���ϳԹ���Դ

Glioblastoma (GBM) is the most aggressive primary malignant brain tumor in adults and is characterized by inevitable recurrence despite extensive first-line treatment. This thesis investigates how glioblastoma can be accurately delineated on radiological images and examines which biomarkers characterize and differentiate patients with long survival from those with poorer outcomes. In this way, the work aims to systematically explore the biological factors underlying the heterogeneous treatment responses observed in patients with this disease.

This was achieved through the application of artificial intelligence (AI) in several forms. Deep learning was used to segment the distinct tumor subregions at multiple time points throughout the disease course. These image-derived segments were subsequently integrated with clinical variables and genetic data, after which machine learning techniques were applied to identify patterns within the combined dataset.

The central hypothesis is that the radiological appearance of the tumor reflects intrinsic biological constraints that influence treatment sensitivity, driven by underlying genetic determinants.

The results demonstrated that the deep learning model delineated tumor volumes with greater accuracy than the inter-observer variability between two independent expert clinicians. Furthermore, the ratio between the contrast-enhancing (CE) and non-contrast-enhancing (NE) tumor components showed prognostic value, both with respect to the distance from the primary tumor site at recurrence and the time to recurrence. A lower CE/NE ratio was associated with more distant recurrences, particularly in younger patients and those with MGMT-methylated tumors. Additionally, a clear association was observed between a low CE/NE volume ratio and the cumulative mutational burden in genes regulating apoptosis, the cell cycle, and NF-κB signaling. We also identified mutations in HSPBP1 as a genetic indicator of response to proteasome inhibitor–based therapy.

The integration of these imaging-based and genetic biomarkers provides a conceptual foundation for the development of future personalized treatment strategies.