Deep Learning Accurately Estimates Pediatric Age from Pelvic Radiographs in Multicenter Study

A deep learning model estimated pediatric age from pelvic radiographs with a mean absolute error of 0.394 years internally and 0.513 years externally; accuracy within 1 year was 0.937 and 0.927, respectively.

• Model trained on 3972 pelvic radiographs (median age 2.5 years) using DenseNet161 with 19 age categories and conditional ordinal regression loss.
• External validation in 744 radiographs from two independent cohorts showed robust performance (MAE 0.513 years).
• Girls had lower error internally (MAE 0.360 vs. 0.419 for boys) but higher externally (0.628 vs. 0.455).

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.