Glioblastoma treatment response assessment is a critical component of tertiary prevention, in which the disease has already manifested, and the primary goal is to accurately monitor tumor evolution to guide therapeutic decisions and optimize ongoing patient management. In this context, this paper proposes a novel deep learning framework that integrates multimodal magnetic resonance imaging with radiomic features to automatically classify glioblastoma treatment response according to the Response Assessment in Neuro-Oncology (RANO) criteria. The proposed approach combines a preprocessing pipeline that ensures spatial standardization and skull stripping, automatic tumor segmentation using state-of-the-art deep learning models, and the extraction of handcrafted radiomic features capturing intensity, texture, and shape information. These features are aggregated and fed into predictive models for both binary (progressive vs. non-progressive disease) and multiclass (CR, PR, SD, PD) classification tasks. An experimental evaluation on the LUMIERE dataset demonstrates that the proposed method outperforms existing baselines while effectively leveraging the complementary strengths of deep and radiomic features. These results highlight the potential of combining deep learning and radiomics for automated, robust, and interpretable assessment of glioblastoma treatment response, ultimately supporting improved clinical decision-making in the advanced phases of disease management and contributing to more effective tertiary prevention, where timely and accurate evaluation of treatment efficacy is essential to guide the continuation or adjustment of therapeutic interventions.
Amato et al. (Mon,) studied this question.