northwestern University, Chicago, IL, USA (No. STU0020711; Approval date: 04/06/2018). Consent was waived for its retrospective manner and the was no identifiable risk for patients included in this analysis.",
  "authors": [
    {
      "affiliations": [
        "Northwestern University, Chicago, IL, USA"
      ],
      "name": "Jungmin Kim"
    },
    {
      "affiliations": [
        "Penn State University College of Medicine, Hershey, PA, USA"
      ],
      "name": "Junho Song"
    },
    {
      "affiliations": [
        "Northwestern University Feinberg School of Medicine, Chicago, IL, USA"
      ],
      "name": "Hangyul Lee"
    },
    {
      "affiliations": [
        "Northwestern University Feinberg School of Medicine, Chicago, IL, USA"
      ],
      "name": "Ronald Min"
    },
    {
      "affiliations": [
        "Northwestern University, Chicago, IL, USA"
      ],
      "name": "Young Kwang Chae"
    }
  ],
  "title": "1121 A combined deep learning-based radiomics model predicts durable clinical benefit to immune checkpoint inhibition in non-small cell lung cancer",
  "uid": "3877c2dd-b00f-50a7-af10-77e7ffb64f08"
}