An Optimized Strategy for Brain Tumor Classification Using SO(3) Equivariant Graph Neural Networks with Snow Geese Algorithm in MRI Imaging.
Authors
Affiliations (4)
Affiliations (4)
- Department of Computer Science and Engineering, MLR Institute of Technology, Dundigal, Hyderabad, India. [email protected].
- School of Computing, SRM Institute of Science & Technology, Tiruchirappalli Campus, Tiruchirappalli, 621105, India.
- Department of Electrical and Electronics Engineering, Chaitanya Bharathi Institute of Technology, Hyderabad, India.
- Department of Computer Science and Engineering, Sri Sairam Engineering College, Sai Leo Nagar,West TambaramPoonthandalam, Village, Chennai, Tamil Nadu, 602109, India.
Abstract
Brain tumors (BT) are actually an uncontrolled growth of cancer cells inside the body that can be classified into several classes according to their characteristics and accessible therapies. Brain tumors require a thorough examination by medical professionals due to their seriousness and risk for death. One of the advanced digital image processing techniques utilized for categorized tumors is magnetic resonance imaging (MRI). Recently, a number of Deep Learning (DL) models have been created to help diagnose BT. Many of these kinds of models have poor accuracy, which could result in incorrect diagnosis. The Robust Peak Guided Filter R2U++ Multilayer Attention SO(3) Equivariant Graph Neural Network with Snow Geese Algorithm (RPGFR2U++MASO(3)EGNN-SGA) is a proposed methodology that uses data from the Contrast-Enhanced MRI (CE-MRI) and BRATS 2018 datasets to improve brain tumor classification. It employs the Iterative Robust Peak-Aware Guided Filter (RPAGF) to reduce noise and preserve critical features. The Multilayer Edge Attention (MEA-Net) enhances feature extraction and refinement, while SO(3)-equivariant Graph Neural Networks ensure precise graph-based feature analysis. The results express how well the proposed method performs, demonstrating its positive potential for cancer diagnosis. The suggested technique, RPGFR2U++MASO(3)EGNN-SGA, demonstrated its efficacy across a range of datasets with impressive accuracy of 99.6% for the BRATS 2018 dataset and 99.7% for the CE-MRI dataset. These results reveal that the suggested methodology outperforms existing methods, demonstrating its capabilities and potential for future breakthroughs in BT identification and classification.