河北水利电力学院学报

河北水利电力学院学报 ›› 2025, Vol. 35 ›› Issue (2): 22-27.DOI: 10.16046/j.cnki.issn2096-5680.2025.02.005

• 人工智能与机器人 • 上一篇    下一篇

AI用于医疗处方多标签分类方法研究——以脊髓损伤患者电子病历为例

赵慧敏1, 郭欣1, 郄博韬1, 刘玉丽2   

  1. 1.河北工业大学人工智能与数据科学学院,天津市北辰区西平道5340号 300401;
    2.山东第一医科大学第三附属医院,山东省济南市无影山路38号 250031
  • 收稿日期:2024-04-03 修回日期:2025-03-10 出版日期:2025-06-30 发布日期:2025-07-04
  • 通讯作者: 郭 欣(1976-),男,天津人,博士,教授,主要研究方向:模式识别与智能系统。E-mail:gxhebut@aliyun.com
  • 作者简介:赵慧敏(2000-),女,河北沧州人,硕士研究生,主要研究方向:智能控制。E-mail:zhaohuiminstu@126.com
  • 基金资助:
    国家重点研发计划(2019YFB1312500)

Research on Multi-label Classification Method of Medical Prescription by Using AI ——Taking electronic medical records of spinal cord injury patients as an example

ZHAO Huimin1, GUO Xin1, QIE Botao1, LIU Yuli2   

  1. 1. School of Artificial Intelligence and Data Science, Hebei University of Technology, 300401, Tianjin, China;
    2. The Third Affiliated Hospital of Shandong First Medical University, 250031, Jinan, Shandong, China
  • Received:2024-04-03 Revised:2025-03-10 Online:2025-06-30 Published:2025-07-04

PDF

摘要： 电子病历(Electronic Medical Records,EMRs)汇集了患者的医疗历史和健康状况数据。利用电子病历对脊髓损伤(Spinal Cord Injury,SCI)患者进行辅助诊断具有重要意义。因此,文中提出了一种基于电子病历的SCI患者康复处方决策模型。首先,构建了一个包含1443名截瘫类SCI患者的EMRs数据集,并相应地完成了数据预处理;其次,针对EMRs不平衡的问题,提出了基于MLSMOTE(Multi-label Synthetic Minority Over-sampling Technique)的多标签分类框架;最后,使用7个多标签分类模型来预测患者的物理治疗(Physical Therapy,PT)处方。所提出的MLSMOTE多标签分类框架可以充分解决类别不平衡的问题。实验结果显示,与其他6个模型相比,RAkEL模型在许多指标上都有显著提高。其中汉明损失和排名损失分别为0.1482和0.2616,精确度、召回率和F1分数分别为82.04%、81.0%和78.07%。文中提出的MLSMOTE多标签分类框架可充分利用EMRs数据,有效提高康复治疗处方的决策准确性。

关键词: 电子病历, MLSMOTE, 多标签分类, 脊髓损伤

Abstract: Electronic medical records (EMRs) are compilation of data concerning a patient's medical history and health status. The utilization of EMRs is imperative in the diagnosis of patients with spinal cord injuries (SCI), thus necessitating the development of an EHR-based rehabilitation prescription model for SCI patients. First, a dataset of EMRs containing 1443 paraplegic SCI patients was constructed, and data preprocessing was completed accordingly. Second, to address the problem of imbalanced EMRs, a multi-label classification based on MLSMOTE (Multi-label Synthetic Minority Over-sampling Technique) was proposed framework. Finally, seven multilabel classification models are used to predict patients' Physical Therapy (PT) prescriptions. The proposed MLSMOTE multi-label classification framework has been shown to adequately address the problem of category imbalance.The experimental results demonstrate that the RAkEL model exhibits significant improvement in numerous metrics when compared to the other six models. Specifically, the Hamming loss and ranking loss were found to be 0.1482 and 0.2616, respectively, while the precision, recall, and F1 score were determined to be 82.04%, 81.0%, and 78.07%, respectively.The MLSMOTE multi-label classification framework proposed in this study has the capacity to fully utilize EMRs data and effectively enhance the precision of rehabilitation therapy prescriptions.

Key words: electronic medical record, MLSMOTE, multi-label classification, spinal cord injury

中图分类号: