Design of an efficient model for brain MRI image classification using graph neural networks

The need for accurate classification of brain MRI images in medical diagnosis has spurred the development of sophisticated methodologies. Existing approaches often encounter limitations in effectively capturing the intricate relationships between images, thereby hindering classification accuracy. To address these challenges, this study proposes a novel methodology leveraging Graph Neural Networks (GNNs) for brain MRI image classification. Conventional methods often struggle to integrate multidomain features extracted from MRI images comprehensively. Our proposed approach overcomes this limitation by employing a series of transformations, including Fourier, Gabor, and Convolution, to extract essential spatial and frequency domain characteristics. These features are then seamlessly integrated into a unified representation, enhancing the discriminative power of the classification model. Central to our methodology is the construction of a graph representation wherein each MRI image is represented as a node, facilitating the modeling of relationships between images. Through quantifying the similarity of feature vectors, the graph structure captures both structural and semantic relationships, providing valuable context for classification. The utilization of Graph Neural Networks (GNNs) enables our model to effectively leverage the rich graph structure for classification. By iteratively aggregating information from neighboring nodes, the GNN learns discriminative features, resulting in superior classification performance compared to traditional methods. The impact of our work is significant, as evidenced by substantial improvements in precision, accuracy, recall, delay, area under the curve (AUC), and specificity. These advancements hold promise for enhancing the efficiency and accuracy of medical diagnoses based on brain MRI images, ultimately benefiting patient care and treatment outcomes.