Comparison of three artificial intelligence-assisted bone age assessment methods for predicting adult height in girls
10.3760/cma.j.cn311282-20241204-00575
- VernacularTitle:三种人工智能骨龄评价方法对女童成年身高预测的比较
- Author:
Jinfeng CHEN
1
;
Huiping SU
;
Shuangyi LIU
;
Shurong HUANG
;
Li WANG
;
Xiu ZHAO
;
Qiru SU
;
Rongfei ZHENG
;
Zhe SU
Author Information
1. 汕头大学医学院附属深圳儿童医院内分泌科，深圳 518038
- Publication Type:Journal Article
- Keywords:
Predicted adult height;
Bone age;
Artificial intelligence;
Girl;
Accuracy
- From:
Chinese Journal of Endocrinology and Metabolism
2025;41(6):460-466
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To compare the accuracy and applicability of three adult height prediction methods based on artificial intelligence-assisted bone age assessment—the Bayley-Pinneau method(BP method), the Tanner-Whitehouse 3 method(TW3 method), and China 05 method—in girls.Methods:This bidirectional cohort study collected clinical data and 690 posteroanterior X-ray images of the left hand from 278 female children who underwent pubertal development assessments at Shenzhen Children′s Hospital between January 2014 and December 2020, with follow-up until adult height was reached. Adult height prediction was performed using BP, TW3, and China 05 methods on artificial intelligence-assisted bone age assessment.Results:The BP and TW3 methods overestimated adult height by(1.7±3.7) cm and(2.6±3.0) cm, respectively, while the China 05 method underestimated adult height by(2.3±3.5) cm. The proportion of PAH within±5 cm of FAH were 80.0% for the TW3 method, 77.0% for the BP method, and 74.2% for the China 05 method, with significant differences among them( P=0.038). Analysis of cases with prediction deviations>10 cm and subgroup comparisons revealed that the TW3 and BP methods were more likely to overestimate adult height in girls aged 6.0-<8.0 years, with delayed bone age, or in the prepubertal stage(all P<0.001). The China 05 method was more prone to underestimate adult height in those with advanced bone age( P<0.001). All three methods showed significantly greater prediction errors(absolute difference between PAH and FAH) in girls with early puberty compared to those with normal pubertal development(all P<0.05). Conclusions:The TW3 and BP methods tend to overestimate adult height in girls, while the China 05 method tends to underestimate it. Caution is warranted when predicting adult height, particularly in girls under 8 years of bone age, with delayed or advanced bone age, and those with early puberty.
