nav emailalert searchbtn searchbox tablepage yinyongbenwen piczone journalimg journalInfo searchdiv qikanlogo popupnotification paper paperNew
2024, 11, v.45 8-14
基于随机森林算法的糖尿病性视网膜病变预测方法研究
基金项目(Foundation): 上海市经济和信息化委员会项目(2023040300189)
邮箱(Email): lijd@sdju.edu.cn;hajfs@126.com;
DOI: 10.19745/j.1003-8868.2024207
摘要:

目的:针对糖尿病随访数据集检测的特征维度多、患病样本某些指标的数值包含异常值和噪声的情况,提出一种基于随机森林(random forest,RF)算法的糖尿病性视网膜病变(diabetic retinopathy,DR)预测方法。方法:首先,使用Weka工具对上海交通大学附属第六人民医院及日本神户大学附属医院糖尿病内分泌科糖尿病患者的随访数据集进行特征选择,以筛选出与DR相关性较大的特征或特征子集;其次,基于特征子集和RF算法构建辅助临床诊断DR的模型;最后,为验证模型性能并判断哪个特征对模型的贡献最大,进行模型对比实验和消融实验。结果:由病程、糖化血红蛋白、促甲状腺激素、总胆红素、低密度脂蛋白、血清肌酐、白蛋白这7种特征构成的特征子集与DR的相关性最大,在此基础上基于RF算法构建的模型精确率为0.92、准确率为0.91、F1分数为0.91、AUC为0.95,均优于其他模型。消融实验结果表明,病程对模型的贡献最大,其次是白蛋白和血清肌酐,然后是低密度脂蛋白、总胆红素、糖化血红蛋白和促甲状腺激素。结论:基于RF算法的预测方法预测效果准确,可以辅助临床诊断DR。

Abstract:

Objective To propose a random forest (RF) algorithm-based prediction method for diabetic retinopathy (DR) to solve the problems due to multi feature dimension for detecting diabetes follow-up data and outliers and noises in the values of some indicators in the disease samples.Methods Firstly,the feature selection of the follow-up dataset of diabetic patients from the endocrinology and metabolism departments of Shanghai Jiao Tong University Affiliated Sixth People's Hospital and Kobe University Hospital in Japan was carried out using the Weka tool to screen out the features or subsets that were hightly correlated with DR;secondly,a model for assisting clinical diagnosis of DR was constructed based on feature subsets and RF algorithm;finally,model comparison experiments and ablation experiments were performed to validate the model performance and to determine which feature contributed the most to the model.Results The feature subset containing disease duration,glycosylated hemoglobin (Hb A1c),thyroid stimulating hormone (TSH),total bilirubin (T-bilirubin),low density lipoprotein(LDL),serum creatinine (s Cr) and albumin (ALB) correlated the most with RF.A model was constructed based on the above findings with RF algorithm,which behaved better than other models in terms of precision (0.92),accuracy (0.91),F1score(0.91) and AUC (0.95).The results of ablation experiments showed that the disease duration contributed the most to the model,followed by albumin and serum creatinine,and then by LDL,total bilirubin,glycosylated hemoglobin and thyroid stimulating hormone.Conclusion The RF algorithm-based prediction method with high accuracy can be used for assisted diagnosis of DR.[Chinese Medical Equipment Journal,2024,45(11):8-14]

参考文献

[1]沈胤忱,马航宇,王育璠,等.重视糖尿病性视网膜病变的全病程综合管理[J].中华预防医学杂志,2022,56(12):1 889-1 892.

[2]黄晓波,张培,林森林,等. 2016-2018年上海市新泾社区糖尿病眼病干预效果分析[J].中华预防医学杂志,2022,56(1):44-48.

[3] TEO Z L,THAM Y C,YU M,et al. Global prevalence of diabetic retinopathy and projection of burden through 2045:systematic review and meta-analysis[J]. Ophthalmology,2021,128(11):1 580-1 591.

[4] SHANKAR K,ZHANG Y Z,LIU Y W,et al. Hyperparameter tuning deep learning for diabetic retinopathy fundus image classification[J]. Access,2020,8:118 164-118 173.

[5] YANG J F,WANG Y,ZHOU L,et al. Allergy test dosage of fluorescein detects diabetic retinopathy changes in fundus fluorescein angiography[J]. Diagnostics,2023,13(23):3519.

[6] TERADA N,MURAKAMI T,ISHIHARA K,et al. Quantification of dilated deep capillaries in diabetic retinopathy on optical coherence tomography angiography[J]. Sci Rep,2023,13(1):17884.

[7] OUYANG J,MAO D,GUO Z,et al. Contrastive self-supervised learning for diabetic retinopathy early detection[J]. Med Biol Eng Comput,2023,61(9):2 441-2 452.

[8]尹曦,钟克丹,陈一玲,等.医联体糖尿病患者眼科随访的健康信念及其影响因素[J].眼科学报,2022,37(4):307-316.

[9] LI Z X,ZHANG J,FONG N,et al. Using artificial intelligence as an initial triage strategy in diabetic retinopathy screening program in China[J]. Natl Med J China,2020,100(48):3 835-3 840.

[10]蒋泽宇,韩荣,刘晓鸿,等.基于深度学习的医学影像高效生成方法研究[J].医疗卫生装备,2023,44(2):1-4.

[11]梁宏,王雅文.人工智能医疗器械的临床应用现状[J].医疗卫生装备,2024,45(2):74-81.

[12] GRZYBOWSKI A,SINGHANETR P,NANEGRUNGSUNK O,et al. Artificial intelligence for diabetic retinopathy screening using color retinal photographs:from development to deployment[J]. Ophthalmol Ther,2023,12(3):1 419-1 437.

[13] LIN D,QIN R,GUO L. Thyroid stimulating hormone aggravates diabetic retinopathy through the mitochondrial apoptotic pathway[J]. J Cell Physiol,2021,237(1):868-880.

[14] ZHONG J B,YAO Y F,ZENG G Q,et al. A closer association between blood urea nitrogen and the probability of diabetic retinopathy in patients with shorter type 2 diabetes duration[J].Sci Rep,2023,13(1):9881.

[15]韩磊,黄瑞龙,范文静,等.基于Weka平台和代价敏感特征选择的基因表达数据分类研究[J].智慧健康,2022,8(17):1-4.

[16] YITAYEH B,LISA M,LAN D,et al. Comparison of machine learning and conventional logistic regression-based prediction models for gestational diabetes in an ethnically diverse population;the Monash GDM machine learning model[J]. Int J Med Inform,2023,179:105228.

[17] PATRICK S,THOMAS R V. Logistic regression in medical research[J]. Anesth Analg,2021,132(2):365-366.

[18] BOURS M J. Bayes’rule in diagnosis[J]. J Clin Epidemiol,2021,131:158-160.

[19] VALKENBORG D,ROUSSEAU A J,GEUBBELMANS M,et al. Support vector machines[J]. Am J Orthod Dentofacial Orthop,2023,164(5):754-757.

[20] SOMASUNDARAM N,AYYASAMY B. Weighted entropy deep features on hybrid RNN with LSTM for glucose level and diabetes prediction[J]. Comput Method Biomec,2023,26(15):1 834-1 858.

[21] LIN K Y,HSIH W H,LIN Y B,et al. Update in the epidemiology,risk factors,screening,and treatment of diabetic retinopathy[J]. J Diabetes Investig,2021,12(8):1 322-1 325.

[22]李琦.快速血糖仪联合便捷式糖化血红蛋白仪检测在社区糖尿病诊治中的应用价值[J].中国医疗器械信息,2020,26(11):135-136.

[23]焦聪,侯超,李蓉. TyG指数与2型糖尿病非增殖性视网膜病变的相关性[J].中国临床研究,2023,36(5):656-660.

基本信息:

DOI:10.19745/j.1003-8868.2024207

中图分类号:R587.2;R774.1

引用信息:

[1]周亚斌,李建敦,陈京京等.基于随机森林算法的糖尿病性视网膜病变预测方法研究[J].医疗卫生装备,2024,45(11):8-14.DOI:10.19745/j.1003-8868.2024207.

基金信息:

上海市经济和信息化委员会项目(2023040300189)

检 索 高级检索

引用

GB/T 7714-2015 格式引文
MLA格式引文
APA格式引文