Attention-Based Radiomics to Predict Histological Grade of Gliomas

This paper proposes an attention-based deep learning framework for automatic glioma histological grade classification using multimodal MRI radiomic features, in accordance with WHO criteria. The pipeline includes image preprocessing, automatic tumor segmentation with state-of-the-art deep learning models, and extraction of handcrafted radiomic features capturing intensity, texture, and shape information. Experiments conducted on the public UCSF-PDGM dataset and on an independent external clinical cohort demonstrate that the proposed approach is aligned with existing baselines. By combining the predictive power of deep learning with the interpretability of radiomics, the proposed method enables accurate and clinically meaningful assessment of glioma histological grade, with potential implications for early detection, and preventive clinical decision-making, supporting prognosis estimation and treatment planning.