18F-FDG PET radiomic analysis to predict outcomes in metastatic melanoma treated with immune checkpoint inhibitors

Karim Amrane^1*Coline Le Meur²David Bourhis³Christian Berthou⁴Olivier Pradier²Laurent Misery⁵Delphine Legoupil⁵Maxime Etienne⁶Georges-Philippe Fontaine⁷Cyril Leleu⁸Romain Floch³Pierre-Yves SALAUN³Ronan ABGRAL³Vincent Bourbonne²

• ¹Departement of Medical Oncology , Morlaix Hospital, Morlaix, France, Morlaix, France
• ²Department of Radiotherapy, University Hospital of Brest, France, Brest, France
• ³Department of Nuclear Medicine, University Hospital of Brest, France, Brest, France
• ⁴Department of Hematology, University Hospital of Brest, France, Brest, France
• ⁵Department of Dermatology, University Hospital of Brest, France, BREST, France
• ⁶Department of Dermatology Centre Hospitalier de Cornouaille, Quimper, France., Quimper, France
• ⁷Nuclear Medicine Centre Georges Charpak, Quimper, France., Quimper, France
• ⁸Department of Radiotherapy, Centre Hospitalier de Cornouaille, Quimper, France., Quimper, France

Abstract Purpose: Cutaneous melanoma (CM) incidence is rising, and despite advances in immune checkpoint inhibitors (ICI), many metastatic patients do not respond or develop resistance. This study aimed to evaluate the prognostic value of a pre-treatment FDG-PET/CT-based radiomic model (MEL-RAD) for predicting 1-year progression-free survival (1y-PFS) in metastatic CM patients treated with first-line ICI. Methods: We retrospectively included 154 metastatic CM patients from two centers who underwent pre-treatment FDG-PET/CT before ICI initiation. Patients were split into a development cohort (n=95) and an independent testing cohort (n=59). Radiomic features were extracted and harmonized to reduce inter-cohort variability. A two-step feature selection identified three key wavelet-transformed texture features used to build the MEL-RAD predictive model. The model's performance was assessed by receiver operating characteristic (ROC) analysis, sensitivity, specificity, and predictive values. Survival analyses (progression-free (PFS) and overall survival (OS)) were performed with Cox regression and Kaplan-Meier methods. Results: In the development cohort, MEL-RAD achieved an AUC of 0.74 (p<0.0001) for predicting 1y-PFS. Using a 55% probability threshold, sensitivity was 93.8%, specificity 31.9%, with positive and negative predictive values of 58.4% and 83.4%, respectively. Patients with MEL-RAD >55% had significantly worse PFS (HR=2.73, p=0.0009) and OS (HR=3.20, p=0.0003). These results were externally validated: in the testing cohort, MEL-RAD positivity remained significantly associated with poorer PFS (HR=2.73, p=0.047), and showed non-significant for OS. Conclusion: The MEL-RAD radiomic model based on pre-treatment FDG-PET/CT offers a non-invasive biomarker to stratify metastatic CM patients treated with immunotherapy.