Optimizing Glioma Classification: Comparative Analysis of Radiologist, Radiomics and Multi-dimensional Deep Learning Using Quantitative MRI

Motivation: Gliomas present complex challenges in neuro-oncology, necessitating improved diagnostic methods that incorporate molecular characteristics alongside traditional imaging. Goal(s): The research aims to optimize adult diffuse glioma classification by integrating advanced quantitative MRI techniques with radiomics and DL models, seeking to surpass conventional radiologist assessments. Approach: A prospective study involving 428 patients employed advanced imaging protocols, including MUSE-DWI, 3D-pCASL, and synthetic MRI, combined with the development of independent binary classifiers for IDH mutation and 1p/19q co-deletion classification. Results: The 2.5D-DL fusion model demonstrated the highest diagnostic performance, significantly outperforming traditional methods and showcasing the effectiveness of multimodal imaging in glioma classification. Impact: This study could significantly enhance glioma diagnostics by employing advanced quantitative MRI alongside DL models, such as 2.5D-DL fusion model, which demonstrates superior classification accuracy. This approach enables personalized treatment strategies based on molecular insights, previously unattainable with traditional methods.