Enhanced brain tumor classification using graph convolutional neural network architecture

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Oncology
• Computational Neuroscience

Background:

• Brain tumors pose a critical health risk, necessitating early detection for effective treatment.
• Conventional models struggle with non-Euclidean image data and pixel proximity analysis.
• Accurate classification of tumor types (Meningioma, Pituitary, Glioma) is crucial for patient management.

Purpose of the Study:

• To develop a novel model for accurate brain tumor detection and classification.
• To address limitations of traditional models in handling non-Euclidean image data.
• To improve diagnostic accuracy using a hybrid Graph Neural Network (GNN) and Convolutional Neural Network (CNN) approach.

Main Methods:

• A novel Convolutional Neural Network (CNN) based Graph Neural Network (GNN) model was developed.
• The model utilizes Graph Convolution operations to modify node features by incorporating information from neighboring nodes.
• A 26-layered CNN integrated with Batch Normalization and Dropout layers processed the graph-based input.

Main Results:

• The proposed Graph based Convolutional Neural Network (GCNN) model successfully addressed non-Euclidean distance considerations in image data.
• Five network variations (Net-0 to Net-4) were evaluated, with Net-2 demonstrating superior performance.
• The Net-2 model achieved a highest accuracy of 95.01% in brain tumor detection and classification.

Conclusions:

• The proposed hybrid GNN-CNN model offers a significant advancement in brain tumor detection and classification.
• This novel technique provides a critical alternative for statistical detection in suspected brain tumor cases.
• The model's effectiveness in handling complex image data suggests potential for broader clinical application.