Delta-radiomics for prediction of pseudoprogression in malignant melanoma treated with immune checkpoint inhibition.

Abstract

9575 Background: Distinguishing progressive disease (PD) from pseudoprogression (PP) in patients treated with immune-checkpoint inhibition (ICI) is challenging and usually requires confirmation follow-up imaging or invasive diagnostic techniques. This project aimed to identify predictive radiomic signatures for PP from CT imaging. Methods: The response to ICI of 105 metastatic melanoma patients with 645 metastases was retrospectively correlated with radiomic signatures (172 total features). All metastatic lesions were delineated at 3 time points: prior to ICI (t0), at 3 (t1) and 6 months (t2). Response was defined individually for each metastasis using RECIST 1.1, comparing baseline t0 to t2. Three prediction models for PP were built: CT radiomics at t0 and t1, as well as the relative difference between both t0 and t1 (delta-radiomics). Results: Median follow-up was 18 months and 2-year OS and PFS were 72% and 25%, respectively. Median OS: not reached, median PFS: 6 months. Response per lesion at t1: 13% complete remission (CR), 19% partial remission (PR), 52% stable disease (SD) and 16% PD. At t2: 16% CR, 31% PR, 38% SD and 15% PD. 106 progressive lesions were identified at t1, of which, 26 changed to SD, 1 to CR and 3 to PR at t2, resulting in 30 PPs (4.7%). Metastasis location significantly influenced response rates but was not associated with PP (p = 0.4). Lung metastases had significantly higher response rates than soft tissue (p < 0.001), liver (p < 0.001) and bone metastases (p = 0.008). Univariate analysis followed by removal of correlated features revealed no significant radiomic features associated with PP at t0. One independent feature was identified at t1 (AUC 0.74), while delta-radiomics was the best performing approach, identifying four independent features (AUC 0.72 to 0.81). A final multivariate delta radiomics logistic regression model was generated and internally validated, achieving an AUC of 0.81 (± 0.11, 10-fold cross-validation). Conclusions: Metastasis location significantly influenced response rates and CT-based delta-radiomics is a promising biomarker for early differentiation between pseudoprogression and true progression in metastatic melanoma patients treated with ICI.