Implementation of a radiomics pipeline for survival analysis in multiple myeloma patients using 18F FDG PET/CT images: unveiling prognostic markers and predictive models

Multiple Myeloma (MM) poses a formidable challenge due to its high mortality rate, necessitating standardized diagnostic criteria, personalized treatment strategies, and robust prognostic assessments. While clinical evaluations and tests are essential, Computed Tomography (CT) and Positron Emission Tomography (PET) scans play a crucial role in detecting MM lesions. This thesis introduces a comprehensive radiomics pipeline for CT and PET images of MM patients. The workflow begins with MOOSE-assisted spine segmentation, focusing on the disease-active region. Three segmentation versions are derived: original, including bone marrow, and considering surrounding areas. Feature extraction follows, shaping medical scans according to these segmentations. Survival analysis, using the Cox model, explores Progression-Free Survival (PFS) in MM patients to identify prognostic biomarkers and build predictive models. Findings suggest that the spine may not be optimal for prognostic insights compared to studies involving the entire skeleton. CT images outperform PET with higher concordance index (C-index) scores. Results differ based on segmentation shapes and image filters. The feature selection model reveals consistent emergence of shape features, especially "flatness." Pairing specific shape and texture features shows promise in Cox model predictions. This study yields valuable MM prognosis insights, emphasizing specific image regions, imaging modalities, and feature selection in enhancing predictive modeling for patient outcomes. Advanced radiomic analysis addresses the complexities of MM, aiding clinicians in diagnosis, treatment, and prognosis for this challenging disease.