检验医学 ›› 2025, Vol. 40 ›› Issue (4): 365-371.DOI: 10.3969/j.issn.1673-8640.2025.04.009

• 论著 • 上一篇    下一篇

基于红细胞形态和相关参数建立地中海贫血和缺铁性贫血机器学习鉴别公式

凌小婷1, 潘丽秋1, 黄运花1, 韦媛媛1, 邓小慧2, 叶丽花3, 林发全1, 黄一芳1()   

  1. 1.广西医科大学第一附属医院检验科 广西高校临床检验诊断学重点实验室,广西 南宁 530021
    2.广西医科大学基础医学院,广西 南宁 530021
    3.来宾市妇幼保健院检验科,广西 来宾 546100
  • 收稿日期:2024-06-09 修回日期:2024-12-08 出版日期:2025-04-30 发布日期:2025-05-08
  • 通讯作者: 黄一芳,E-mail:YFY004462@sr.gxmu.edu.cn
  • 作者简介:凌小婷,女,1999年生,学士,主要从事临床检验诊断学研究。
  • 基金资助:
    国家自然科学基金项目(82201913);广西自然科学基金项目(2024GXNSFBA010209);广西卫生健康委自筹课题(Z-A20230482);广西研究生教育创新计划项目(YCSW2024268);来宾市科技计划项目(来科转220831)

Establishment of machine learning discriminatory formulas for thalassemia and iron deficiency anemia based on red blood cell morphology and parameters

LING Xiaoting1, PAN Liqiu1, HUANG Yunhua1, WEI Yuanyuan1, DENG Xiaohui2, YE Lihua3, LIN Faquan1, HUANG Yifang1()   

  1. 1. Department of Clinical Laboratory,the First Affiliated Hospital of Guangxi Medical University,Key Laboratory of Clinical Laboratory Medicine of Guangxi Department of Education,Nanning 530021,Guangxi,China
    2. School of Basic Medical Sciences,Guangxi Medical University,Nanning 530021,Guangxi,China
    3. Department of Clinical Laboratory,Laibin Maternal and Child Health Hospital,Laibin 546100,Guangxi,China
  • Received:2024-06-09 Revised:2024-12-08 Online:2025-04-30 Published:2025-05-08

摘要:

目的 探究基于机器学习的红细胞(RBC)形态分析和RBC参数检测在地中海贫血和缺铁性贫血鉴别诊断中的应用价值。方法 选取2021年1月—2023年4月广西医科大学第一附属医院确诊的162例地中海贫血患者(地中海贫血组)和75例缺铁性贫血患者(缺铁性贫血组)。采用Trainable Weka Segmentation(简称TWS)算法对外周血涂片中的RBC进行自动计数和分类。比较地中海贫血组和缺铁性贫血组血液学指标和异常RBC百分比的差异。通过逐步判别分析建立鉴别公式,采用受试者工作特征(ROC)曲线评价鉴别公式的效能。结果 地中海贫血组RBC计数、血红蛋白(Hb)、平均红细胞血红蛋白含量(MCH)、平均红细胞血红蛋白浓度(MCHC)和球形RBC、口形RBC、靶形RBC、泪滴形RBC、棘形RBC、锯齿形RBC、裂RBC百分比均高于缺铁性贫血组(P<0.001)。基于TWS算法和逐步判别分析构建地中海贫血和缺铁性贫血鉴别公式,其曲线下面积(AUC)、敏感性、特异性和准确性分别为0.93、88.75%、84.02%和86.91%。进一步构建α-地中海贫血和β-地中海贫血鉴别公式,敏感性、特异性和准确性分别为70.3%、81.4%和72.8%。结论 基于异常RBC形态指数和RBC分析指标,通过机器学习算法构建鉴别公式,可作为鉴别诊断地中海贫血和缺铁性贫血,以及辅助α-地中海贫血和β-地中海贫血分型的有效工具。

关键词: 红细胞, 形态学, 机器学习, 地中海贫血, 缺铁性贫血

Abstract:

Objective To investigate the application role of machine learning-based red blood cell(RBC) morphological analysis combined with RBC parameters in the differential diagnosis of thalassemia(TT) and iron deficiency anemia(IDA).Methods Totally,162 patients with TT and 75 patients with IDA diagnosed at the First Affiliated Hospital of Guangxi Medical University from January 2021 to April 2023 were enrolled. The Trainable Weka Segmentation(TWS) algorithm was employed for automatic counting and the classification of RBC. Differences in hematological indicators and the percentages of abnormal RBC between groups were compared. Discriminatory formulas were established using stepwise discriminant analysis(SDA),and the diagnostic performance of the formula was evaluated by receiver operating characteristic(ROC) curve.Results Compared with IDA group,the RBC count,hemoglobin(Hb),mean corpuscular hemoglobin(MCH) and mean corpuscular hemoglobin concentration(MCHC) in TT group were higher,and The percentages of spherocytes,stomatocytes,target erythrocytes,teardrop erythrocytes,acanthocytes,serrated erythrocytes and schistacytes in TT group were higher than those in IDA group(P<0.001). The identification formulas for TT and IDA were established based on TWS algorithm and SDA. The area under curve(AUC),sensitivity,specificity and accuracy were 0.93,88.75%,84.02% and 86.91%,respectively. The sensitivity,specificity accuracy of the formula for α-TT and β-TT were 70.3% and 72.8%,respectively.Conclusions The discriminatory formulas established by the use of abnormal RBC morphology and parameters combined with machine learning can be used as an effective tool for the differential diagnosis of TT and IDA,as well as for assisting in the differentiation of α-TT and β-TT.

Key words: Red blood cell, Morphology, Machine learning, Thalassemia, Iron deficiency anemia

中图分类号: