Low-Grade Glioma Segmentation Based on CNN with Fully Connected CRF.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuro-oncology

Background:

• Glioma is a common and aggressive brain tumor requiring accurate clinical diagnosis.
• Accurate segmentation of brain tumors in MRI is crucial for treatment planning and monitoring.
• Existing segmentation methods may struggle with low-contrast tumors like low-grade gliomas.

Purpose of the Study:

• To propose a novel automatic 3D MRI segmentation method for glioma.
• To enhance the accuracy of low-grade glioma recognition and boundary delineation.
• To improve upon existing state-of-the-art methods for brain tumor segmentation.

Main Methods:

• A multipathway convolutional neural network (CNN) incorporating 3D information was developed.
• A fully connected conditional random field (CRF) was employed as a postprocessing step for refined segmentation.
• The method was evaluated on T2-FLAIR MRI images from 160 low-grade glioma patients.

Main Results:

• The proposed method achieved a Dice Similarity Coefficient (DSC) of 0.85 on a test set of 101 MRI images.
• This performance surpasses a state-of-the-art CNN method, which obtained a DSC of 0.76 on the same dataset.
• The results demonstrate superior performance in segmenting low-grade gliomas.

Conclusions:

• The novel 3D CNN-CRF method offers improved accuracy for low-grade glioma segmentation.
• This technique shows significant potential for clinical diagnosis of glioma.
• The method provides more delicate delineation of glioma boundaries compared to previous approaches.