# Final height prediction of girls at menarche: a combined model using left hand and wrist bone age, knee radiomic scores, and clinical characteristics

## TL;DR

This study improves final height prediction for girls after menarche by combining bone age, knee radiomic scores, and clinical data in a new model.

## Contribution

The study introduces a novel method using knee radiomic scores alongside traditional bone age for more accurate final height prediction.

## Key findings

- The new model achieved an R2 of 0.733 with low prediction residuals.
- The model outperformed traditional methods in Bland–Altman agreement analysis.
- Knee radiomic scores from the distal femur significantly improved prediction accuracy.

## Abstract

Accurate final height prediction for girls with menarche is important, yet traditional Greulich–Pyle (GP) and Bayley–Pinneau predictions based on left hand-wrist bone age (BA) and target height demonstrate limited accuracy. This study aims to develop a method to more accurately predict final height.

One hundred and seventy-three girls with menarche from August 2018 to June 2023 were analyzed retrospectively. BAs in Greulich and Pyle and Hoerr knee atlases were evaluated. Knee radiomic features were extracted using PyRadiomics; least absolute shrinkage and selection operator regression was utilized to develop radiomic scores of the distal femur and proximal tibia. Ordinary least squares regression with stepwise selection was used to build a multilinear equation. This was further compared with traditional methods in fivefold cross-validation (CV = 5) using residual distribution and Bland–Altman agreement analysis.

Height gain in our Chinese cohort after menarche was 8.94 ± 2.99 cm. A stepwise multilinear equation was built with height at menarche, BA of GP and radiomic score of the distal femur (R2 = 0.733, F statistic = 115.1, P < 0.05). Compared with traditional methods, a multilinear equation displayed the lowest residuals (residual range: − 5.677 cm to + 6.444 cm) and best Bland–Altman agreement (the mean difference: − 0.01 cm, 95% limits of agreement: − 3.96 to + 3.93 cm).

A robust linear regression model that incorporates knee radiomic scores, BA of GP, height at menarche, and father’s height demonstrated the best final height prediction in our cohort. This research is an innovative application of radiomic score of the distal femur to final height prediction. Further validation is warranted to test robustness across populations and scenarios.

The online version contains supplementary material available at 10.1007/s12519-025-01002-5.