This invention discloses a urodynamic intelligent diagnostic method based on a multi-tree model fusion of prior regularization and an age-modulated network, belonging to the field of medical intelligent
diagnostic technology. This method integrates four tree models—XGBoost, LightGBM,
Random Forest, and CatBoost—combining five-fold cross-validation and multiple random seeds to obtain feature prior weights, and assigns positive or negative weights using the Spearman
correlation coefficient. The age-modulated sub-network dynamically adjusts the feature weights according to the patient's age. The main network with prior regularization constraints uses a
cosine similarity regularization term to guide the direction of learnable weights, improving
diagnostic accuracy and
interpretability. This invention achieves an average AUC of over 0.94 for diagnosing bladder outlet obstruction, with significantly better accuracy, recall, and F1
score than existing methods. It effectively solves the problems of poor
interpretability, insufficient integration of individual age factors, and unreasonable feature weight design in existing diagnostic models, thus adapting to the needs of clinical urodynamic diagnosis.