ESTRO 2024 - Abstract Book

S443

Clinical - Biomarkers

ESTRO 2024

Purpose/Objective:

Treatment of metastatic melanoma entails a multi-modal concept including chemotherapy, targeted therapy, immunotherapy, surgery and radiotherapy, which has substantially improved survival in the last decade. Despite immune checkpoint-inhibitors achieving a dramatic increase in overall survival in metastatic melanoma, more than 50% of the patients are non-responders. 1 Therefore, there is an urgent need to identify biomarkers that can stratify patients for the most promising therapy. 2 This study is an imaging sub-project of the “ Tumor Profiler Study ”, which performed an integrated, multi-omic, functional tumor profiling of metastatic melanomas. 3 The aim was to correlate multimodal CT and FDG-PET-based radiomic features of resected metastatic melanoma lesions with high-dimensional molecular tumor data in order to better understand their biological correlates.

Material/Methods:

All metastatic melanoma patients treated with immunotherapy, who underwent FDG-PET/CT imaging within 8 weeks before resection of a non-cerebral metastasis of interest (MOI) for molecular analysis within the Tumor Profiler Study, were included. We manually segmented the MOIs and extracted radiomic signatures (n_intensity=16, n_texture=95) of the MOIs for correlation with molecular data (CyTOF and histology) ( Figure 1 ). Based on molecular data retrieved by CyTOF and conventional pathology, each metastasis was categorized in (1) melanoma tumor type , (2) immunotype , (3) whole-tumor cell content and ( 4) inflamed/non-inflamed type . Descriptive analysis, Cox regression, Kaplan-Meier method and log-rank test were employed to evaluate overall survival (OS) and progression-free survival (PFS). Preprocessing and extraction of radiomic features from the baseline medical images were done using an in-house developed software, Z-Rad. Mann-Whitney U test and bootstrapping was used to evaluate differences in radiomic feature distributions among different classes of molecular data. A radiomic feature was considered discriminatory if it was able to differentiate a molecular class with p-value < 0.05 and the 95% confidence interval of the area under the ROC curve (AUC) was not including 0.5. Eighty-one metastatic melanoma patients treated with immunotherapy were included in this study. The median age at the time of inclusion was 63.9 (range: 29.7-87) years; the most common melanoma type was cutaneous (n=61/81, 73.3%) followed by ocular melanoma (n=8/81, 9.9%). At the time of inclusion, 28.4 % of the patients presented with only a single tumor lesion, while oligometastatic disease with up to five lesions and polymetastatic disease was present in 26.0% and 45.6% of the cases, respectively. At a median follow-up time after inclusion into the study of 22.3 (range: 0.2-53.6) months, the median overall survival after start of immunotherapy was 28.2 (range: 20.5-43) months and median progression-free survival was 6 (range: 1-51) months. Distinct melanoma immunotypes identified in the Tumor Profiler Study correlated with overall survival and progression-free survival in our cohort (Fig 2A-B). Phenotype 1 was characterized by bystander CD4/8, naive B-cells, and quiescent cells. Phenotype 2 was characterized by exhausted CD8 cells whereas phenotype 3 by lack of B cells. Phenotype 1 (OS: HR=2.5 (1.2-5.1), p=0.013; PFS: HR=2.4 (1.1-4.9), p=0.019) had significantly shorter overall survival than Phenotype 2 and Phenotype 3 with p=0.014 and p=0.008, respectively. Similar pattern was observed for progression-free survival with p=0.005 and p=0.036 for Phenotype 2 and Phenotype 3, respectively. Radiomic features extracted from PET/CT images of MOIs were able to univariately differentiate between the immunotypes with AUC values in range 0.67-0.77 ( Fig. 2C-D ). Phenotype 1 and 2 had significantly different signatures of GLCM Dissimilarity and GLCM Inverse Difference extracted from PET images. Phenotype 1 and 3 had significantly different signatures of various CT-based GLSZM and GLDZM features as well as PET-based histogram features. We did not find radiomic signatures discriminating between Phenotype 2 and 3, however, these two phenotypes were characterized by a comparable patient survival. Results: