Pituitary radiomics combined with MRI features for predicting growth hormone status in pediatric short stature
10.13929/j.issn.1003-3289.2025.07.009
- VernacularTitle:垂体影像组学联合MRI特征预测矮小症儿童生长激素状态
- Author:
Fukun SHI
1
;
Lan ZHANG
;
Yu GAO
;
Xiaoyang ZHAI
;
Qian XU
;
Jiaxu LIANG
;
Shengli SHI
;
Ling WU
Author Information
1. 河南中医药大学第一附属医院磁共振科,河南郑州 450000;河南中医药大学第一临床医学院,河南郑州 450046
- Publication Type:Journal Article
- Keywords:
short stature;
magnetic resonance imaging;
radiomics;
growth hormone deficiency
- From:
Chinese Journal of Medical Imaging Technology
2025;41(7):1073-1078
- CountryChina
- Language:Chinese
-
Abstract:
Objective To observe the value of pituitary radiomics and MRI features combined model for predicting growth hormone(GH)status in pediatric short stature.Methods Totally 300 children with short stature were enrolled as training set,while other 73 cases were taken as external validation set.Based on growth hormone stimulation test,the children were divided into GH deficiency(GHD)group(n=228)and non-GHD group(n=145).The training set included 196 cases in GHD subgroup and 104 cases in non-GHD subgroup,while the validation set included 32 cases in GHD subgroup and 41 cases in non-GHD subgroup.Radiomics features of pituitary were extracted from T1WI.The key features were selected using least absolute shrinkage and selection operator(LASSO)regression,and machine learning models were subsequently constructed using support vector machine(SVM),logistic regression(LR),naive Bayes(NB)and K-nearest neighbor(KNN),respectively.Then combined models were constructed combining with MRI features,and the efficacy of each model was evaluated.Results The area under the curve(AUC)of SVM,LR,NB,and KNN radiomics model for predicting GH status in pediatric short stature was 0.860,0.831,0.838 and 0.901 in training set,0.788,0.829,0.823 and 0.770 in validation set,while of the relative combined SVM,LR,NB and KNN model was 0.924,0.903,0.859 and 0.920 in training set,and 0.827,0.881,0.836 and 0.718 in validation set.LRcombined model had the best overall performance,with sensitivity of 84.94％,specificity of 80.56％and accuracy of 83.61％in training set,and 80.95％,72.22％and 80.00％in validation set,respectively.Conclusion Pituitary radiomics and MRI features combined model could effectively predict GH status in pediatric short stature.
