A whole body Radiomics approach in patients with metastatic melanoma undergoing systemic therapy: Fully automated longitudinal segmentation and Deep Learning-based outcome prediction

• Applicants: Professor Dr. Thomas Eigentler; Dr. Annika Gerken; Professor Dr. Ahmed Othman; Dr. Felix Peisen
• Subject Area: Nuclear Medicine, Radiotherapy, Radiobiology
• Term: from 2019 to 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 428216905

The incidence of melanoma has steadily been increasing over the past decades. In recent years, the treatment of patients with advanced melanoma has undergone a revolution owing to the introduction of the checkpoint inhibitors ipilimumab (CTLA-4), nivolumab / pembrolizumab (PD-1) and relatlimab (LAG-3) as well as targeted therapy using BRAF and MEK inhibitors. Due to the use of immunotherapeutic agents, a significant improvement of the patients’ overall survival and progression-free survival could be achieved. However, the identification of those patients who do not benefit from immunotherapy remains challenging. To date, no biomarker is widely accepted for routine clinical use; only unspecific clinical parameters such as lactate dehydrogenase (LDH), the presence of lung or liver metastases are applied as well as experimental biomarkers. Thus, the evaluation of additional imaging biomarkers from whole-body cross-sectional imaging may increase the clinical significance of prediction models. During the first funding period of SPP2177, we were able to set the basis for further evaluations by implementing improved segmentation algorithms as well as quantitative image analysis algorithms. In the upcoming funding period, we aim at the adaption and implementation of novel applications regarding artificial intelligence, especially Deep Learning and explainable AI for advanced image analysis. Particularly, the segmentation of highly complex CTs of metastatic melanoma patients may represent a use case for clinically feasible algorithms that will serve the purpose of automated follow-up RECIST evaluation and Radiomic analysis. Since segmentation algorithms trained on metastatic melanoma patients cover a broad range of lesion types, they can potentially be transferred to other entities. Furthermore, the advance of Radiomics towards clinical applicability by reducing manual steps that implicates a minimal effort for clinical implementation, may potentially be even more facilitated through the refinement of our already developed pipelines. In summary, our herein described objectives for the second funding period are firstly, to adapt and evaluate segmentation and registration algorithms for automated lesion follow-up incorporating whole-body CT as well as brain MRI data, and to optimize the workflow for clinical use; secondly, to develop explainable Deep Learning-based methods that combine clinical biomarkers and imaging data for a reliable prediction of response to therapy and patient outcome in metastatic melanoma; and thirdly, to implement our refined and evaluated automated segmentation pipeline and Deep Learning Delta Radiomics for a patient-based outcome prediction of metastatic melanoma into clinical routine.

• DFG Programme: Priority Programmes
• Subproject of: SPP 2177: Radiomics: Next Generation of Biomedical Imaging
• Co-Investigators: Professorin Dr. Bettina Baeßler; Professor Dr. Thorsten Persigehl