Publication

Abstract : Brain tumors are one of the top diseases with less survival rate that affect humans across the globe and the development of brain tumors is influenced by many factors. The abnormal growth of cell masses in different categories of brain tumors are detected using brain Imaging techniques such as Magnetic Resonance Imaging(MRI). The purpose of this study is to examine the potential of Graph Convolutional Networks (GCNs) combined with pre-trained Convolutional Neural Network(CNN) models for brain tumor classification. Novel architectures, including VGG16, Densenet121, and Resnet18, are used to harness hierarchical features learned by CNNs and exploit graph-based representations for tumor classification. The models showed accuracy ranging from 64 to 74% and Resnet18 showed improved feature extraction compared to other models. The findings of this study highlight the models’ efficacy in leveraging image features by using the pre-trained CNN models like VGG16, DenseNet121, and ResNet18 to extract rich hierarchical features from brain tumor MRI images. These extracted image features are then transformed into graph structures, enabling the application of Graph Convolutional Networks (GCNs) for classification tasks. By incorporating spatial dependencies among features through GCNs, the models can effectively learn complex patterns and relationships within the image data, leading to accurate brain tumor classification results. Visual representations using PCA offer insights into distinctive feature representations learned by each model, underscoring the significance of model architecture and backbone selection in GCN based medical image analysis. Future work aims to explore larger datasets, refine hyperparameters, and delve into interpretability aspects, further advancing GCNs’ role in medical imaging and diagnostics.