Keynote: A deep learning model accurately estimates children’s ages from chest X-rays, offering a viable complement to conventional wrist radiograph-based bone age assessment.
Keywords: paediatric age estimation, chest X-ray, bone age assessment, deep learning, coordinate attention, medical imaging
This study presents a novel approach to paediatric age estimation using chest X-rays and deep learning. Age assessment is a crucial component of monitoring growth and development in children, with wrist radiographs being the standard imaging method for evaluating bone age. The authors highlight that chest X-rays are already the most frequently performed radiological examination in paediatric care, making them an abundant and accessible source of diagnostic data. By applying artificial intelligence to this existing resource, additional imaging can potentially be avoided, lowering radiation exposure while providing valuable clinical insights.
Researchers retrospectively collected 128,008 regular chest X-rays from healthy Chinese children aged 0–16 years across two tertiary hospitals in Shanghai between 2021 and 2024. A modified ResNet-18 convolutional neural network, integrated with a Coordinate Attention mechanism, was trained and validated on data from Xinhua Hospital, with independent testing on an external dataset from Shanghai Tenth People’s Hospital. The model’s task was to predict chronological age from image data without requiring additional clinical annotations.
Performance was evaluated using the Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Spearman correlation coefficient. Internal validation achieved MAEs of 5.86 months for males and 5.80 months for females. On the external dataset, the results were 7.40 months and 7.29 months, respectively. Spearman correlation coefficients exceeded 0.98 across all groups, indicating robust agreement between predicted and actual ages. Subgroup analysis showed minimal error (~1.3 months) in infants under one year, with MAPE values below 10% for children over six years of age.
Attention heatmaps revealed that the model focused on anatomical areas, including the spine, mediastinum, heart, great vessels, and surrounding bones. These findings suggest that both skeletal and soft tissue features contribute to age estimation in chest radiographs.
The authors note that this approach offers a complementary method to traditional bone age assessment. It could be integrated into existing workflows to provide a secondary reference point, particularly in cases where wrist imaging is not available or advisable. Limitations include the potential for undiagnosed growth disorders and slightly lower accuracy compared to the most refined wrist radiograph models.
In conclusion, deep learning applied to chest X-rays provides an accurate, accessible, and low-radiation method for paediatric age estimation, with clear potential for future clinical adoption.
Reference: Li, M., Zhao, J., Liu, H. et al. Predicting pediatric age from chest X-rays using deep learning: a novel approach. Insights Imaging 16, 184 (2025). https://doi.org/10.1186/s13244-025-02068-5
Disclaimer: This article summarises findings from the referenced peer-reviewed study and is provided for educational and informational purposes only. It is not intended to replace professional medical advice, diagnosis, or treatment. Clinicians should consult the original publication and use their own judgement when applying these findings in practice.
You are here: home »