Multiparametric Magnetic Resonance Imaging Information Fusion Using Graph Convolutional Network for Glioma Grading

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate preoperative glioma grading is crucial for patient treatment and prognosis.
• Multiparametric magnetic resonance imaging (mpMRI) offers detailed, non-invasive tumor tissue information.
• Existing convolutional neural network (CNN) methods for glioma grading using mpMRI have limitations in fully exploiting contextual data.

Purpose of the Study:

• To propose a novel graph convolutional network (GCN)-based module (MMIF-GCN) for comprehensive fusion of mpMRI data.
• To enhance glioma grading by effectively integrating rich tumor contextual information from multi-parametric MR images.
• To improve the accuracy of preoperative glioma grading through advanced information fusion techniques.

Main Methods:

• Constructing a graph where vertices represent glioma grading features from CNN-extracted slices and edges denote spatial distances.
• Updating vertex information by considering interactions between adjacent slices within a 3D volume.
• Implementing a GCN-based module (MMIF-GCN) for nonlinear representation learning and information fusion.

Main Results:

• The MMIF-GCN module effectively fuses grading-relevant information from mpMRI data.
• The proposed method demonstrated superior glioma grading performance on both public (BraTS2020) and private (GliomaHPPH2018) datasets.
• The GCN approach successfully maintained positional relationships between adjacent slices during fusion.

Conclusions:

• The novel MMIF-GCN module offers an effective approach for fusing mpMRI information for glioma grading.
• This method enhances the accuracy of preoperative glioma grading by leveraging comprehensive contextual data.
• The GCN-based fusion strategy represents a significant advancement in the application of AI for neuro-oncology diagnostics.