Laser induced breakdown spectroscopy (LIBS) has attracted extensive attention in the field of biomedical detection due to its advantages of rapid, in situ, micro-damage and multi-element synchronous analysis. However, its inherent spectral noise, matrix effect and other problems restrict the analysis accuracy. In recent years, the powerful nonlinear fitting and feature extraction capabilities of neural network models provide effective solutions to overcome these bottlenecks. This paper systematically reviews the research progress of LIBS technology combined with neural network model in biomedical field, focusing on the key research achievements of this combined strategy in tumor diagnosis and classification, cancer staging and treatment monitoring, bacterial and pathogen identification, as well as traditional Chinese medicine analysis and traceability. Studies have shown that by combining artificial neural network, convolutional neural network, spiking neural network and other models, it can extract deep features from complex LIBS spectra, significantly improve the accuracy and robustness of classification and quantitative analysis. It is concluded that the integration of LIBS and neural network forms a synergistic effect of “rapid detection+intelligent analysis”, which shows great clinical application potential and lays a solid technical foundation for the development of intelligent, high-precision point-of-care diagnosis system in the future.