Predicting immunotherapy response in advanced solid tumors using quantitative imaging features from CD8 PET/CT exams.

Abstract

3098 Background: CD8-PET/CT imaging with 89 Zr crefmirlimab berdoxam (ImaginAb, Inc), which targets CD8-expressing T-lymphocytes, is being explored as an imaging tool to predict responses and monitor immune checkpoint inhibitors (ICI) in patients with advanced solid malignancies. Here we explore how quantitative imaging features from CD8 PET/CT may predict ICI responses. Methods: We studied quantitative imaging features (PET parameters and radiomics) in 45 patients from the ImaginAb IAB-CD8-201 phase II trial (NCT03802123). Tumoral lesions, peritumoral ring (ring shaped margin extending 0.5 cm inwards and outwards from the segmented tumor surface), healthy tissue, benign and pathological lymph nodes were segmented from baseline and first on-treatment (4-6 weeks after standard of care treatment including ICI blockade). Imaging features from CD8-PET/CT scans were extracted. Predictive models for best overall response (BOR) according to RECIST 1.1 were developed. Models’ performance was evaluated by the ability to distinguish responders (complete or partial response, n = 13) from non-responders (stable or progressive disease, n = 32). A survival random forest analysis was also conducted to estimate time to BOR. Results: Significantly greater delta values were identified in the tumor and peritumoral ring compared to healthy tissues, suggesting the tumor and peritumoral ring may reveal early treatment-induced changes important for predicting response. Eighteen predictive models were developed, with models using imaging features from the peritumoral ring showing comparable performance to those using features from lesions and pathological lymph nodes. The simplest BOR predictive model, which yielded the highest performance, used delta values extracted from the peritumoral ring (AUCs = 0.895, sensitivity = 0.900, specificity = 0.615). The inclusion of clinical variables (including age, sex, body mass index, cancer type, received treatments, number of lines of received treatment, white blood cell count) did not significantly enhance model accuracy, emphasizing the robustness of the imaging data alone. A final model integrating key imaging features from multiple regions successfully predicted time to BOR with a C-index, a generalizable AUC that considers censored data, of 0.86. Conclusions: This study highlights the potential of quantitative imaging analysis of CD8-PET/CT scans as a tool for predicting responses to cancer immunotherapy. Results suggest the effectiveness of delta imaging features from the peritumoral ring as a potential indicator of patient’s ability to respond to ICI treatment, simplifying the analysis without sacrificing accuracy. Further validation in trials with more homogeneous populations and treatment regimens (eg. NCT05013099) is warranted, with the potential to advance personalized cancer care.