基于集成神经网络的类风湿关节炎中医证候分类研究

目的 针对基于AI技术的类风湿关节炎（RA）中医证候多标签分类中存在标签关联性差、泛化性能低等问题，提出构建一种集成神经网络模型来实现RA中医证候分类，并探究其中的特征重要性和风险因素，为RA的诊断和治疗提供参考。方法 本文提出一种集成神经网络模型实现RA中医证候分类。该模型采用一种基于多层神经网络的基分类器提取临床RA多标签样本的深层特征，增强RA特征区分度，根据协方差理论衡量标签相关性，调节分类器链的输入空间，减少RA错误信息传播和冗余度，采用集成学习方法减小分类器链中不合理标签序列对RA特征分类的影响。结果 该模型在10折交叉验证性能参数中表现出优秀的性能，其中汉明损失、1-错误率、准确率和F1值分别为0.0036、0.0248、97.52%、99.18%。与其他常用多标签分类器相比，该模型的性能更为优秀具有更好的分类性能。此外，本文分析了RA中医证候特征重要性，并挖掘了潜在的风险因素。结论 基于集成神经网络模型的RA中医证候分类器具有较高的分类精度和效率，对于RA的临床诊断和治疗具有重要参考价值。

Research on Classification of Chinese Medicine Syndromes in Rheumatoid Arthritis Based on Ensemble Neural Networks

Objective: To address issues such as poor label correlation and low generalization performance in the multi-label classification of traditional Chinese medicine (TCM) syndromes in rheumatoid arthritis (RA) based on AI technology, this study proposes the construction of an integrated neural network model to achieve TCM syndrome classification in RA. The research also aims to explore the importance of features and risk factors, providing valuable references for the diagnosis and treatment of RA. Methods: We introduce an integrated neural network model for TCM syndrome classification in RA. The model utilizes a base classifier based on a multi-layer neural network to extract deep features from clinical multi-label RA samples, enhancing the discriminative power of RA features. By measuring label correlation based on covariance theory, the input space of the classifier chain is adjusted to reduce the spread of RA error information and redundancy. An ensemble learning approach is employed to mitigate the impact of unreasonable label sequences in the classifier chain on RA feature classification. Results: The model demonstrates excellent performance in a 10-fold cross-validation, with Hamming loss, 1-error rate, accuracy, and F1 score being 0.0036, 0.0248, 97.52%, and 99.18%, respectively. Compared to other commonly used multi-label classifiers, this model exhibits superior performance and better classification capabilities. Additionally, the study analyzes the importance of TCM syndrome features in RA and explores potential risk factors. Conclusion: The RA TCM syndrome classifier based on the integrated neural network model exhibits high classification accuracy and efficiency, providing important reference value for the clinical diagnosis and treatment of RA.