A supported decision-making model for idiopathic pulmonary fibrosis based on feature screening and optimized neural network

Idiopathic pulmonary fibrosis (IPF) severely impairs human respiratory function, with an increasing incidence and mortality. Treatment based on syndrome differentiation is the characteristics and essence of traditional Chinese medicine (TCM). This study proposes an interpretable intelligent classification model-the MIV-GA-LM-BP (MGLB) model-to support syndrome differentiation in TCM for IPF. Based on 956 real-world clinical cases, the mean impact value (MIV) algorithm was employed for feature screening to identify key symptom-syndrome relationships and improve model interpretability. A hybrid optimization strategy combining the Levenberg-Marquardt (LM) and genetic algorithm (GA) was applied to enhance the convergence speed and generalization performance of the BP neural network. Comparative experiments with GRA, PCA, and PSO demonstrated that the MGLB model achieved the highest classification accuracy of 81.22%, outperforming other models in both accuracy and stability. More importantly, the MIV-based feature screening enables transparent mapping between symptoms and syndromes, aligning well with TCM diagnostic logic. The proposed model not only provides a standardized reference for IPF diagnosis but also offers a new methodological framework for developing AI-driven TCM diagnostic tools. It contributes to the modernization and standardization of TCM and supports the integration of intelligent systems into clinical decision-making.