Construction and Validation of a Risk Prediction Model for Brucellosis Based on Deep Neural Networks
10.13471/j.cnki.j.sun.yat-sen.univ(med.sci).2025.0417
- VernacularTitle:基于深度神经网络的布鲁氏菌病风险预测模型的构建和验证
- Author:
Siyuan LIU
1
;
Biao SONG
;
Guizhi LIU
;
Jun WANG
;
Lan XUE
;
Jie SU
;
Hongli WANG
;
Xin SHEN
Author Information
1. 呼和浩特市职业病防治院,内蒙古自治区 呼和浩特 010020
- Publication Type:Journal Article
- Keywords:
Brucellosis;
deep neural network;
blood routine indices;
Shapley Additive exPlanations(SHAP);
risk prediction model
- From:
Journal of Sun Yat-sen University(Medical Sciences)
2025;46(4):700-707
- CountryChina
- Language:Chinese
-
Abstract:
[Objective]To construct a prediction model for brucellosis by using a deep neural network algorithm to improve the early detection.[Methods]We collected the clinical data of 202 brucellosis patients and 319 non-brucellosis patients admitted to Hohhot Occupational Disease Prevention and Treatment Hospital in 2023,and analyzed data such as gender,age,blood routine indices and clinical diagnosis.A prediction model for brucellosis was constructed by using a deep neural network algorithm and optimized through 10-fold cross-validation.Performance metrics included sensitivity,false negative rate,specificity,false positive rate,accuracy,positive predictive value,negative predictive value,F1 score,and area under the receiver operating characteristic curve(AUC).The optimal model was interpreted by using SHapley Additive exPlanations(SHAP)to clarify decision-making logic and feature influencing mechanisms.[Results]Data visualization analysis revealed no significant difference between brucellosis and non-brucellosis groups.The optimal model demonstrated good performance:sensitivity(85.3%),specificity(92.1%),accuracy(89.5%),AUC(96.6%),95%CI(0.937,0.977).SHAP analysis identified age,platelet count,mean platelet volume,basophil ratio,red blood cell distribution width,and absolute basophil count as significant predictors of brucellosis.[Conclusions]The deep neural network prediction model constructed in this study has good performance and can provide reliable support for the early diagnosis,prevention and control of brucellosis.Identification of key brucellosis-related influencing features will help further understand the pathogenesis of the disease,and this model holds promise for broad clinical application in the future.
