MmMv

Deep MultiModal MRI Fusion Model for Brain Tumor Grading

1. Introduction

Glioma is a type of brain tumor that occurs in glial cells and is classified into two types: high grade glioma(HGG) with a poor prognosis and low grade glioma(LGG).

We developed a Multimodal & Multiview (MmMv) model that applied input-level fusion to MRI of four modalities (T1, T1Gd, T2, T2-FLAIR) for glioma grading.

2. Materials & Methods

Dataset

Brain Tumor Segmentation Challenge (BraTS) 2020

Training set : 248 subjects (HGG 197, LGG 51)
Test set : 121 subjects (HGG 96, LGG 25)

Multimodal model

Trained for axial, coronal, and sagittal planes respectively.

3. Results

Classification Performances

	Accuracy	Sensitivity	Specificity	F1 Score	AUC
Axial	0.9008	0.9479	0.7200	0.9381	0.8340
Coronal	0.9256	0.9063	1.000	0.9531	0.9508
Sagittal	0.8595	0.8750	0.8000	0.9081	0.8375
`MmMv`	0.8926	0.9688	0.6400	0.9347	0.9467

nainye/MmMv