A functionally guided fusion Vision Transformer for predicting IDH status in gliomas: a multicenter study with...

Area of Science:

• Neuroimaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Accurate preoperative prediction of isocitrate dehydrogenase (IDH) genotype in gliomas is vital for treatment and prognosis.
• Current imaging methods face challenges in generalization due to variability across modalities and institutions.

Purpose of the Study:

• To develop and evaluate a functionally guided fusion Vision Transformer (FGF-ViT) network for glioma IDH genotype prediction.
• To assess the model's generalization across multicenter datasets and its robustness with incomplete multimodal inputs.

Main Methods:

• Retrospective multicenter study involving glioma patients.
• Construction of four FGF-ViT networks using combinations of conventional MRI (cMRI), diffusion-weighted imaging (DWI), and perfusion-weighted imaging (PWI).
• Fusion of cMRI, DWI, and DSC-PWI features using transformer attention; performance evaluated by AUC, accuracy, sensitivity, and specificity.

Main Results:

• The FGF-ViT achieved a robust AUC of 0.822 in an independent external validation cohort.
• Model performance remained stable despite the absence of one functional imaging modality.
• The network demonstrated generalizability across additional multicenter datasets with varying modality availability.

Conclusions:

• The FGF-ViT offers a clinically relevant, generalizable framework for preoperative IDH genotype prediction in gliomas.
• This multimodal imaging approach enables reliable application across different centers and under conditions of incomplete data.
• The study highlights the potential of AI in improving glioma diagnosis and management.