Background: An early biomarker of response to immunotherapy (IO) is needed urgently to identify the patients (pts) who will derive benefit. We reported the first clinical trial of neoadjuvant IO (nIO) in resectable non-small cell lung cancer (NSCLC) (NCT02259621). In this study, we investigated whether there was an association between MPR and radiomic features (RF) in [¹⁸F]-fluorodeoxyglucose ([¹⁸F]-FDG) PET and standard CT images obtained at baseline and after nIO in early stage NSCLC tumors. Methods: Prior to receiving neoadjuvant nivolumab or nivolumab/ipilimumab, patients with Stage I-IIIA NSCLC underwent two [¹⁸F]-FDG PET-CTs and/or plain CTs: a baseline scan at enrollment (PRE), and after nIO (POST). After neoadjuvant treatment, tumors were resected and evaluated for MPR. Volumes of interest (VOIs) were drawn around primary tumors on the scans. Using our novel radiomic software, Imager-4D, VOIs were evaluated for 20 RFs assessing [¹⁸F]-FDG standard uptake value (SUV) or Hounsfield unit (HU) heterogeneity and spatial distribution in PET and CT images respectively. The baseline, post-treatment, and percent change in RFs before and after nIO were compared between tumors with and without MPR. Wilcoxon test was used for the comparisons. Results: The PRE and POST scans of 24 pts were analyzed. All pts had PRE and POST CTs performed, and 17 pts had PRE and POST [¹⁸F]-FDG PET-CT scans. 7 of 24 (29%) had MPR. In the CT scan analysis, HU-based RFs of voxel count, total volume, energy, entropy, homogeneity, contrast, and dissimilarity in POST CT scans each significantly association with MPR. In the PET scan analysis, SUV mean and voxel count RFs in the POST scans, and the percent change in the cluster shade RF between PRE and POST scans were significantly associated with MPR. Conclusions: Collectively, we identified a significant increase in heterogeneity in the POST CT images of NSCLC tumors that had MPR. This association may reflect increased T cell infiltration or tumor necrosis. In contrast, most [¹⁸F]-FDG-based RFs did not distinguish MPR vs non-MPR tumors, although the sample size was limited. We will further investigate these HU-based RFs as non-invasive markers of response to IO in conjunction with pathologic markers of IO response and in a larger patient cohort.