Integration of Radiomics and Deep Learning for Advanced MRI-Based Brain Tumor Characterization

Accurate brain tumor classification is critical for early diagnosis and effective treatment planning. This study proposes a hybrid framework that integrates handcrafted radiomics descriptors with deep learning features extracted from MRI scans using MobileNetV2. Three models were systematically evaluated: a radiomics-only model, a deep-only model, and a fusion model combining both feature sets. The radiomics-only approach achieved moderate performance with 84.5% accuracy, an F1-score of 0.84, and an AUC of 0.97 on the test set. The deep learning model markedly improved results, attaining 95.9% accuracy, an F1-score of 0.96, and an AUC of 0.996. The proposed fusion model delivered the best outcomes, with 97.1% accuracy, an F1-score of 0.97, and an AUC of 0.998, confirming its robustness and superior discriminative power. These findings demonstrate that combining radiomics with deep representations enhances classification stability and reliability compared to single-modality models. Future research should focus on multi-center validation, integration of clinical data, and explainable AI techniques to advance clinical adoption.