SCAG-Net: Automated Brain Tumor Prediction from MRI Using Cuttlefish-Optimized Attention-Based Graph Networks

Area of Science:

• Medical imaging analysis
• Artificial intelligence in oncology
• Computational neuroscience

Background:

• Automated systems are crucial for accurate and timely brain tumor recognition, reducing diagnostic delays and human error.
• High variability in tumor characteristics (location, size, shape) and infiltrative nature of gliomas pose significant segmentation and identification challenges.
• Existing automated systems struggle with complex tumor features and heterogeneity, impacting diagnostic efficiency.

Purpose of the Study:

• To develop an advanced automated system for improved brain tumor recognition from MRI images.
• To address the complexities of tumor variability, infiltrative gliomas, and feature redundancy.
• To enhance diagnostic efficiency and accuracy in clinical decision-making.

Main Methods:

• Integration of Swin-UNet for initial image region identification and error minimization.
• Application of a cuttlefish-optimized attention-based Graph Neural Network (SCAG-Net) for feature exploration and selection.
• Utilizing attention graph networks to process structural and heterogeneous information for robust classification.

Main Results:

• The SCAG-Net approach achieved high recognition accuracy on public datasets (BRATS 2018-2020, Figshare).
• Key performance metrics include a Dice coefficient of 0.989, Intersection over Union of 0.969, and classification accuracy of 0.992.
• The proposed system demonstrated statistically significant improvements over recent benchmark models (p < 0.05).

Conclusions:

• SCAG-Net offers a robust, efficient, and clinically deployable framework for brain tumor recognition from MRI.
• The approach effectively handles tumor heterogeneity and infiltrative gliomas, crucial for accurate diagnosis.
• This method supports rapid and precise diagnosis, maintaining expert-level performance in medical applications.