ESTRO 2024 - Abstract Book

S5148

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2024

[4] N. Beig et al., Radiogenomic-Based Survival Risk Stratification of Tumor Habitat on Gd-T1w MRI Is Associated with Biological Processes in GlioblastomaRadiogenomic Analysis of Tumor Habitat on MRI in Glioblastoma, Clinical Cancer Research. 26 (2020) 1866–1876. https://doi.org/10.1158/1078-0432.CCR-19-2556.

[5] C. Sun et al., Neighboring gray level dependence matrix for texture classification, Comput Vis Graph Image Process. 23 (1983) 341–352. https://doi.org/https://doi.org/10.1016/0734-189X(83)90032-4.

3052

Digital Poster

Immune signature of radiomic overall survival model in melanoma patients treated with immunotherapy

Hubert S Gabryś 1 , Maiwand Ahmadsei 1,2 , Stephanie Tanadini-Lang 1 , Reinhard Dummer 3 , Mitchell P Levesque 3 , Matthias Guckenberger 1 1 University Hospital Zurich, Department of Radiation Oncology, Zurich, Switzerland. 2 University Hospital Bern, Department of Radiation Oncology, Bern, Switzerland. 3 University Hospital Zurich, Department of Dermatology, Zurich, Switzerland

Purpose/Objective:

Early differentiation into well-responding and non-responding patients in metastatic melanoma treated with immune checkpoint inhibitors still poses a challenge. This work was a segment of the Tumor Profiler Study, which entails a comprehensive molecular and imaging analysis of metastatic melanomas. The goal of this work was to develop a radiomic-based patient survival model and use the molecular data to establish its immune profile.

Material/Methods:

The patient cohort comprised metastatic melanoma patients treated with immunotherapy with FDG-PET/CT imaging available within 8 weeks before resection of the non-cerebral metastasis of interest (MOI). The MOIs were manually segmented and radiomic features were extracted (n_intensity=16, n_texture=95) using an in-house developed software, Z-Rad. Patient overall survival at 24 months after last immunotherapy was defined as the clinical endpoint. The model was built in a setting of nested repeated cross-validation to achieve a stable model with robust performance estimates. Feature selection was realized with selection of the most informative features from a gradient tree boosting model that was fitted to the data. The final classification was performed with logistic regression to facilitate model interpretation. Model predictions were allowed to define “low risk” and “high risk” groups of patients. Subsequently, Mann-Whitney U tests and the area under the ROC curve (ROC-AUC) were used to relate immune cell profiles with the risk groups.

Results: