Mycoplasma Pneumonia Diagnosis Algorithm Based on Multimodal Data Fusion

Abstract

The incidence of Mycoplasma Pneumonia (MP) has been on the rise. Traditional diagnostic methods have limitations, and existing machine - learning - based diagnostic studies also have deficiencies. In this study, a diagnostic system for Mycoplasma Pneumonia based on a multimodal visual hypergraph neural network (Trans - HGNN) was proposed. Firstly, basic processing such as denoising and standardization, as well as optimization using the VAE - GAN algorithm, were carried out on lung CT images and biochemical indicator data. Then, a hypergraph was constructed to fuse multimodal data, the VHNNs architecture was designed, and the Transformer algorithm was applied. Experiments show that the Trans - HGNN model performs best in terms of accuracy, recall rate, and F1 - score, reaching 88.69%, 0.8569, and 0.9229 respectively. The training process has good convergence and strong generalization ability. Although there may be a slight overfitting problem, it still provides reliable technical support for the diagnosis of Mycoplasma Pneumonia and has broad application prospects in the field of medical diagnosis.