Research on Deep Learning-Based Algorithms for Medical Image Characterisation

Lung disease is a worldwide health problem, and accurate image diagnosis of lung disease is crucial for patient treatment and management. In recent years, deep learning has made significant progress in the field of medical image analysis, in which Graph Neural Networks (GNNs), as a special deep learning model, show great potential in lung disease diagnosis. In this study, we propose a graph neural network-based model for lung disease diagnosis, aiming to automate the analysis of lung images and lesion detection. First, we represent the lung images as a graph structure, where each pathogenic region represents a node in the graph. Then, the adjacency matrix and node features of the graph are constructed, and the image is subjected to feature extraction and representation learning using a graph neural network, which enables the model to capture the structural and contextual information in the image. To further improve the performance of the model, we introduce multilayer graph convolution and multilayer perceptron to enhance the sensitivity and differentiation ability of the graph neural network for lung lesions. We conducted experimental evaluations based on publicly available lung image datasets and compared them with traditional machine learning methods as well as other deep learning models. The experimental results show that the proposed graph neural network model achieves excellent performance on the lung disease diagnosis task.