Scholarlink

<p>This paper proposes an innovative NLP hybrid model for diagnosing and predicting common types of anemia in Traditional Chinese Medicine (TCM), addressing challenges of limited medical text data and suitable NLP models. Utilizing the pre-trained LERT model as a foundation, we integrate it with the GRU deep learning model to enhance text feature extraction, improving prediction accuracy and reducing computational complexity. The introduced AGN(Adaptive Gate Network) Module mitigates long-term dependency challenges associated with GRU, offering flexible data attention control and enhancing classification performance. A novel loss function, FCS, is designed to handle unbalanced datasets, focusing on difficult-to-classify samples for improved training stability and reduced misclassifications. Evaluating on a TCM anemia diagnosis dataset, our model outperforms benchmarks with an accuracy of 0.9430, precision(MA/WA) of 0.9517/0.9459, recall(MA/WA) of 0.9326/0.9430, and F1-score(MA/WA) of 0.9408/0.9430, demonstrating its potential to intelligently assist doctors in anemia diagnosis and prediction.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Research on a multi-classification diagnosis method for anemia text medical records in Traditional Chinese Medicine based on improved pre-trained LERT model

  • 期刊论文

  • 发表日期：

    2025年3月26日

查看全文
  • Chongxiao Peng,
  • Zhijun Gao,
  • Jinhuan Wang,
  • Xin Yue,
  • Lili Sun,
  • Yinhuan Sun,
  • Fuquan Du

摘要

This paper proposes an innovative NLP hybrid model for diagnosing and predicting common types of anemia in Traditional Chinese Medicine (TCM), addressing challenges of limited medical text data and suitable NLP models. Utilizing the pre-trained LERT model as a foundation, we integrate it with the GRU deep learning model to enhance text feature extraction, improving prediction accuracy and reducing computational complexity. The introduced AGN(Adaptive Gate Network) Module mitigates long-term dependency challenges associated with GRU, offering flexible data attention control and enhancing classification performance. A novel loss function, FCS, is designed to handle unbalanced datasets, focusing on difficult-to-classify samples for improved training stability and reduced misclassifications. Evaluating on a TCM anemia diagnosis dataset, our model outperforms benchmarks with an accuracy of 0.9430, precision(MA/WA) of 0.9517/0.9459, recall(MA/WA) of 0.9326/0.9430, and F1-score(MA/WA) of 0.9408/0.9430, demonstrating its potential to intelligently assist doctors in anemia diagnosis and prediction.