An Attention-Guided CNN Framework for Segmentation and Grading of Glioma Using 3D MRI Scans

Area of Science:

• Neuro-oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Glioma is a deadly brain tumor requiring timely diagnosis for effective treatment.
• Manual Magnetic Resonance Imaging (MRI) analysis for gliomas is time-consuming and prone to errors.
• Automated diagnosis systems are crucial for efficient clinical management and surgical planning of gliomas.

Purpose of the Study:

• To develop a Convolutional Neural Network (CNN)-based framework for non-invasive glioma grading using 3D MRI scans.
• To improve the accuracy and efficiency of glioma diagnosis and classification.

Main Methods:

• Proposed a novel CNN framework with two architectures: one for tumor segmentation and another for multi-task glioma grading.
• The segmentation network incorporates spatial and channel attention modules.
• The classification network performs multi-task learning for low-grade/high-grade classification, 1p19q codeletion, and Isocitrate Dehydrogenase (IDH) status identification.

Main Results:

• The proposed framework demonstrated superior performance compared to several state-of-the-art methods in glioma grading.
• Experimental results showed high accuracy in tumor segmentation and classification tasks.
• Welch's t-test confirmed the statistical significance of the grading results.

Conclusions:

• The developed CNN framework offers an effective automated solution for non-invasive glioma grading from 3D MRI.
• This approach has the potential to significantly aid clinicians in diagnosing and managing glioma patients.
• The study provides a valuable tool for advancing neuro-oncology research and practice.