Automated Brain Tumor Classification and Grading Using Multi-scale Graph Neural Network with Spatio-Temporal...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Magnetic Resonance Imaging (MRI) is crucial for diagnosing brain conditions.
• Manual assessment of brain tumors in MRI scans is challenging due to feature heterogeneity and irregular shapes, leading to inaccuracies.
• Accurate brain tumor detection and grading are vital for effective treatment planning.

Purpose of the Study:

• To develop an Automated Classification and Grading Diagnosis Model (ACGDM) for enhanced brain tumor detection and grading using MRI.
• To improve the accuracy and efficiency of brain tumor diagnosis compared to conventional methods.
• To leverage advanced AI techniques for analyzing complex MRI data.

Main Methods:

• Proposed an Automated Classification and Grading Diagnosis Model (ACGDM) integrating a Multi-Scale Graph Neural Network (MSGNN) and a Spatio-Temporal Transformer Attention Mechanism (STTAM).
• MSGNN captures hierarchical and multi-scale dependencies in MRI data for superior feature representation.
• STTAM models spatial patterns and temporal evolution by incorporating cross-frame dependencies for enhanced sensitivity to disease progression.

Main Results:

• The ACGDM demonstrated 99.8% accuracy in detecting various brain tumor types across multiple datasets (BRATS 2018-2020, Br235H).
• The model effectively analyzed multi-modal MRI sequences, dynamically focusing on salient spatial and temporal features.
• Achieved high precision in identifying subtle disease progression and tumor characteristics.

Conclusions:

• The ACGDM significantly enhances brain tumor detection and grading accuracy using MRI.
• The proposed MSGNN and STTAM components offer a powerful approach for analyzing complex medical imaging data.
• This model has the potential to revolutionize diagnostic practices and improve patient outcomes in neuro-oncology.